Water supply intelligent dispatching decision system fusing multi-factor coupling analysis

Abstract

The application discloses a water supply intelligent scheduling decision system based on fusion multi-factor coupling analysis, relates to the technical field of water supply intelligent scheduling, and is characterized in that: original data of a pipe network is collected and pretreated through a unified standard, is divided into a space, a time sequence and a relationship database, and is summarized to a scheduling data subject database; a parameter-fixed offline hydraulic model and a locally dynamically corrected online hydraulic model are constructed in the data in the scheduling data subject database; the water retention time of a pipe section is measured and calculated, and dead water and slow-flow pipe sections are divided; three types of machine learning sub-models of time-sharing water demand, energy consumption-leakage correlation and water quality constraints are trained based on historical samples; water supply and water quality are taken as constraints, energy consumption reduction and leakage reduction are taken as targets, a plurality of scheduling alternative schemes are generated by linkage and coupling of the hydraulic model, the optimal scheduling scheme is output by two rounds of screening through multi-dimensional weighted scoring, and fine intelligent scheduling of the pipe network is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent water supply scheduling technology, and in particular to an intelligent water supply scheduling decision system that integrates multi-factor coupling analysis. Background Technology

[0002] With the acceleration of urbanization and the improvement of residents' living standards, the scale of water supply systems is becoming increasingly large and the structure increasingly complex. Water supply companies not only need to ensure sufficient "quantity" but also ensure the safety of "quality" and the efficiency of "energy". The traditional scheduling mode based on manual experience or a single data source is no longer able to cope with the needs of modern water supply system operation and management that are large-scale, multi-constrained, and dynamically changing.

[0003] Currently, water supply network scheduling technology mainly faces the following challenges: Traditional scheduling lacks systematic predictive models, making it impossible to accurately predict changes in water demand for residents and businesses during different times of the day. This can easily lead to insufficient water supply during peak water usage periods and overpressure redundancy in the pipeline network during off-peak periods. Furthermore, scheduling decisions do not take into account the water quality risks caused by pipeline leakage, unit energy consumption, and water stagnation. The control scheme focuses solely on water supply security, which can easily result in excessively high water pressure at the plant and unreasonable pump matching, exacerbating overpressure leakage and ineffective energy consumption in the pipeline network. Meanwhile, conventional scheduling scheme generation lacks multi-condition simulation optimization methods, making it difficult to generate multiple alternative control schemes in batches. There is also no standardized quantitative evaluation mechanism, making it impossible to objectively score and screen schemes from multiple dimensions such as economy, water supply safety, operational reliability, and water supply security. Scheme selection is highly subjective, and manual scheduling makes it difficult to balance multiple conflicting objectives such as reliable water supply, water quality compliance, optimal energy consumption, and minimum leakage. This has long resulted in high energy consumption in pipeline operation, the emergence of dead water sections in the pipeline leading to water quality exceeding standards, and persistently high leakage in the pipeline, thus restricting the improvement of the refined energy-saving management level of the water supply network.

[0004] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0005] To address the shortcomings of existing water supply networks, such as scattered and difficult-to-unify multi-source data, hydraulic model parameters relying on experience, redundant online modeling calculations, scheduling depending on manual experience, and lack of quantitative evaluation standards for scheme selection, this invention provides a smart water supply scheduling and decision-making system that integrates multi-factor coupled analysis. This system achieves integrated data governance, hierarchical and accurate verification of hydraulic models, multi-dimensional intelligent prediction, multi-constraint optimization to generate schemes, and quantitative optimal output, thereby reducing energy consumption and network leakage while ensuring reliable water supply and water quality compliance.

[0006] The objective of this invention can be achieved through the following technical solution: a smart water supply scheduling and decision-making system integrating multi-factor coupling analysis, comprising: Multi-source aggregation module: Collects and preprocesses multi-source heterogeneous raw data from the water supply network, and stores them into the spatial database, time-series database, and relational database in the scheduling data subject database respectively; Offline-Model Building Module: Based on the scheduling data topic library, the measured data sets of the pipeline network are selected according to the four seasons and peak-valley-flat periods. The steady-state hydraulic mechanism calculation framework of the water supply pipeline network is built. The least squares iterative calibration of solidified parameters is adopted. Five typical working conditions are simulated and the benchmark parameters are organized and stored in the library. Online Model Building Module: Real-time acquisition of instantaneous operating data of the pipeline network, using offline fixed parameters as initial values, combined with real-time benchmark sample collection of field measured data, only iteratively corrects parameters for abnormal pipe sections and pump sets exceeding limits, generates an online hydraulic model, distinguishes dead water and slow flow pipe sections based on the retention time of pipe sections, and forms a ledger of retention pipe sections; The scheduling scheme management module extracts and preprocesses all historical sample data of the pipeline network, selects features to construct three types of prediction sub-models to form a machine learning prediction model group, combines hard constraints, aims to minimize comprehensive power consumption and pipeline leakage, and generates multiple scheduling alternative schemes by combining offline and online coupled hydraulic models. Multi-level scheme screening module: The scheduling candidate schemes are scored from four aspects: economy, safety, reliability and guarantee. Inferior schemes are screened out by weighted economic and safety factors, and then ranked by weighted safety and guarantee factors. The top three ranked schemes are selected as recommended scheduling schemes.

[0007] Preferably, the spatial library records the GIS topology data of the entire regional water supply network, the time series library collects time series data at the hour / day / month / year time granularity, and the relationship library binds the subordinate relationship between pump stations, pipelines and water use blocks, and the relationship between meteorology and water consumption.

[0008] Preferably, the process of storing and analyzing the normalized benchmark parameters is as follows: T1: Select network topology and historical operational test datasets that cover the four seasons and distinguish between off-peak / peak / low-peak water usage periods from the spatial, temporal, and relational databases of the scheduling data subject library; T2: Establish an offline steady-state hydraulic mathematical model with undetermined parameters; T3: Select the measured data of typical historical working conditions in four seasons, peak, flat and valley as fitting samples, take the water pressure at the measuring point and the measured flow rate of the pipe section as the true value benchmark, and adopt the least squares combined with multi-round reverse iterative optimization algorithm to iterate until the simulation relative error of the key measuring points of the entire pipe network converges to the pre-set error allowable threshold. All undetermined parameters are locked and fixed to complete the static solidification of the parameters of the entire pipe network of the offline hydraulic model. T4: Divide into five typical steady-state operating conditions, and based on the offline hydraulic model with solidified parameters, summarize the basic parameters of the entire pipeline network operation under each typical simulation condition, including the steady-state water loss of each pipe section, the pressure loss along the pipeline, the steady-state pressure distribution at nodes, and the core benchmark parameters of the water supply energy consumption and efficiency characteristics of a single pumping station unit.

[0009] Preferably, the analysis process of the online hydraulic model is as follows: S1: Real-time acquisition of instantaneous operational data; S2: Read the friction coefficient and local loss coefficient of the entire pipeline segment of the offline hydraulic model, which have been verified and solidified by historical operating conditions, from the scheduling data subject library, and use them as the initial parameter set of the online hydraulic model; S3: Extract the measured water pressure, measured flow rate of pipe sections, actual water outflow of pumping stations and measured power consumption of units at key nodes of the pipeline network collected online at the same time, and build a real-time benchmarking sample set; S4: Calculate the theoretical pressure and theoretical flow rate, perform difference calculation with the real-time benchmark sample set on-site measured data, and statistically analyze the hydraulic calculation deviations of each pipe section and each pump station location; S5: Based on the preset dynamic fitting optimization algorithm, parameter optimization iteration is carried out. The range of the threshold exceeding the limit is defined according to the hydraulic deviation value of each point. Local pipe sections and corresponding operating pump groups with calculated deviations exceeding the allowable threshold are screened out and marked as abnormal objects. For abnormal objects, the friction coefficient, local loss coefficient and efficiency correction coefficient of a single pump unit are adjusted one by one by setting a small step size iterative method. For normal pipe sections and pump units in operation with hydraulic deviation within the threshold range, the baseline parameters completed by the offline hydraulic model are directly used. S6: Continuously iterate and optimize the local pipe section resistance coefficient and pump group efficiency correction coefficient of the abnormal object until the overall calculation error converges and is less than the preset accuracy threshold, then terminate the parameter iteration calculation of this round and obtain the real-time parameters of the online hydraulic model that are adapted to the current instantaneous pipe network conditions.

[0010] Preferred options also include: S7: Based on the online hydraulic model, obtain the flow direction of water within the pipeline network, the spatial distribution of instantaneous pressure at each node, and the water transport rate of the pipeline section; S8: Read the geometric parameters of each pipe segment from the scheduling data theme library space library, obtain the volume of a single pipe, calculate the theoretical retention time of water for each pipe segment based on the ratio of volume to instantaneous flow rate, and compare and output the slow-flow retention pipe segment and the dead water segment by combining the preset dead water judgment threshold and the overflow flow threshold. S9: Bind the spatial number, water supply zone, and geographical location information to the identified dead water section and slow flow pipe section to form a ledger of the stagnant pipe section.

[0011] Preferably, the process for generating the multiple scheduling alternatives is as follows: Historical full-volume sample data is extracted from the spatial, temporal, and relational databases of the scheduling data subject library, and divided into training, validation, and test sets. Based on the training set, a time-sharing water demand prediction sub-model, an energy consumption-leakage correlation prediction sub-model, and a water quality constraint prediction sub-model are constructed. The validation set is used to verify the three types of sub-models respectively, and the model hyperparameters are repeatedly optimized until the prediction error of the test set meets the preset error requirements, forming a machine learning prediction model group. Relying on the optimization algorithm, the model is iteratively optimized to generate multiple scheduling alternative schemes under the premise that all constraints are met.

[0012] Preferably, the analysis process for the recommended scheduling scheme is as follows: The alternative scheduling schemes are marked as Fg, where g is a natural number greater than zero. The three objective evaluation indicators of each alternative scheduling scheme are obtained: economic score, safety score, reliability score and availability score. The safety and low-power fusion score is obtained by multiplying the economic score by the preset weight coefficient a and the safety score by the preset weight coefficient b. The scheduling candidate schemes corresponding to the safety and low-power fusion scores that are lower than the preset safety and low-power fusion score threshold are eliminated to obtain intermediate scheduling candidate schemes. The security score × preset weight coefficient c + the assurance score × preset weight coefficient d is used to obtain the assurance low leakage adjudication score. Based on the assurance low leakage adjudication score, the intermediate scheduling candidate schemes are sorted from high to low, and the top three in the sorted list are output as recommended scheduling schemes.

[0013] The beneficial effects of this invention are as follows: Breaking down data silos: By unifying data models and interfaces, we achieve the integration of multi-source heterogeneous data in spatial, temporal, and relational data, build a scheduling data theme library, support one-click retrieval and correlation query of multi-source data, and achieve a balance between model accuracy and efficiency: We adopt a collaborative strategy of "offline benchmark solidification + online local correction" to perform small-step iterative correction only on abnormal objects with deviations exceeding the limit, which greatly reduces computational redundancy and improves real-time response capabilities. Based on training time-of-use water demand using full-dimensional historical samples, constructing three types of machine learning sub-models for energy consumption-leakage correlation and water quality constraints, the system accurately predicts changes in water demand, energy consumption and leakage, and water quality risks. Under multiple constraints of water supply and water quality, the system uses a coupled hydraulic model for optimization and solution, generating compliant scheduling schemes in batches. The system is quantitatively scored from four dimensions: economy, safety, reliability, and security. After two rounds of weighted screening, the optimal scheduling scheme is selected and output. Under the premise of ensuring sufficient water supply throughout the region and meeting the water quality standards of the pipeline network, the system effectively reduces system energy consumption and pipeline leakage, achieving safe, energy-saving, efficient, and refined intelligent scheduling of the water supply network. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings; Figure 1 This is a flowchart of the system of the present invention; Figure 2 This is a schematic diagram of the scheduling data subject library structure of the present invention; Figure 3 This is a schematic diagram of the offline model building module analysis. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments; Example 1: Please refer to Figures 1 to 3 As shown, the present invention is a smart water supply scheduling and decision-making system that integrates multi-factor coupling analysis, including a water supply scheduling platform, a multi-source aggregation module, an offline-model building module, an online-model building module, a scheduling scheme management module, and a multi-level scheme screening module; Specifically, the multi-source collection module uniformly collects and processes heterogeneous data from multiple sources in the pipeline network and constructs three types of databases; the offline-model building module establishes a parameter-fixed offline hydraulic model based on the data; the online-model building module realizes local dynamic correction of the online hydraulic model and identifies stranded pipe sections; the scheduling scheme management module trains three types of machine learning sub-models; multi-constraint simulation generates alternative scheduling schemes; and the multi-level scheme screening module selects the optimal scheme through two rounds of weighted screening using four-dimensional indicators, thereby achieving intelligent scheduling of the pipeline network with energy saving and quality assurance.

[0017] Multi-source aggregation module: Collects and preprocesses heterogeneous raw data from multiple sources in the water supply network, and stores it into the spatial, temporal, and relational databases of the scheduling data subject database, respectively. Based on a unified data model and standardized interface, multi-source heterogeneous raw data of water supply network is collected. Missing value completion, abnormal data removal, and full data dimension unification preprocessing are performed on the multi-source heterogeneous raw data in sequence. After preprocessing, the data are stored in spatial database, time series database, and relational database respectively. The data in spatial database, time series database, and relational database are uniformly summarized and stored in the scheduling data theme database, realizing one-click retrieval of multi-source data and cross-database association query. Through a unified data model and interface, the system collects multi-source heterogeneous data of the target water supply network, completes missing value completion, abnormal data removal, and unit unification processing of the multi-source heterogeneous data, and constructs spatial, temporal, and relational databases. The data from the spatial, temporal, and relational databases are then uniformly integrated into the scheduling data theme database, enabling one-click retrieval and related queries of multi-source data, and providing unified data for subsequent model calculations. Spatial Library: Input the GIS topology data of the entire regional water supply network, including the location of water plants, coordinates of pump stations at all levels, pipe diameter / length / material, valve locations, spatial boundaries of zonal metering blocks, and geographical information of water quality monitoring stations, and complete the archiving of spatial attributes of all water supply facilities; Time Series Library: Time series data is collected by hour / day / month / year time granularity, covering historical and real-time water output, water pressure at pipeline nodes, pump station start-up and shutdown sequence, unit power consumption, end-point water flow, pipeline water quality indicators (residual chlorine, turbidity, pH value), meteorological parameters (temperature, precipitation, seasonal climate), and time series data of residential / industrial and commercial time-of-use water demand. Relationship database: Establish coupling and mapping relationships between various data, bind the subordinate relationship between pump station-pipeline-water use block, the relationship between meteorology and water consumption, the linkage relationship between unit operation parameters-energy consumption and pipeline pressure, and the corresponding relationship between water retention time and water quality changes.

[0018] Offline Model Building Module: Based on the scheduling data topic library, this module selects measured datasets of water supply networks for each season and peak / valley / flat period, builds a calculation framework for the steady-state hydraulic mechanism of water supply networks, uses least squares iterative calibration to solidify parameters, simulates five typical operating conditions, and standardizes and stores the baseline parameters in the database. Specifically, it includes: T1: From the spatial, temporal, and relational databases of the scheduling data subject library, filter the pipeline topology and historical operation measurement datasets that cover the four seasons and distinguish between off-peak / peak / low-peak water usage periods. The datasets include information such as pipeline material, pipe diameter, laying years, historical node pressure, pipe section flow, and measured records of pump station unit output and power consumption. T2: Establish an offline steady-state hydraulic mathematical model with undetermined parameters: Based on the fluid continuity equation, energy equation, and head loss calculation formula, a calculation framework for the steady-state hydraulic mechanism of the water supply network is built. The model follows the hydraulic constraints of nodal flow conservation and the algebraic sum of head loss in closed loops being zero. The pipeline friction coefficient, valve elbow local loss coefficient, and pump station unit head-efficiency characteristic parameters are taken as unknown variables to be checked, and an offline steady-state hydraulic mathematical model with undetermined parameters is established (representing the offline hydraulic model). T3: Select typical historical operating conditions of four seasons and peak, flat and valley as fitting samples, and use the water pressure at the measuring point and the measured flow rate of the pipe section as the true value benchmark. Use the least squares combined with multi-round reverse iterative optimization algorithm to compare the deviation of the model simulation output pressure and flow rate calculation values ​​with the field measured data node by node and pipe section by pipe section. Step-by-step iterative correction of the friction coefficient of the pipeline network, the local loss coefficient of valves, elbows and pipe fittings and the head-efficiency characteristic parameters of the pump station unit. Iterative processing is carried out until the simulation relative error of the key measuring points of the entire pipeline network converges to the pre-set error allowable threshold. All undetermined parameters are locked and fixed, and the offline hydraulic model of the entire pipeline network parameters is statically solidified to obtain the offline hydraulic model. T4: Five typical steady-state operating conditions are defined: dry season, wet season, daytime peak water consumption, daytime low water consumption, and nighttime low water consumption. Based on the offline hydraulic model with fixed parameters, steady-state hydraulic simulation calculations of the entire pipeline network are carried out for each typical steady-state operating condition. The basic operating parameters of the entire pipeline network under each typical simulation condition are summarized, including the steady-state water loss of each pipe section, the pressure loss value along the pipeline, the steady-state pressure distribution of nodes, and the water consumption and efficiency characteristics of a single pumping unit. The core benchmark parameters of all scenarios are classified, standardized, and quantified, and then archived in batches into the integrated scheduling data theme library. The fixed offline static benchmark parameters under steady-state conditions of multiple scenarios are set as the core reference basis, providing a standardized and traceable benchmark data source for subsequent real-time dynamic parameter correction of the online hydraulic model, iterative calculation of multi-factor coupled scheduling optimization algorithms, and intelligent scheduling scheme derivation.

[0019] Example 2: Online Model Building Module: Real-time acquisition of instantaneous operating data of the pipeline network, using offline fixed parameters as initial values, combined with real-time benchmarking sample data from on-site measurements, iteratively correcting parameters only for abnormal pipe sections and pump sets exceeding limits, generating an online hydraulic model, distinguishing between dead and slow-flowing pipe sections based on the retention time of pipe sections, forming a ledger of retained pipe sections, with specific analysis process as follows: S1: Real-time operation data is collected through the water plant PLC control system, pump station online acquisition terminal, pipeline pressure / flow / water quality online monitoring points, and DMA zone metering equipment. The real-time operation data includes instantaneous outflow, measured water pressure at nodes, pump unit operating frequency and instantaneous power consumption, and flow rate at key sections of the pipeline network. S2: Read the friction coefficient, local loss coefficient, etc. of the entire pipeline segment of the offline hydraulic model after historical working condition verification from the scheduling data subject library, and use them as the initial parameter set of the online hydraulic model; S3: Extract the measured water pressure, measured flow rate of pipe sections, actual water outflow of pumping stations and measured power consumption of units at key nodes of the pipeline network collected online at the same time, and build a real-time benchmarking sample set; S4: Substitute the initial reference parameters from the initial parameter set into the hydraulic equations (such as the pipeline energy equation, the head loss formula, and the pump characteristic equation) to calculate the theoretical pressure and theoretical flow rate. Perform a difference calculation with the field measured data in the real-time benchmark sample set, and statistically analyze the hydraulic calculation deviations for each pipe section and each pump station location. S5: Based on the preset dynamic fitting optimization algorithm (one or more of the following: gradient descent iterative fitting algorithm, least squares fitting algorithm, Kalman filter dynamic correction algorithm), carry out parameter optimization iteration, define the over-limit threshold range according to the hydraulic deviation value of each point, screen out the local pipe section and corresponding operating pump group whose calculated deviation exceeds the allowable threshold, and mark them as abnormal objects. For abnormal objects, the friction coefficient, local loss coefficient and efficiency correction coefficient of a single pump unit are adjusted one by one by setting a small step size iterative method. For normal pipe sections and pump units in operation with hydraulic deviation within the threshold range, the baseline parameters completed by the offline hydraulic model are directly used without performing parameter correction operations, thereby reducing the computational redundancy caused by global parameter adjustment while ensuring the overall accuracy of the model. S6: Continuously iterate and optimize the local pipe section resistance coefficient and pump group efficiency correction coefficient of the abnormal object, and compare the error value of the hydraulic parameters calculated by the model theory with the actual measured hydraulic parameters on site until the overall calculation error converges and is less than the preset accuracy threshold. Then, terminate the parameter iteration calculation and obtain the real-time parameters of the online hydraulic model that are adapted to the current instantaneous pipe network conditions. That is, the offline static reference parameters are corrected in real time through the above dynamic correction mechanism, and the instantaneous operating parameters of the online hydraulic model with high fitting degree are output, which are adapted to the current instantaneous operating conditions of the pipeline network. S7: Based on the online hydraulic model, obtain information such as the flow direction of water within the pipeline network, the spatial distribution of instantaneous pressure at each node, and the water transport rate of the pipeline section; S8: Read the geometric parameters of each pipe segment from the scheduling data theme library space library, obtain the volume of a single pipe, and calculate the theoretical retention time of water for each pipe segment based on the ratio of volume to instantaneous flow rate. Formula: Retention time = Effective volume of pipe segment / Real-time flow rate of pipe segment; The system presets a dead water threshold and an overflow threshold. When the water retention time in a pipe section is less than the preset dead water threshold and the real-time overflow is greater than the preset overflow threshold, the water in that pipe section is judged to be flowing slowly and is marked as a slow-flowing stagnant pipe section. When the water retention time in a pipe section is greater than or equal to the preset dead water threshold, or the real-time overflow is less than or equal to the preset overflow threshold, the water in that pipe section is judged to be stagnant for a long time and is marked as a dead water section. S9: Bind spatial number, water supply zone, and geographical location information to the identified dead water section and slow flow pipe section to form a ledger of the stagnant pipe section; The dwell time and dwell segment marking information of each pipe section of the entire pipeline network will be stored in the scheduling data subject database time series database, which will serve as the input basis for water quality constraint verification and pipeline network scheduling pressure regulation.

[0020] Example 3: Scheduling Scheme Management Module: Extracts and preprocesses all historical sample data of the pipeline network, selects features to construct three types of prediction sub-models to form a machine learning prediction model group, and combines hard constraints with the goal of minimizing overall power consumption and pipeline leakage. It also combines offline and online coupled hydraulic models to generate multiple alternative scheduling schemes, specifically including: Historical full sample data is extracted from the spatial, temporal, and relational databases of the scheduling data subject library. The sample data includes: historical pipeline hydraulic simulation benchmark parameters, hourly outflow water volume, node pressure time series, pump station unit start-up and shutdown and energy consumption data, whole pipeline water quality monitoring indicators, daily meteorological information, time-sharing water demand of residents and industrial and commercial users, historical scheduling execution plans, and actual leakage and energy consumption results after the plans were implemented. The extracted sample data were uniformly processed by noise reduction, outlier removal, and feature normalization. Feature label association was established based on the relational database. Water consumption, temperature, rainfall, season, electricity price, pipeline retention time, water quality limit, and unit operating efficiency were selected as model input features. System comprehensive energy consumption, pipeline leakage rate, water quality compliance status, and water supply guarantee margin were used as model output labels. Training set, validation set, and test set were divided. Based on the training set, construct a time-of-use water demand prediction sub-model, an energy consumption-leakage correlation prediction sub-model, and a water quality constraint prediction sub-model; Time-of-use water demand prediction sub-model: Using a time-series machine learning algorithm, the water demand prediction model is trained based on the correlation characteristics between historical water use and meteorological features to achieve accurate prediction of future daily / time-of-use water consumption for residents and industrial and commercial users, and output the time-of-use water demand boundary conditions for different areas. Construct an energy consumption-leakage correlation prediction sub-model: Based on sample data, train a regression learning model to explore the coupling mapping law between plant pressure, pump station start-up combination, water load, seasonal operating conditions and system power consumption, pipeline overpressure leakage, and establish a quantitative prediction relationship between energy consumption loss and leakage. Water quality constraint prediction sub-model: Using water retention time, water conveyance conditions and influent water quality as inputs, a water quality change prediction model is trained to predict the trend of residual chlorine and turbidity changes at key points in the pipeline network under different scheduling conditions, and to identify the conditions at which water quality exceeds the standard. The validation set is used to verify the three types of sub-models respectively, and the hyperparameters of the models are repeatedly optimized until the prediction error of the test set meets the preset error requirements, forming a machine learning prediction model group. The hard constraints are to meet the water supply demand of the whole region and the water quality of the whole pipeline network. The optimization objectives are to minimize the comprehensive power consumption and the pipeline network leakage. Multi-dimensional constraint factors are embedded in the optimization function: water quantity constraint (lower limit of water supply flow at the end of each zone), water quality constraint (national standard limit of key pipeline network indicators, maximum allowable retention time of water body), energy consumption constraint (upper limit of unit water production energy consumption of single pump unit, peak and valley electricity price constraint of the power grid, etc.). The time-of-use water demand and meteorological change parameters for the future target period, predicted by the machine learning prediction model group, are used as boundary inputs. Simultaneously, the offline + online coupled hydraulic model is called to carry out multi-condition iterative simulation calculations. Within the optimization solution space, adjustable scheduling variables such as the water plant outlet pressure setpoint, the number of pump station units in operation, the combination of unit start-up and shutdown time periods, and the opening degree of zone valves are changed. Each combination of variables corresponds to a set of pipeline network operating conditions. Based on the optimization algorithm, multiple sets of differentiated scheduling alternatives are generated in batches, all of which meet the bottom line requirements of water supply and water quality, under the premise that all constraints are compliant. Each alternative is accompanied by supporting simulation indicators: estimated total power consumption, predicted leakage of the pipeline network, water quality compliance of the entire pipeline network, start-up details of each pumping station, and factory pressure control parameters, etc. Multi-level scheme screening module: This module scores candidate scheduling schemes based on four criteria: economy, safety, reliability, and assurance. Inferior schemes are eliminated through a weighted average of economic and safety scores. Then, a weighted average of safety and assurance scores is used to rank the remaining schemes, and the top three are selected as recommended scheduling schemes. Specifically, these include: The alternative scheduling schemes are marked as Fg, where g is a natural number greater than zero. The three objective evaluation indicators of each alternative scheduling scheme are obtained: economic score, safety score, reliability score and availability score. Economic efficiency score: Obtain the comprehensive power consumption of the scheme. Using the benchmark energy consumption as a reference, the value obtained by subtracting the benchmark energy consumption from the comprehensive power consumption of the scheme is set as a score evaluation item. Positive values ​​represent deductions and negative values ​​represent additions. The absolute value of the score evaluation item is matched with a preset interval, and the score value set in the preset interval to which the match belongs is output. Based on the positive or negative value of the score evaluation item, the preset benchmark score (comprehensive power consumption of the scheme = preset benchmark score of benchmark energy consumption) is added or subtracted to output the economic efficiency score. Safety rating: Set the compliance rate range of the benchmark water quality points in the pipeline network, set the score within each benchmark water quality point compliance rate range, match the compliance rate of the pipeline network water quality points with the benchmark water quality point compliance rate range, output the matching score, and mark it as the safety rating. The higher the safety rating value, the higher the compliance rate of the pipeline network water quality points. The safety and low-power fusion score is obtained by multiplying the economic score by the preset weight coefficient a and the safety score by the preset weight coefficient b. The scheduling candidate schemes corresponding to the safety and low-power fusion scores that are lower than the preset safety and low-power fusion score threshold are eliminated to obtain intermediate scheduling candidate schemes. Reliability Score: Obtain the predicted pipeline leakage of the proposed solution. Using the baseline leakage as a reference, subtract the baseline leakage from the predicted pipeline leakage and set the result as a leakage score evaluation item. Positive values ​​represent deductions, and negative values ​​represent additions. Match the absolute value of the leakage score evaluation item with a preset interval and output the leakage score value set for the preset interval to which the match belongs. Based on the sign of the leakage score evaluation item, add or subtract from the preset baseline score (predicted pipeline leakage = preset baseline score of baseline leakage) and output the reliability score. Guarantee score: Based on the total water demand of each zone, the actual water supply satisfaction ratio is obtained. The guarantee score is calculated by multiplying the actual water supply satisfaction ratio by 100. The higher the guarantee score, the better the guarantee of the plan. The minimum leakage protection score is obtained by combining the safety score × preset weight coefficient c and the security score × preset weight coefficient d. The intermediate scheduling candidates are sorted from high to low according to the minimum leakage protection score. The top three in the sorted list are output as recommended scheduling schemes, and the water supply scheduling platform directly displays the recommended scheduling schemes in sequence.

[0021] In summary, by unifying and aggregating multi-source heterogeneous water supply data, establishing separate spatial, temporal, and relational databases, and summarizing them into a scheduling theme database, a regular and traceable basic data source is provided for various models. Based on the measured data of multiple working conditions in four seasons, the solidified parameters of the offline hydraulic model are iteratively verified to form multi-scenario benchmark parameters. The online hydraulic model reuses the offline benchmark parameters, and only dynamically adjusts the parameters of local pipe sections and pump groups with deviations exceeding the limits, which greatly reduces computational redundancy, balances computational efficiency and simulation accuracy, and can also quantify the water retention time of pipe sections, accurately identify dead water and slow-flowing pipe sections, and support the water quality management of the pipe network. Based on training time-of-use water demand using full-dimensional historical samples, constructing three types of machine learning sub-models for energy consumption-leakage correlation and water quality constraints, the system accurately predicts changes in water demand, energy consumption and leakage, and water quality risks. Under multiple constraints of water supply and water quality, the system uses a coupled hydraulic model for optimization and solution, generating compliant scheduling schemes in batches. The system is quantitatively scored from four dimensions: economy, safety, reliability, and security. After two rounds of weighted screening, the optimal scheduling scheme is selected and output. Under the premise of ensuring sufficient water supply throughout the region and meeting the water quality standards of the pipeline network, the system effectively reduces system energy consumption and pipeline leakage, achieving safe, energy-saving, efficient, and refined intelligent scheduling of the water supply network.

[0022] The threshold is set for result comparison and analysis to determine whether it is good or bad. The value of the threshold is determined by a combination of large-scale model analysis of the sample data and human experience, and can also be adjusted appropriately based on seasonal or common-sense influencing factors. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart water supply scheduling and decision-making system integrating multi-factor coupling analysis, characterized in that, include: Multi-source aggregation module: Collects and preprocesses multi-source heterogeneous raw data from the water supply network, and stores them into the spatial database, time-series database, and relational database in the scheduling data subject database respectively; Offline-Model Building Module: Based on the scheduling data topic library, the measured data sets of the pipeline network are selected according to the four seasons and peak-valley-flat periods. The steady-state hydraulic mechanism calculation framework of the water supply pipeline network is built. The least squares iterative calibration of solidified parameters is adopted. Five typical working conditions are simulated and the benchmark parameters are organized and stored in the library. Online Model Building Module: Real-time acquisition of instantaneous operating data of the pipeline network, using offline fixed parameters as initial values, combined with real-time benchmark sample collection of field measured data, only iteratively corrects parameters for abnormal pipe sections and pump sets exceeding limits, generates an online hydraulic model, distinguishes dead water and slow flow pipe sections based on the retention time of pipe sections, and forms a ledger of retention pipe sections; The scheduling scheme management module extracts and preprocesses all historical sample data of the pipeline network, selects features to construct three types of prediction sub-models to form a machine learning prediction model group, combines hard constraints, aims to minimize comprehensive power consumption and pipeline leakage, and generates multiple scheduling alternative schemes by combining offline and online coupled hydraulic models. Multi-level scheme screening module: The scheduling candidate schemes are scored from four aspects: economy, safety, reliability and guarantee. Inferior schemes are screened out by weighted economic and safety factors, and then ranked by weighted safety and guarantee factors. The top three ranked schemes are selected as recommended scheduling schemes.

2. The intelligent water supply scheduling and decision-making system integrating multi-factor coupling analysis according to claim 1, characterized in that, The spatial database records the GIS topology data of the entire regional water supply network, the time series database collects time series data at the hourly / daily / monthly / yearly time granularity, and the relational database binds the subordinate relationship between pump stations, pipelines and water use blocks, and the relationship between meteorology and water consumption.

3. The intelligent water supply scheduling and decision-making system integrating multi-factor coupling analysis according to claim 1, characterized in that, The process of inputting and analyzing the normalized benchmark parameters into the database is as follows: T1: Select network topology and historical operational test datasets that cover the four seasons and distinguish between off-peak / peak / low-peak water usage periods from the spatial, temporal, and relational databases of the scheduling data subject library; T2: Establish an offline steady-state hydraulic mathematical model with undetermined parameters; T3: Select the measured data of typical historical working conditions in four seasons, peak, flat and valley as fitting samples, take the water pressure at the measuring point and the measured flow rate of the pipe section as the true value benchmark, and adopt the least squares combined with multi-round reverse iterative optimization algorithm to iterate until the simulation relative error of the key measuring points of the entire pipe network converges to the pre-set error allowable threshold. All undetermined parameters are locked and fixed to complete the static solidification of the parameters of the entire pipe network of the offline hydraulic model. T4: Divide into five typical steady-state working conditions, and summarize the core benchmark parameters under each typical simulation working condition based on the offline hydraulic model with solidified parameters.

4. The intelligent water supply scheduling and decision-making system integrating multi-factor coupling analysis according to claim 1, characterized in that, The analysis process of the online hydraulic model is as follows: S1: Real-time acquisition of instantaneous operational data; S2: Read the friction coefficient and local loss coefficient of the entire pipeline segment of the offline hydraulic model, which have been verified and solidified by historical operating conditions, from the scheduling data subject library, and use them as the initial parameter set of the online hydraulic model; S3: Extract the measured water pressure, measured flow rate of pipe sections, actual water outflow of pumping stations and measured power consumption of units at key nodes of the pipeline network collected online at the same time, and build a real-time benchmarking sample set; S4: Calculate the theoretical pressure and theoretical flow rate, perform difference calculation with the real-time benchmark sample set on-site measured data, and statistically analyze the hydraulic calculation deviations of each pipe section and each pump station location; S5: Based on the preset dynamic fitting optimization algorithm, parameter optimization iteration is carried out. The range of the threshold exceeding the limit is defined according to the hydraulic deviation value of each point. Local pipe sections and corresponding operating pump groups with calculated deviations exceeding the allowable threshold are screened out and marked as abnormal objects. For abnormal objects, the friction coefficient, local loss coefficient and efficiency correction coefficient of a single pump unit are adjusted one by one by setting a small step size iterative method. For normal pipe sections and pump units in operation with hydraulic deviation within the threshold range, the baseline parameters completed by the offline hydraulic model are directly used. S6: Continuously iterate and optimize the local pipe section resistance coefficient and pump group efficiency correction coefficient of the abnormal object until the overall calculation error converges and is less than the preset accuracy threshold, then terminate the parameter iteration calculation of this round and obtain the real-time parameters of the online hydraulic model that are adapted to the current instantaneous pipe network conditions.

5. The intelligent water supply scheduling and decision-making system integrating multi-factor coupling analysis according to claim 4, characterized in that, Also includes: S7: Based on the online hydraulic model, obtain the flow direction of water within the pipeline network, the spatial distribution of instantaneous pressure at each node, and the water transport rate of the pipeline section; S8: Read the geometric parameters of each pipe segment from the scheduling data theme library space library, obtain the volume of a single pipe, calculate the theoretical retention time of water for each pipe segment based on the ratio of volume to instantaneous flow rate, and compare and output the slow-flow retention pipe segment and the dead water segment by combining the preset dead water judgment threshold and the overflow flow threshold. S9: Bind the spatial number, water supply zone, and geographical location information to the identified dead water section and slow flow pipe section to form a ledger of the stagnant pipe section.

6. The intelligent water supply scheduling and decision-making system integrating multi-factor coupling analysis according to claim 1, characterized in that, The process of generating the multiple scheduling alternatives is as follows: Historical full-volume sample data is extracted from the spatial, temporal, and relational databases of the scheduling data subject library, and divided into training, validation, and test sets. Based on the training set, a time-sharing water demand prediction sub-model, an energy consumption-leakage correlation prediction sub-model, and a water quality constraint prediction sub-model are constructed. The validation set is used to verify the three types of sub-models respectively, and the model hyperparameters are repeatedly optimized until the prediction error of the test set meets the preset error requirements, forming a machine learning prediction model group. Relying on the optimization algorithm, the model is iteratively optimized to generate multiple scheduling alternative schemes under the premise that all constraints are met.

7. The intelligent water supply scheduling and decision-making system integrating multi-factor coupling analysis according to claim 1, characterized in that, The analysis process for the recommended scheduling scheme is as follows: The alternative scheduling schemes are marked as Fg, where g is a natural number greater than zero. The three objective evaluation indicators of each alternative scheduling scheme are obtained: economic score, safety score, reliability score and availability score. The safety and low-power fusion score is obtained by multiplying the economic score by the preset weight coefficient a and the safety score by the preset weight coefficient b. The scheduling candidate schemes corresponding to the safety and low-power fusion scores that are lower than the preset safety and low-power fusion score threshold are eliminated to obtain intermediate scheduling candidate schemes. The security score × preset weight coefficient c + the assurance score × preset weight coefficient d is used to obtain the assurance low leakage adjudication score. Based on the assurance low leakage adjudication score, the intermediate scheduling candidate schemes are sorted from high to low, and the top three in the sorted list are output as recommended scheduling schemes.

Images