Intelligent pump distribution analysis method and system based on flow prediction

Abstract

The invention relates to the field of water pump control, and particularly discloses an intelligent pump distribution analysis method and system based on flow prediction, and the method comprises the steps: constructing and continuously updating a pump unit individual digital file representing the real health state of a water pump; cooperatively performing water demand prediction and pipe network explosion and leakage risk probability pre-judgment based on the pump unit individual digital archive, and generating a first predicted flow value and an explosion and leakage risk coefficient; taking the first predicted flow value and the explosion and leakage risk coefficient as input, dynamically selecting a mode, and generating a pump distribution strategy with conventional energy efficiency optimization or pressure interference resistance; performing closed-loop optimization on the archive and the prediction model through comparison feedback of virtual deduction and actual execution; the system comprises five functional modules for realizing the steps. According to the method, the crossing from static optimization to dynamic adaptation, from passive response to active pre-judgment and from a fixed algorithm to a self-learning closed loop is realized, and the scientificity of a pump distribution decision, the safety of a water supply system and the overall energy efficiency are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of water pump control technology, and in particular to an intelligent pump allocation analysis method and system based on flow prediction. Background Technology

[0002] In urban water supply systems, the secondary pumping stations of waterworks are key power units responsible for pressurizing and delivering treated water to the municipal water network. Their core control objective is to minimize pump energy consumption while meeting rapidly changing water demands and ensuring stable network pressure. Currently, most waterworks still rely on manual experience or simple constant pressure control modes for pump scheduling, making it difficult to achieve refined and intelligent management.

[0003] With the development of automation technology, some intelligent pump allocation methods based on flow prediction have begun to be applied. Existing technical solutions usually follow a two-stage logic of "prediction-optimization": First, the demand for future periods is predicted using historical water consumption data; then, with this prediction value as a constraint, optimization calculations are performed based on the rated performance curve of the water pump, with the goal of minimizing energy consumption, to provide pump start-up, shutdown, and speed regulation schemes.

[0004] However, in practical applications, especially under the complex operating conditions of waterworks, existing technologies have the following deep-seated problems that have not yet been effectively resolved: The forecasting model is disconnected from the decision-making process and ignores the actual state of the equipment: it outputs only an isolated flow rate value, without considering the underlying operating characteristics of the demand (such as whether there are accompanying pressure anomalies). More importantly, subsequent optimization calculations generally assume that all pumps always operate on the ideal performance curve designed at the factory. In reality, the efficiency and head characteristics of pumps will continuously decline due to impeller cavitation, bearing wear, etc. The "optimal" strategy found based on "ideal performance" may actually force pumps to operate in the inefficient range during actual implementation, thereby increasing energy consumption and accelerating equipment wear.

[0005] Lack of ability to identify and respond to abnormal operating conditions in the pipeline network: This is a thorny and serious but often overlooked problem in waterworks operations. When a pipe bursts or experiences a major leak, the monitoring system detects a drop in pressure and an abnormal increase in flow. Existing methods are highly prone to misinterpreting this phenomenon as a surge in regional water demand, leading to erroneous instructions to "turn on more pumps." This not only fails to alleviate the leak but also results in a huge waste of water and electricity, and may even exacerbate the leak due to the impact of water flow.

[0006] The system lacks self-learning and adaptive capabilities: most existing solutions are open-loop control systems with fixed predictive models and optimization algorithm parameters. They cannot learn from long-term operating results and are difficult to adapt to the gradual decline in equipment performance, seasonal changes in water usage patterns, and partial modifications to the pipeline network structure, resulting in a gradual decrease in system efficiency over time.

[0007] Therefore, there is an urgent need for a new generation of intelligent pump allocation analysis method that can deeply integrate equipment status, intelligently identify operating risks, and continuously self-optimize, in order to break through the existing technical bottlenecks. Summary of the Invention

[0008] The purpose of this invention is to solve the problems of separation between prediction and decision-making, inability to identify pipeline anomalies, and neglect of dynamic degradation of equipment performance in the prior art, and to propose an intelligent pump distribution analysis method and system based on flow prediction.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A smart pump allocation analysis method based on flow prediction includes: Collect historical and real-time operating data of water pumps, and build and update individual digital profiles for each pump group; Based on the individual digital profiles of the pump units, the water demand is predicted, and a first predicted flow rate value is generated; at the same time, the pipeline pressure is simulated, and the risk of bursting is predicted by comparing with the real-time pressure, generating a bursting risk coefficient. Based on the first predicted flow rate value as the demand benchmark and the burst risk coefficient as the decision correction factor, an anti-interference pump allocation strategy is generated. Based on the individual digital files of the pump group and the anti-interference pump allocation strategy, a virtual simulation is performed to obtain the system state simulation value; The system state projection values ​​are compared with the actual data to generate model deviation feedback. The model deviation feedback is then used to adaptively optimize the file generation model for individual pump unit digital files and the water demand prediction model for generating the first predicted flow rate value.

[0010] As a further technical solution of the present invention, the collection of historical and real-time operating data of the water pumps, and the construction and updating of individual digital profiles for each pump group, specifically includes: For each water pump in the pumping station, historical performance curve data is obtained from the monitoring and data acquisition system. The historical performance curve data includes the relationship between flow rate, head and efficiency under different loads. Real-time operating current is collected in real time by Hall sensor installed at the pump motor, and vibration spectrum data is collected in real time by acceleration sensor installed on the pump casing. The cumulative maintenance records of the water pump are synchronized from the maintenance management system. The cumulative maintenance records include mechanical seal replacement, bearing lubrication and impeller repair events and their occurrence time. The collected data is input into the archive generation model, which calculates the current health score of the water pump based on real-time operating current, vibration spectrum data and cumulative maintenance records. The historical performance curve is dynamically corrected based on the current health score to generate an individual digital archive of the pump group that represents the actual flow-head-efficiency relationship of the water pump under the current health condition. Each time a water pump completes an operating cycle or a maintenance event occurs, the mapping relationship or coefficient in the individual digital file of the pump unit is iteratively updated using the latest operation and maintenance data.

[0011] As a further technical solution of the present invention, the water demand is predicted based on the individual digital profiles of the pump set, and a first predicted flow rate value is generated; simultaneously, the pipeline pressure is simulated, and the risk of leakage is predicted by comparing with the real-time pressure, generating a leakage risk coefficient; specifically including: Using the individual digital profiles of the pump units, the total water demand for future scheduling cycles is predicted, and a first predicted flow rate value is generated. Meanwhile, based on the actual flow-head relationship of the pumps in the current healthy state in the individual digital files of the pump set, combined with the pipeline topology and hydraulic model, the pressure distribution of the pipeline under the current pump set combination is simulated, and the simulated pressure is dynamically compared with the real-time monitoring pressure to calculate the pressure deviation value of each node. A pressure deviation threshold is set, and the number and duration of nodes whose pressure deviation values ​​continuously exceed the threshold are counted. The explosion risk coefficient is then calculated using a risk quantification function.

[0012] As a further technical solution of the present invention, the total water demand for future scheduling cycles is predicted using the individual digital files of the pump set, generating a first predicted flow rate value, specifically including: Call the collection of individual digital profiles of pump units ,in: To determine the number of operating pump units, extract the actual efficiency curve of each pump under its current health condition. Simultaneously collect external data from multiple sources: in the past Historical water consumption time series vector for days Weather forecast data for the next 24 hours, including temperature. and precipitation probability Calendar event data, including weekday and holiday markers. and special event markers The above data is input into the water demand prediction model, which performs a two-stage calculation: Predicting future scheduling cycles using Long Short-Term Memory (LSTM) networks The baseline water demand : ,in: The function for the baseline water demand prediction model; Calculate the overall efficiency attenuation coefficient of the pump set: ,in: For the first The flow load probability density function of the pump. and These are the upper and lower limits of the pump set's operating flow rate. To improve the overall efficiency of the pump system; The corrected first predicted flow value is: ,in: The first predicted flow value; This is the baseline efficiency value for a standard pump set system; Adjust the weighting coefficients for efficiency. .

[0013] As a further technical solution of the present invention, an anti-interference pump allocation strategy is generated using the first predicted flow rate value as the demand benchmark and the leakage risk coefficient as the decision correction factor, specifically including: Using the first predicted flow value as the demand benchmark and the leakage risk coefficient as the key correction factor for strategy generation, the leakage risk coefficient is compared with the preset decision threshold. When the risk coefficient is below the threshold, the conventional energy efficiency optimization is performed; when the risk coefficient exceeds the threshold, it is determined that there is a high probability of abnormal leakage in the pipeline network, and an anti-interference pump matching strategy with pressure stability as the core objective and compatible with the health status in the individual digital files of pump units is generated first.

[0014] As a further technical solution of the present invention, the conventional energy efficiency optimization and anti-interference pump matching strategy specifically includes: Conventional energy efficiency optimization uses the first predicted flow rate value as the total system demand constraint that must be met, and uses the actual efficiency of each pump represented in the individual digital files of the pump group as the core input parameter to construct and solve a multi-objective optimization problem. The goal is to minimize the total power consumption of the pump group and prioritize the use of the pump combination with the better health status in the file, and output an optimal energy efficiency pump matching scheme. When generating the anti-interference pump allocation strategy, maintaining stable pipeline pressure is set as the first priority objective. Based on the individual digital profiles of pump units, pumps with stable actual head-flow characteristics and good health status are selected to form the basic support unit. With the first predicted flow value as a reference, a flow safety margin that is dynamically adjusted according to the magnitude of the leak risk coefficient is added to form the pressure stability control objective. Finally, the anti-interference pump allocation strategy is generated, which aims to quickly compensate for pressure loss caused by suspected leakage.

[0015] As a further technical solution of the present invention, a virtual simulation is performed based on the individual digital files of the pump group and the anti-interference pump allocation strategy to obtain the system state simulation value, specifically including: The actual performance mapping relationship of the pumps in the individual digital files of the pump sets, as well as the pump set start-up and shutdown and frequency commands in the anti-interference pump allocation strategy, are input into the preset pipeline network digital twin model. In the digital twin model of the pipeline network, the hydraulic state of the system after the strategy is executed is dynamically simulated using two time scales: short-term minute-level and medium-term scheduling cycle-level. The system state projection values ​​of pressure, flow rate and total energy consumption of key nodes in the future period are obtained.

[0016] As a further technical solution of the present invention, obtaining the system state deduction value specifically includes: Collection of individual digital files for pump units Pump matching strategy Digital twin model of injection pipeline network Pump matching strategy Including start and stop commands With frequency conversion command Pipeline network digital twin model It consists of a coupled dynamic hydraulic equation set and a pump performance mapping module; Parallel execution of dual-timescale dynamic simulation in the digital twin model of the pipeline network: Using time step Duration of simulation Capture transient pressure fluctuations during pump start-up and shutdown: ,in: It is a numerical solver for transient hydraulic equations, used to solve nonlinear partial differential algebraic equations (PDAE) for pipe networks. The total head at the node, This is the transient hydraulic gradient function, which includes inertial and drag terms; Using time step Duration of simulation Predicting steady-state energy consumption and flow distribution: ,in: This is a steady-state hydraulic balance solver used to solve the hydraulic balance equations of a pipe network under constant operating conditions. The system state projection values ​​are defined as an aggregation of the two-scale results: in: This is a short-term stress projection matrix. This is a matrix for projecting medium-term stress. For short-term projection weights, Adjusted according to the pipeline network's inertial time constant; These are the aggregated values ​​of pressure and traffic projections for key nodes. To estimate the total energy consumption of the system, To extrapolate node traffic, For a moment No. The actual start-stop status of the water pump. No. The water pump has a flow rate With health status Instantaneous shaft power.

[0017] As a further technical solution of the present invention, the system state projection value is compared with the actual data to generate model deviation feedback, and the model deviation feedback is used to adaptively optimize the file generation model of the pump group individual digital file and the prediction model for generating the first predicted flow value, specifically including: After the anti-interference pump allocation strategy is actually implemented, real flow, pressure and energy consumption monitoring data are collected at the same time and location. The system state projection values ​​are compared with the real monitoring data item by item in real time to calculate the flow deviation rate, pressure consistency and energy consumption error. Based on the deviation index, a structured model deviation feedback is generated. The model deviation feedback clearly indicates which pumps in the individual digital files of the pump group have large deviations from the projection assumptions in actual performance, and under what operating conditions the collaborative prediction model has significant prediction errors. By utilizing the model bias feedback, the parameters of the file generation model are optimized. Specifically, the weights of the current distortion feature value, vibration degradation index, and time decay coefficient in the weighted fusion algorithm are adjusted to make the updated individual digital files of pump units closer to the actual decay trajectory of the water pump. At the same time, the working condition-error correspondence in the model bias feedback is used to perform incremental training or efficiency compensation logic correction on the water demand prediction model, thereby simultaneously improving the file fidelity and prediction accuracy.

[0018] A smart pump allocation analysis system based on flow prediction, used to implement a smart pump allocation analysis method based on flow prediction, includes: The data acquisition and communication interface module is used to collect real-time operating data of water pumps and real-time monitoring data of pipeline network, and synchronize historical performance curve data and cumulative maintenance records. The individual digital profile management module is connected to the data acquisition and communication interface module and is used to build and continuously update the individual digital profile of each pump group based on the received data. The collaborative prediction and risk assessment module is connected to the individual digital file management module and is used to call the individual digital files of the pump group to generate a first predicted flow rate value and a burst risk coefficient. The intelligent pump allocation strategy generation module is connected to the collaborative prediction and risk assessment module, and is used to generate an anti-interference pump allocation strategy or a conventional energy-efficient pump allocation scheme based on the first predicted flow value and the explosion risk coefficient. The virtual simulation and adaptive optimization module is connected to the individual digital file management module and the intelligent pump allocation strategy generation module, respectively. It is used to perform virtual simulation based on the individual digital files of the pump group and the generated pump allocation strategy, and optimize the individual digital file management module and the collaborative prediction and risk assessment module based on the comparison feedback between the simulation results and real data.

[0019] The beneficial effects of this invention are as follows: 1. Achieved a leap from static theoretical optimization to dynamic and precise adaptation in pump allocation decision-making: By constructing dynamically updated individual digital profiles for pump sets, the system can make decisions based on the actual health status and real-time performance of the pumps, rather than relying on fixed factory curves. This ensures better system energy efficiency and more balanced equipment load while guaranteeing water supply needs, thus extending the overall service life.

[0020] 2. Overcame the technical challenge of misjudging pipeline anomalies as normal demand, significantly improving system safety: The collaborative prediction and risk assessment mechanism, while predicting flow, uses archives to simulate pressure and compares it with real-time data to generate a burst risk coefficient. This enables the system to intelligently distinguish between changes in water demand and pipeline leaks, and automatically switch to an anti-interference pump allocation strategy centered on pressure stability when there is a high risk, thereby avoiding misoperation and greatly enhancing the resilience and reliability of the water supply network.

[0021] 3. An intelligent closed loop with self-learning and continuous optimization capabilities has been constructed: Through strategy virtual deduction and model adaptive optimization steps, the system can compare the decision results with the actual effects, and use the generated model deviation feedback to optimize the digital archives and prediction models in reverse. This enables the system to continuously self-correct and evolve, adapt to the long-term degradation of equipment performance and changes in water usage patterns, and realize the fundamental transformation from a fixed algorithm tool to an adaptive intelligent agent. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the structure of an intelligent pump allocation analysis method based on flow prediction proposed in this invention; Figure 2 This is a block diagram of an intelligent pump distribution analysis system based on flow prediction proposed in this invention; Figure 3 An architecture diagram of the archive production model; Figure 4 This is a comparison diagram of the effects of the method of the present invention and the prior art method in Example 1. Detailed Implementation

[0023] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0024] Please see the appendix Figure 1 A smart pump allocation analysis method based on flow prediction includes: S1. Collect historical and real-time operating data of the water pumps, and build and update individual digital profiles for each pump group; specifically including: S11. Regarding the first pump room A water pump performs multi-source heterogeneous data acquisition: S111. Collect historical performance curve data. Obtain the standard performance curve of the pump in the full load operating range from the Supervisory Control and Data Acquisition (SCADA) system, including the flow rate-head-efficiency relationship at rated speed, and denote it as a discrete dataset. ,in: For the first The water pump is in the first Rated flow rate under standard operating conditions For the first The water pump is in the first Rated head at a standard operating point For the first The water pump is in the first Rated efficiency at a standard operating point This represents the number of sampling points.

[0025] S112. Real-time operating current is collected via a Hall sensor installed in the power supply circuit of the water pump motor at a sampling frequency. Acquire instantaneous three-phase current signals of the motor, and obtain the effective current timing data through Park transform. ,in: , The initial time, For the monitoring cycle.

[0026] S113. Collect vibration spectrum data by using a triaxial accelerometer installed at the pump casing bearing housing to collect vibration acceleration signals. The vibration spectrum matrix is ​​obtained by Fast Fourier Transform (FFT). ,in: For frequency components.

[0027] S114. Collect cumulative maintenance records and synchronize the historical maintenance event sequence of the water pump from the Maintenance Management System (CMMS). ,in: For the first Sub-maintenance event types include mechanical seal replacement, bearing lubrication, and impeller repair. For the first The number of days since this event occurred This represents the total number of maintenance events that have occurred.

[0028] S12. Input the collected data into the document generation model, and the document generation model performs the following calculation process: S121. Real-time operating current Harmonic analysis was performed to extract the total harmonic distortion (THD) of the current as an electrical health characteristic. ,in: The effective value of the fundamental current is extracted using FFT. Component amplitude; For the first RMS value of subharmonic current For harmonic order; To determine the highest harmonic order for analysis, it is usually taken as... or ; This is a characteristic value of current distortion; the larger the value, the more severe the abnormality in the motor windings or frequency converter.

[0029] S122. Vibration spectrum Extract the characteristic frequency amplitude and calculate the comprehensive vibration degradation index: ,in: For the first Characteristic frequencies, including frequency conversion Leaf frequency bearing characteristic frequency wait; , for frequency The magnitude of the vibration amplitude vector at that location; This is the reference vibration amplitude of the water pump in its new state or after the last maintenance; For the first The weighting coefficients of each feature frequency satisfy the following conditions: Assigned according to the importance of failure modes; The total number of characteristic frequencies; The vibration degradation index, This indicates severe deterioration of the mechanical parts.

[0030] S123. To maintain the event sequence Constructing time-decaying health impact factors: ,in: Event type The attenuation coefficient, for example: mechanical seal replacement bearing lubrication Impeller repair ; To maintain the attenuation coefficient, The smaller the value, the more significant the maintenance effect decays over time.

[0031] S124. Calculate the overall health score of the water pump using the health fusion function: ,in: For the first The current health rating of the water pump. 1 indicates perfect health; The upper limit of allowable current distortion is usually taken as... ; For the weighting coefficients, satisfying Based on operational experience, optimization is recommended. In the formula Ensure that the higher the vibration index, the smaller the health contribution.

[0032] S125. Based on health score The standard performance curve is dynamically corrected by performing a nonlinear degradation mapping, thereby generating an individual digital profile for the pump unit. in: This refers to the actual performance curve, which is the core data of the individual pump unit's digital profile; For standard performance curves; The degradation coefficient for flow rate and head is determined based on pump type test calibration, typically... ; The efficiency decay semi-empirical exponent is generally taken as... ; The final files are stored in the form of discrete mapping tables or coefficient matrices: or ,in: This is the coefficient matrix of the corrected performance curve.

[0033] like Figure 3 As shown: The architecture of the archive generation model is as follows: Multi-source heterogeneous data fusion (input layer): Incorporating real-time electrical signals (current), mechanical vibration signals, and maintenance business data into the analysis scope, providing a three-dimensional data foundation for comprehensively assessing the "health status" of water pumps.

[0034] Health-oriented feature engineering (processing layer): Harmonic analysis of current can detect early faults such as motor electrical imbalance and rotor bar breakage. Spectral analysis of vibration can accurately locate the wear status of mechanical components such as bearings and impellers. A time decay model transforms discrete maintenance events into continuous performance impact coefficients. Through a weighted fusion algorithm, the above features are integrated into an authoritative and quantitative "current health score".

[0035] Dynamic file generation and iteration (output layer): The dynamic performance correction engine receives "historical performance curves" and "current health scores," and corrects the original curves in real time according to predetermined rules (such as the mapping relationship between performance degradation and health) to obtain the current true performance mapping. The output is a "digital mapping table or coefficient matrix" that can be directly called by the algorithm, i.e., the individual digital file of the pump group.

[0036] S13. Set file update trigger conditions: Automatically repeat processes S11 to S12 whenever any of the following events are met: the cumulative running time of the water pump increases. Hours; a new maintenance event occurred. And enter it into the CMMS system; Changes in health score ; Updated archives Replace the old version and synchronize it with the water demand prediction model in step S2 and the pipeline digital twin model in step S4 to achieve spatiotemporal consistency and timeliness of the archives.

[0037] S2. Based on the individual digital profiles of pump units, predict water demand and generate the first predicted flow rate value; simultaneously simulate pipeline pressure, and predict the risk of leaks by comparing with real-time pressure, generating a leak risk coefficient; specifically including: S21. Retrieve the collection of individual digital files for each pump unit. ,in: To determine the number of operating pump units, extract the actual efficiency curve of each pump under its current health condition. Simultaneously collect external data from multiple sources: in the past Historical water consumption time series vector for days Weather forecast data for the next 24 hours, including temperature. and precipitation probability Calendar event data, including weekday and holiday markers. and special event markers The above data is input into the water demand prediction model, which performs a two-stage calculation: Phase 1: Predicting future scheduling cycles using Long Short-Term Memory (LSTM) networks The baseline water demand : ,in: The function for the baseline water demand prediction model; Phase Two: Calculate the overall efficiency attenuation coefficient of the pump set: ,in: For the first The flow load probability density function of the pump. and These are the upper and lower limits of the pump set's operating flow rate. For the overall efficiency of the pump system, the corrected first predicted flow rate value is: ,in: The first predicted flow rate value is the corrected prediction of the total water demand of the system. The standard pump set system efficiency benchmark value is the average efficiency of all pumps in their new operating condition. Adjust the weighting coefficients for efficiency. The adjustment is made dynamically based on the pipeline leakage rate.

[0038] S22. Based on the currently active pump group combination index set. Extract the actual flow rate-head function of the corresponding pump from the individual pump set digital file. Based on the adjacency matrix of the pipeline network topology Using a hydraulic model, solve the nonlinear equations of the pipe network. The hydraulic equilibrium equations are as follows: in: The node-pipeline topology correlation matrix. For pipeline flow vectors, This represents the node traffic demand vector. Let the nodal head vector be... This is the pipeline resistance coefficient matrix. For nodes A collection of interconnected pipes; For the set of pipeline nodes, , Total number of nodes; pump head vector The head of each pump is obtained by superimposing their actual heads: ; The simulated pressure distribution is obtained by solving the problem. ,in: For nodes The simulated pressure value; Synchronous acquisition of pipeline network Real-time pressure monitoring of each pressure sensor node Calculate the pressure deviation value at each node: ,in: For sensors The simulated pressure value, For sensors The collected real-time pressure values, For a moment The pressure deviation value.

[0039] S23. Set pressure deviation threshold (Adjustable according to pipeline network level), define the anomaly indication function: ; Statistics within the judgment time window Within this range, the number of nodes satisfying the continuous abnormal conditions and their duration are as follows: ,in: This is the time window for risk assessment. For a moment The equivalent number of abnormal nodes; Calculate the pressure propagation consistency coefficient to identify multi-node coordinated pressure drop that matches the characteristics of a burst leak: ,in: The pressure propagation consistency coefficient, ; For sensor nodes and The distance of the pipeline topology; The propagation attenuation characteristic distance is usually taken as... ; Finally, the location is calculated using a risk quantification function. Risk coefficient of leakage in the range: ,in: To account for the risk factor of leakage, The closer the value is to 1, the higher the risk; activation function Map the output to , The slope coefficient of the sigmoid function controls risk sensitivity. This is the offset of the center of the Sigmoid function, usually taken as... ; For risk factor weights, satisfying .

[0040] S3. Using the first predicted flow rate as the demand benchmark and the leak risk coefficient as the decision correction factor, generate an anti-interference pump allocation strategy; specifically including: S31. Setting Decision Thresholds (usually taken) The leakage risk coefficient output in step S2) In comparison: in: For the current moment Decision-making model In standard energy efficiency mode, This is an anti-interference emergency mode.

[0041] S32. When Then, construct a multi-objective mixed-integer nonlinear programming (MINLP) problem: The decision variable is a binary variable for pump start-up and shutdown. Pump set flow distribution variables ,in: This represents the total number of water pumps in the pump house. The objective function is to satisfy the first predicted flow value. Under the premise that, i.e., satisfying Under the premise of minimizing total power consumption and prioritizing the use of water pumps in good health: ,in: For the first The water pump has a flow rate Shaft power below, , The density of water, It is the acceleration due to gravity; As a health penalty coefficient, The lower the health level, the greater the power consumption weight. For the first Health rating of the water pump (from S1 output); This is the penalty coefficient for the number of pump units. This is to avoid excessive pump starts and stops; The minimum number of operating pumps is typically taken as the baseline value. , To round up; The optimal solution can be obtained by using the branch and bound method or a genetic algorithm to solve the above MINLP. Output conventional energy-efficient pump matching scheme ,in: For the first Optimal flow distribution for the water pump. For the first The optimal start-stop state for the water pump.

[0042] S33. When At that time, an emergency strategy with pressure stabilization as the first priority will be generated: S331. From the individual digital files of pump units Select a set of candidate stable pumps that meet the following conditions : Among them: health threshold Slope threshold (This indicates that the absolute value of the slope of the flow rate-head curve does not exceed 0.05). As a reference flow point, it is usually taken as , For the first The maximum effective operating flow rate of the water pump under its current healthy condition.

[0043] S332. Based on the risk coefficient of leakage Calculate the emergency flow safety margin: Among them: risk amplification factor Basic margin coefficient , To ensure a safety margin for emergency flow; To achieve stable pressure control of the target flow rate: ,in: To stabilize the pressure and control the target flow rate.

[0044] S333. Construct a quadratic programming problem with the objective of minimizing the pressure deviation at critical nodes: ,in: For key monitoring nodes (such as the end of the pipeline network, high-risk areas); For nodes Pressure weighting coefficient ; For nodes The set pressure target value; Solving the above optimization problem yields an anti-interference pump allocation strategy. ,in: For the first Emergency flow distribution of the water pump. For the first The emergency start-stop strategy for the water pump; its core control commands are: prioritize maintaining stable pipeline pressure, quickly compensate for leakage losses, and increase the flow demand benchmark to [priority to [specific value]]. .

[0045] S4. Based on the individual digital profiles of pump units and the anti-interference pump allocation strategy, a virtual simulation is performed to obtain the system state simulation values; specifically including: S41. Collect individual digital files of pump units Pump matching strategy Digital twin model of injection pipeline network Pump matching strategy Including start and stop commands With frequency conversion command Pipeline network digital twin model It consists of a coupled dynamic hydraulic equation set and a pump performance mapping module; S42. Parallel execution of dual-timescale dynamic simulation in the digital twin model of the pipeline network: Using time step Duration of simulation Capture transient pressure fluctuations during pump start-up and shutdown: ,in: It is a numerical solver for transient hydraulic equations, used to solve nonlinear partial differential algebraic equations (PDAE) for pipe networks. The total head at the node, This is the transient hydraulic gradient function, which includes inertial and drag terms; Using time step Duration of simulation Predicting steady-state energy consumption and flow distribution: ,in: This is a steady-state hydraulic balance solver used to solve the hydraulic balance equations of a pipe network under constant operating conditions. S43. The system state projection value is defined as the aggregation of two-scale results: in: This is a short-term stress projection matrix. This is a matrix for projecting medium-term stress. For short-term projection weights, Adjusted according to the pipeline network's inertial time constant; These are the aggregated values ​​of pressure and traffic projections for key nodes. To estimate the total energy consumption of the system, To extrapolate node traffic, For a moment No. The actual start-stop status of the water pump. No. The water pump has a flow rate With health status Instantaneous shaft power.

[0046] S5. Compare the system state projection values ​​with actual data to generate model deviation feedback, and use the model deviation feedback to adaptively optimize the file generation model for individual pump unit digital files and the water demand prediction model for generating the first predicted flow rate; specifically including: S51. Pump matching strategy After actual implementation, real monitoring data is collected simultaneously: pressure monitoring data. Traffic monitoring data Energy consumption monitoring data Data is obtained from the main incoming line meter; a real-time comparison is performed item by item to calculate the deviation index. Statistical flow deviation rate ,in: For the number of flow meters, For the first Estimated flow rate values ​​for each flow meter location; Calculate pressure fit ,in: For the number of pressure sensors, To allow for a baseline value for pressure deviation, it is typically taken as... ; For the first The projected pressure values ​​of each pressure sensor node; Assessment of energy consumption error ,in: To estimate the total energy consumption of the system; Model bias feedback is constructed as structured triples: ,in: List of deviations in pump records; To maintain a list of recorded deviations; This represents the error range of the prediction model.

[0047] S52. Utilizing Feedback Perform bidirectional synchronous tuning: S521. Online learning of the health-weighted fusion function in step S12: Where: loss function Defined as the exponentially weighted moving average of the efficiency deviation rate: , For the first Table pump efficiency deviation rate. For the number of deviation pumps; These are the first two terms in the weighted fusion algorithm. New and old values ​​of the item weights (corresponding to current distortion, vibration degradation, and time decay); For the learning rate of the archival model, ; S522. When the error operating condition area When the water demand prediction model is not empty, retraining is triggered: Among them: augmented dataset Includes true data under high-error operating conditions. These are the water demand prediction models before and after the update. Represents the model retraining function; S523. Synchronously adjust the efficiency correction weighting coefficient in S21: in: and These are the efficiency correction weights before and after the update. To compensate for efficiency, an adaptive gain coefficient is used. ; The target energy consumption error rate is typically set at 3%.

[0048] Please see the appendix Figure 2 A smart pump allocation analysis system based on flow prediction is used to implement a smart pump allocation analysis method based on flow prediction, comprising: The data acquisition and communication interface module is used to collect real-time operating data of the water pump and real-time monitoring data of the pipeline network, and to synchronize historical performance curve data and cumulative maintenance records; specifically, it includes: A unit used to collect real-time operating current and vibration spectrum data from field sensors; A unit used to acquire real-time monitoring pressure from pipeline pressure sensors; This unit is used to synchronize historical performance curve data from the monitoring system and cumulative maintenance records from the maintenance management system.

[0049] An individual digital profile management module, connected to the data acquisition and communication interface module, is used to build and continuously update an individual digital profile for each water pump based on the received data; specifically including: The archive generation unit is used to perform harmonic analysis on real-time operating current, extract features from vibration spectrum data, and calculate the current health score of the water pump by combining cumulative maintenance records and using a weighted fusion algorithm. The file construction unit is used to dynamically correct the historical performance curve based on the current health score and generate individual digital files of pump groups that represent the actual performance mapping relationship of the water pump under the current health state. The file update unit is used to iteratively update the individual digital files of pump units after receiving new operation or maintenance data.

[0050] The collaborative prediction and risk assessment module, connected to the individual digital file management module, is used to access the individual digital files of the pump group to generate a first predicted flow rate value and a leak risk coefficient; specifically including: The collaborative water demand prediction unit is used to combine external data with the actual pump efficiency in the individual digital files of the pump group to predict and output the first predicted flow rate value. The pipeline pressure simulation unit is used to simulate the pressure distribution of the pipeline network based on the actual performance of the current pump set combination and the individual pump set digital files; The risk prediction unit is used to dynamically compare the simulated pressure distribution with the real-time monitored pressure, and calculate and output the explosion risk coefficient through a risk quantification function.

[0051] A smart pump allocation strategy generation module, connected to the collaborative prediction and risk assessment module, is used to generate an anti-interference pump allocation strategy or a conventional energy-efficient pump allocation scheme based on the first predicted flow rate value and the leak risk coefficient; specifically including: The decision mode determination unit is used to compare the leakage risk coefficient with the preset decision threshold in order to select the decision mode; The conventional energy efficiency optimization unit is used in a low-risk mode to optimize the pump matching scheme by taking the first predicted flow rate value as the demand constraint and the efficiency data in the individual digital files of the pump group as the input. The anti-interference strategy generation unit is used to generate anti-interference pump matching strategies in high-risk modes, with pressure stability as the primary goal, by combining the individual digital profiles of pump units.

[0052] The virtual simulation and adaptive optimization module is connected to the individual digital file management module and the intelligent pump allocation strategy generation module, respectively. It is used to perform virtual simulations based on the individual digital files of the pump group and the generated pump allocation strategies, and to optimize the individual digital file management module and the collaborative prediction and risk assessment module based on the comparison feedback between the simulation results and real data. Specifically, it includes: The virtual simulation unit is used to perform multi-timescale simulations in the digital twin model of the pipeline network based on the individual digital files of pump groups and the generated pump allocation strategy to obtain the system state simulation values. The feedback generation unit is used to compare the system state projection values ​​with the actual monitoring data after the strategy is actually executed, and generate model deviation feedback. The adaptive optimization unit is used to adjust the parameters of the archive update algorithm in the individual digital archive management module by utilizing model bias feedback, and to correct and train the prediction model in the collaborative prediction and risk assessment module.

[0053] Example 1 To verify the effectiveness of the method of the present invention, a simulation case of an urban water supply pumping station is used to compare and verify the actual effect of the method of the present invention with that of existing typical methods.

[0054] 1. Test Environment Setup Hardware and Data Foundation: A simulation system was built based on a medium-sized secondary pumping station of a municipal water supply company, consisting of 6 pumps in parallel (4 in operation and 2 on standby, including 2 variable frequency pumps). The pipeline model includes 12 key pressure monitoring nodes. Historical SCADA data (flow rate, pressure, current, and equipment maintenance records) from the pumping station over the past year was used as the training and testing dataset.

[0055] Existing technical method (comparative example): The "prediction + static optimization" model, which is currently more advanced in the industry, is adopted. Its core is: using a time series model to predict the hourly water consumption of the next day; based on the rated performance curve of the water pumps at the factory, with the predicted water consumption as a constraint and the minimum energy consumption as the goal, a genetic algorithm is used to optimize the combination of pump sets.

[0056] The method of this invention (example): adopts the complete scheme of steps S1-S5. The key steps are set as follows: the health score of the individual digital profile is updated every 24 hours; the decision threshold for the exposure risk coefficient is set to 0.7; virtual simulation is performed once a day to optimize the model.

[0057] 2. Test Scenario and Process Two representative scenarios were designed for comparative testing, with each scenario running continuously for 30 days.

[0058] Scenario 1: Comparison of comprehensive energy efficiency under normal operating conditions The simulation of a stable water supply period in late spring and early summer showed that water consumption exhibited regular daily cyclical fluctuations, with no major abnormal events.

[0059] Comparative example (existing technology): Based on the predicted water volume, the "optimal" pump set is selected strictly according to the rated efficiency. For example, if the predicted demand for a certain period is 1000 m³ / h, the system calculates that the total efficiency of pumps No. 1 (variable frequency) and No. 3 (industrial frequency) is the highest under rated operating conditions, and this is followed.

[0060] This invention (example): System records show that pump No. 3 has experienced slight bearing wear due to long-term operation, resulting in an 8% decrease in actual efficiency compared to the rated value. Meanwhile, the records indicate that pump No. 2 is currently in better health. Therefore, the system's strategy is to activate a combination of pumps No. 1 and No. 2, fine-tuning the frequency to meet the requirements.

[0061] Comparison Focus: Under the premise of meeting the same water supply demand, compare the total system energy consumption and pump unit operational health of the two methods.

[0062] Scenario 2: Comparison of safety responses under abnormal operating conditions On the 15th day of running Scenario 1, a simulated "sudden burst" event (a branch pipe ruptures and leaks continuously) is injected into the pipeline model.

[0063] Comparative example (existing technology): Pressure sensors detected a drop in pipeline pressure. The predictive model misinterpreted the resulting abnormal increase in water consumption (actually leakage) as a surge in residential water demand. To meet this artificially inflated predicted water volume, the optimization algorithm issued an instruction to "activate an additional large pump (pump No. 4)," leading to further expansion of the leak, increased water waste, and a surge in electricity consumption.

[0064] In this invention (example): The pressure simulation unit detected a synchronous abnormal drop in pressure monitoring values ​​at nodes 3 and 7, and the pattern matched the characteristics of a leak. The leak risk coefficient was calculated in real time and rose to 0.85 (> the 0.7 threshold). The system immediately switched to "anti-interference mode." At this time, the first predicted flow rate value was only used as a reference, and the core objective became "stabilizing the pressure at key nodes." Based on the data, the system selected pumps 1 and 2, which had the most stable head characteristics, as the main pumps and slightly increased their frequency to accurately compensate for pressure loss, rather than blindly increasing the number of large pumps. Simultaneously, the system automatically generated a leak alarm and located the abnormal range.

[0065] Comparison Focus: Comparing the two methods in terms of decision-making correctness after an anomaly occurs, water waste, and the speed at which the system recovers to stability.

[0066] 3. Verification Results Comparison Table The key performance indicators of the method of this invention and the existing technology in a 30-day simulation operation are compared in Table 1 below. Figure 4 As shown: Table 1: Comparison of Results 4. Conclusion The quantitative comparison of the above embodiments confirms that, compared with existing technologies, the intelligent pump allocation analysis method provided by this invention achieves comprehensive superiority in terms of economy, safety, and reliability. Its core innovative value lies in: breaking through the decision-making bottleneck relying on rated performance curves through "individual pump unit digital profiles," making optimization more closely aligned with the actual state of the equipment; and through the "collaborative prediction and risk assessment" mechanism, adding the ability to perceive pipeline integrity risks in addition to traditional demand forecasting, thus shifting the safety threshold forward. The "virtual simulation and adaptive optimization" module endows the system with the ability to learn from historical decisions, making the entire analysis system a continuously evolving intelligent agent.

[0067] This embodiment fully demonstrates that the present invention effectively solves the industry's obscure pain point of misjudging pipeline leaks as water demand, and significantly enhances the resilience and safe operation level of the water supply system while improving energy efficiency.

[0068] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects: This invention represents a fundamental shift from "static optimization based on rated parameters" to "dynamic and precise adaptation based on real-time status," significantly improving system energy efficiency and extending equipment lifespan. It constructs individual digital profiles for each pump unit. These profiles are not static factory data but integrate historical performance, real-time current / vibration characteristics, and maintenance records, dynamically depicting the unique "health status" and true efficiency curve of each pump. This innovation completely resolves the industry paradox of existing technologies that optimize based on fixed performance curves while actual equipment efficiency has declined—leading to "optimal on paper, but high consumption in reality." When making pump allocation decisions, the system proactively avoids operating pumps with declining efficiency in high-load ranges and prioritizes scheduling equipment combinations with better health status. The result is that, while meeting the same water supply demand, the overall system energy efficiency can be improved by more than 8%. Simultaneously, through load balancing and preventative scheduling, the wear rate of key pump components is effectively reduced, achieving a unification of energy saving and asset lifespan management.

[0069] Breaking away from the traditional framework where "prediction" and "safety" are often separated, this invention pioneers a collaborative mechanism for flow demand forecasting and pipeline leak risk prediction, significantly enhancing the safety and resilience of water supply systems. Utilizing unified individual pump unit digital profiles, this invention executes collaborative water demand forecasting and pipeline leak risk prediction in parallel during step S2. By dynamically comparing the pressure distribution simulated in the profiles with real-time monitoring data, it can intelligently distinguish whether an increase in water consumption stems from genuine demand or pipeline leakage. This mechanism directly addresses the industry's uncommon problem of misjudging leaks as peak water usage, leading to incorrect pump increases and exacerbating water waste. When the system calculates a high leak risk coefficient, it immediately triggers a decision-making mode switch. The effect is that, in the event of a suspected leak, the system can quickly (within minutes) generate an anti-interference pump allocation strategy with pressure stability as the core objective, proactively compensating for pressure loss rather than blindly increasing water supply. Practice has proven that this mechanism can reduce the scale of abnormal leakage events by more than 80% and increase the pressure guarantee rate of key nodes in the pipeline network from about 65% with traditional methods to 98%, greatly ensuring the safety and stability of water supply.

[0070] A continuous self-evolutionary closed loop of "decision-verification-learning" is constructed, enabling the system to evolve from a fixed algorithm tool into an adaptive intelligent agent. This invention adds a "virtual strategy deduction and model adaptive optimization" step at the end of the decision chain. This step deduces the files and strategies generated in the preceding steps in a digital twin environment and generates model deviation feedback by comparing them with actual results. This closed-loop design solves the deep-seated problems of traditional intelligent systems, such as rigid models and inability to adapt to long-term equipment performance drift and changes in water usage patterns. Feedback is used not only to calibrate individual pump group digital files, enabling them to more accurately track equipment aging trajectories, but also to continuously optimize the collaborative prediction model. The effect is that the entire analysis system possesses online learning capabilities; the experience of each decision execution is used to improve the accuracy of the next prediction and the fidelity of the files. This allows the system's energy efficiency and decision accuracy to continuously improve over time, with increasingly obvious long-term operational advantages, achieving a leap from "one-time programming" to "lifelong learning."

[0071] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0072] This invention is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this specification. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A smart pump allocation analysis method based on flow prediction, characterized in that, include: Collect historical and real-time operating data of water pumps, and build and update individual digital profiles for each pump group; Based on the individual digital profiles of the pump units, the water demand is predicted, and a first predicted flow rate value is generated. Simultaneously, the pipeline pressure is simulated, and the risk of leakage is predicted by comparing with the real-time pressure, generating a leakage risk coefficient; Based on the first predicted flow rate value as the demand benchmark and the burst risk coefficient as the decision correction factor, an anti-interference pump allocation strategy is generated. Based on the individual digital files of the pump group and the anti-interference pump allocation strategy, a virtual simulation is performed to obtain the system state simulation value; The system state projection values ​​are compared with the actual data to generate model deviation feedback. The model deviation feedback is then used to adaptively optimize the file generation model for individual pump unit digital files and the water demand prediction model for generating the first predicted flow rate value.

2. The intelligent pump allocation analysis method based on flow prediction according to claim 1, characterized in that, The collection of historical and real-time operating data of the water pumps is used to build and update an individual digital profile for each pump group, specifically including: For each water pump in the pump room, historical performance curve data is obtained from the monitoring and data acquisition system, real-time operating current is collected in real time by Hall sensor installed at the pump motor, and vibration spectrum data is collected in real time by acceleration sensor installed on the pump casing. The cumulative maintenance records of the water pump are synchronized from the maintenance management system. The collected data is input into the archive generation model, which calculates the current health score of the water pump based on real-time operating current, vibration spectrum data and cumulative maintenance records. The historical performance curve is dynamically corrected based on the current health score to generate an individual digital archive of the pump group that represents the actual flow-head-efficiency relationship of the water pump under the current health condition. Each time a water pump completes an operating cycle or a maintenance event occurs, the mapping relationship or coefficient in the individual digital file of the pump unit is iteratively updated using the latest operation and maintenance data.

3. The intelligent pump allocation analysis method based on flow prediction according to claim 1, characterized in that, Based on the individual digital profiles of the pump units, the water demand is predicted, and a first predicted flow rate value is generated. Simultaneously, the pipeline pressure is simulated, and the risk of leakage is predicted by comparing it with real-time pressure, generating a leakage risk coefficient; specifically including: Using the individual digital profiles of the pump units, the total water demand for future scheduling cycles is predicted, and a first predicted flow rate value is generated. Meanwhile, based on the actual flow-head relationship of the pumps in the current health state in the individual digital files of the pump set, combined with the pipeline network topology and hydraulic model, the pressure distribution of the pipeline network under the current pump set combination is simulated, and the simulated pressure is dynamically compared with the real-time monitoring pressure to calculate the pressure deviation value of each node. A pressure deviation threshold is set, and the number and duration of nodes whose pressure deviation values ​​continuously exceed the threshold are counted. The explosion risk coefficient is then calculated using a risk quantification function.

4. The intelligent pump allocation analysis method based on flow prediction according to claim 3, characterized in that, Using the individual digital profiles of the pump units, the total water demand for future scheduling cycles is predicted, generating a first predicted flow rate value, specifically including: Call the collection of individual digital profiles of pump units ,in: To determine the number of operating pump units, extract the actual efficiency curve of each pump under its current health condition. Simultaneously collect external data from multiple sources: in the past Historical water consumption time series vector for days Weather forecast data for the next 24 hours, including temperature. and precipitation probability Calendar event data, including weekday and holiday markers. and special event markers ; The above data is input into the water demand prediction model, which performs a two-stage calculation: Predicting future scheduling cycles using Long Short-Term Memory (LSTM) networks The baseline water demand ; Calculate the overall efficiency attenuation coefficient of the pump set: ,in: For the first The flow load probability density function of the pump. and These are the upper and lower limits of the pump set's operating flow rate. To improve the overall efficiency of the pump system; The corrected first predicted flow value is: ,in: The first predicted flow value; This is the baseline efficiency value for a standard pump set system; Adjust the weighting coefficients for efficiency.

5. The intelligent pump allocation analysis method based on flow prediction according to claim 1, characterized in that, Using the first predicted flow rate as the demand benchmark and the leak risk coefficient as the decision correction factor, an anti-interference pump allocation strategy is generated, specifically including: Using the first predicted flow value as the demand benchmark and the leakage risk coefficient as the key correction factor for strategy generation, the leakage risk coefficient is compared with the preset decision threshold. When the risk coefficient is below the threshold, the conventional energy efficiency optimization is performed; when the risk coefficient exceeds the threshold, it is determined that there is a high probability of abnormal leakage in the pipeline network, and an anti-interference pump matching strategy with pressure stability as the core objective and compatible with the health status in the individual digital files of pump units is generated first.

6. The intelligent pump allocation analysis method based on flow prediction according to claim 5, characterized in that, The conventional energy efficiency optimization and anti-interference pump matching strategies specifically include: Conventional energy efficiency optimization uses the first predicted flow rate value as the total system demand constraint that must be met, and uses the actual efficiency of each pump represented in the individual digital files of the pump group as the core input parameter to construct and solve a multi-objective optimization problem. The goal is to minimize the total power consumption of the pump group and prioritize the use of the pump combination with the better health status in the file, and output an optimal energy efficiency pump matching scheme. When generating the anti-interference pump allocation strategy, maintaining stable pipeline pressure is set as the first priority objective. Based on the individual digital profiles of pump units, pumps with stable actual head-flow characteristics and good health status are selected to form the basic support unit. With the first predicted flow rate as a reference, a flow safety margin that is dynamically adjusted according to the magnitude of the leak risk coefficient is added to form the pressure stability control objective. Finally, the anti-interference pump allocation strategy is generated, which aims to quickly compensate for pressure loss caused by suspected leakage.

7. The intelligent pump allocation analysis method based on flow prediction according to claim 1, characterized in that, Based on the individual digital profiles of the pump sets and the anti-interference pump allocation strategy, a virtual simulation is performed to obtain the system state simulation values, specifically including: The actual performance mapping relationship of the pumps in the individual digital files of the pump sets, as well as the pump set start-up and shutdown and frequency commands in the anti-interference pump allocation strategy, are input into the preset digital twin model of the pipeline network. In the digital twin model of the pipeline network, the hydraulic state of the system after the strategy is executed is dynamically simulated using two time scales: short-term minute-level and medium-term scheduling cycle-level. The system state projection values ​​of pressure, flow rate and total energy consumption of key nodes in the future period are obtained.

8. The intelligent pump allocation analysis method based on flow prediction according to claim 7, characterized in that, The obtained system state projection values ​​specifically include: Collection of individual digital files for pump units Pump matching strategy Digital twin model of injection pipeline network Pipeline network digital twin model It consists of a dynamic hydraulic equation set coupled with a pump performance mapping module; Parallel execution of dual-timescale dynamic simulation in the digital twin model of the pipeline network: using time step Duration of simulation Capture transient pressure fluctuations during pump start-up and shutdown; employ time step Duration of simulation Predict steady-state energy consumption and flow distribution; The system state projection values ​​are defined as an aggregation of the two-scale results: in: This is a short-term stress projection matrix. This is a matrix for projecting medium-term stress. For short-term projection weights, These are the aggregated values ​​of pressure and traffic projections for key nodes. To estimate the total energy consumption of the system, To extrapolate node traffic, For a moment No. The actual start-stop status of the water pump. No. The water pump has a flow rate With health status Instantaneous shaft power.

9. The intelligent pump allocation analysis method based on flow prediction according to claim 8, characterized in that, The system state projection values ​​are compared with actual data to generate model deviation feedback. This model deviation feedback is then used to adaptively optimize the file generation model for individual pump unit digital files and the prediction model for generating the first predicted flow rate value. Specifically, this includes: After the anti-interference pump allocation strategy is actually implemented, real flow, pressure and energy consumption monitoring data are collected at the same time and location; the system state projection values ​​are compared with the real monitoring data item by item in real time to calculate the flow deviation rate, pressure consistency and energy consumption error; based on the deviation index, a structured model deviation feedback is generated. The parameters of the data generation model are optimized using the model deviation feedback; at the same time, the working condition-error correspondence in the model deviation feedback is used to perform incremental training or efficiency compensation logic correction on the water demand prediction model.

10. An intelligent pump allocation analysis system based on flow prediction, characterized in that, A method for implementing an intelligent pump allocation analysis method based on flow prediction as described in any one of claims 1-9 includes: The data acquisition and communication interface module is used to collect real-time operating data of water pumps and real-time monitoring data of pipeline network, and synchronize historical performance curve data and cumulative maintenance records. The individual digital profile management module is connected to the data acquisition and communication interface module and is used to build and continuously update the individual digital profile of each pump group based on the received data. The collaborative prediction and risk assessment module is connected to the individual digital file management module and is used to call the individual digital files of the pump group to generate a first predicted flow rate value and a burst risk coefficient. The intelligent pump allocation strategy generation module is connected to the collaborative prediction and risk assessment module, and is used to generate an anti-interference pump allocation strategy or a conventional energy-efficient pump allocation scheme based on the first predicted flow value and the explosion risk coefficient. The virtual simulation and adaptive optimization module is connected to the individual digital file management module and the intelligent pump allocation strategy generation module, respectively. It is used to perform virtual simulation based on the individual digital files of the pump group and the generated pump allocation strategy, and optimize the individual digital file management module and the collaborative prediction and risk assessment module based on the comparison feedback between the simulation results and real data.

Images