Multilevel optimization scheduling method and system for sensing operation state of pump station

Abstract

The invention provides a pump station operation state sensing multi-level optimization scheduling method and system, relates to the technical field of pump station operation state data processing, and constructs a pump station optimization scheduling model considering optimal energy consumption, equipment loss balance and water supply stability by taking a pump station historical operation state record as training data. Three types of algorithms of system-level comprehensive optimization, unit-level operation optimization and pipeline system optimization are introduced into the model, and an input association mechanism is adopted, so that flow quota, unit combination and pipe network distribution are subjected to collaborative optimization under a unified framework, and local optimization caused by mutual separation of system scheduling, single machine operation and pipe network operation is avoided. Through real-time sensing of the operation state and the total flow of each water pump, the model adaptively outputs the optimal water pump combination and flow distribution at the current moment, the accuracy of energy consumption evaluation and flow configuration is improved, meanwhile, unnecessary start and stop and local overload are reduced, and the continuity of the water supply process and the performability of a scheduling result are enhanced.

Description

Technical Field

[0001] This invention relates to the field of pump station operation status data processing technology, and in particular to a multi-level optimized scheduling method and system for pump station operation status perception. Background Technology

[0002] The pump station operation and dispatch system is the core hub for ensuring the stable and efficient operation of the water supply network. Its core function is to dynamically coordinate the operating parameters of the pump units, valves and frequency converters in the pump station based on the real-time operating status of the water supply network (including network pressure, reservoir water level and equipment condition) and time-sharing water demand, so as to achieve a reasonable allocation of water lifting volume and supply-demand balance, while meeting basic constraints such as network end pressure safety and equipment operation and maintenance specifications.

[0003] To further improve the accuracy of scheduling, a rolling optimization scheduling process of prediction-optimization-issuance-feedback is typically adopted: Supported by the existing SCADA / PLC data acquisition system and historical database, the system aggregates time-of-use water demand and electricity price information, combines water level, pressure, and equipment status, and triggers optimization calculations according to fixed time windows to form a water extraction volume and quota plan for the next time period. Subsequently, instructions are issued to the frequency converter and valve execution layer via a host computer or gateway, and the planned and actual results are compared through monitoring feedback for verification and correction in the next window. This process aims to produce executable plans on time, emphasizing reducing overall electricity costs while meeting the daily total water extraction volume and end-pressure safety boundaries. However, its accuracy highly depends on the timeliness and consistency of the input data, as well as the accuracy of the characterization of the actual equipment performance.

[0004] To support the above process, existing pump station operation optimization and scheduling systems typically consist of a data acquisition and preprocessing layer, a prediction layer, an optimization decision-making layer, and a human-machine interaction / execution layer. The data acquisition and preprocessing layer collects field data through sensors such as pressure, flow, level, and electrical parameters, as well as PLCs / RTUs, primarily performing basic noise reduction, anomaly removal, and simple interpolation. The prediction layer often uses statistical methods or empirical models to make short-term predictions of time-of-use water demand and time-of-use electricity prices, and based on this, it calls upon pump station design data or factory sample curves to provide the approximate available operating range of the units. The optimization decision-making layer, under given demand forecasts and empirical characteristic constraints, uses mixed-integer linear programming or conventional heuristic algorithms to solve for water pumping quotas and start / stop / frequency combinations for each time period. The human-machine interaction / execution layer is responsible for displaying the optimization results to dispatchers, supporting manual review and one-click deployment, and feeding back key monitoring data during the execution process for subsequent window corrections. While the overall architecture emphasizes ease of implementation and maintenance, the reliance on data governance and equipment characteristic modeling directly impacts the reliability and precision of the scheduling results in actual operation.

[0005] Under the above-described process and system composition, the existing technology still has the following technical problems regarding the accuracy of results:

[0006] On the one hand, while traditional scheduling has accumulated a large amount of historical data on start / stop status, frequency, flow rate, and pressure, it largely remains at the level of real-time monitoring and post-event querying. It lacks systematic processing mechanisms such as data cleaning, time alignment, and feature extraction for optimization modeling. This results in the inability to reconstruct the actual operating conditions and energy efficiency levels of each time period stably and accurately from the original operation logs, leading to significant uncertainties and biases in the data foundation used for modeling and decision-making. On the other hand, existing optimization models largely rely on design or ideal operating condition curves, making it difficult to reflect the performance degradation and characteristic drift of pumps due to wear and changes in operating conditions during long-term operation. This creates a significant discrepancy between the head-flow-efficiency relationship used in scheduling calculations and the actual characteristics of the equipment. The persistent systemic biases significantly distort the assessment of the electricity cost impact and energy-saving potential of different start-stop combinations and frequency configurations under time-of-use pricing. Furthermore, existing technologies lack quantitative trade-offs and global optimization capabilities for multiple objectives such as peak shaving and valley filling, high-efficiency pump operation, and start-stop shock suppression. The generated operating schemes do not match the future demand curve and electricity price periods adequately in terms of time series, easily leading to inaccurate assessments of water supply security margins, increased end-pressure deviations in localized periods, and persistently high long-term energy consumption levels. This makes it difficult to fully exploit energy-saving potential and hinders the stable reuse of scheduling strategies across different operating conditions and pump stations. Consequently, this poses a substantial constraint on the accuracy, reliability, and replicability of pump station operation scheduling results in practical engineering applications. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a multi-level optimized scheduling method and system for pump station operation status perception, which can achieve high accuracy and reproducibility of pump station operation scheduling results.

[0008] The technical solution of this invention is implemented as follows:

[0009] This invention provides a multi-level optimized scheduling method for pump station operation status perception. The method includes: using historical operation status records of the pump station as training data to construct a pump station optimized scheduling model that can achieve optimal pump station operation energy consumption, balanced equipment loss, and improved water supply stability; providing algorithmic support for the pump station optimized scheduling model, which includes a system-level comprehensive optimization algorithm for optimizing pump station resource allocation, a unit-level operation optimization algorithm for reducing pump station unit energy consumption, and a pipeline system optimization algorithm for reducing pump station energy loss, all working together for the global scheduling and management of the pump station. An input association mechanism is used between the algorithms; the output of the system-level comprehensive optimization algorithm serves as part of the input to the unit-level operation optimization algorithm, and also serves as part of the input to the pipeline system optimization algorithm. The system-level comprehensive optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm all require additional acquisition of the operation status of each pump belonging to the pump station; perceiving the operation status and total flow of each pump belonging to the pump station and inputting them into the pump station optimized scheduling model; the pump station optimized scheduling model outputs the optimal pump combination and pump flow, achieving multi-level optimized scheduling of the pump station.

[0010] This application embodiment also provides a multi-level optimized scheduling system for pump station operation status awareness. The system includes: a model building module, used to construct an optimized scheduling model for pump stations using historical operation status records as training data, capable of optimizing pump station operation energy consumption, balancing equipment losses, and improving water supply stability; and an algorithm composition module, used to provide algorithmic support for the optimized scheduling model, which includes a system-level comprehensive optimization algorithm for optimizing pump station resource allocation, a unit-level operation optimization algorithm for reducing pump station unit energy consumption, and a pipeline system optimization algorithm for reducing pump station energy loss, all working together to optimize the pump station's operation. Global scheduling and management employ an input correlation mechanism between algorithms. The output of the system-level comprehensive optimization algorithm serves as part of the input to the unit-level operation optimization algorithm, and also as part of the input to the pipeline system optimization algorithm. The system-level comprehensive optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm all require additional information on the operating status of each pump belonging to the pumping station. The optimization scheduling module is used to sense the operating status and total flow of each pump belonging to the pumping station and input it into the pumping station optimization scheduling model. The pumping station optimization scheduling model outputs the optimal pump combination and pump flow, realizing multi-level optimization scheduling of the pumping station.

[0011] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0012] (1) This invention uses historical operating status records of pumping stations as training data to construct an optimized pumping station scheduling model that can simultaneously achieve optimal energy consumption, balanced equipment losses, and improved water supply stability. This overcomes the technical problems in existing technologies, such as excessive reliance on human experience and ideal operating condition curves for scheduling decisions, difficulty in timely and accurately reflecting the actual operating conditions and energy efficiency levels of each pump, and resulting in significant deviations between scheduling results and actual operation. On this basis, the model is supported by algorithms consisting of a system-level integrated optimization algorithm, a unit-level operation optimization algorithm, and a pipeline system optimization algorithm. An input association mechanism is adopted, using the flow quota and economic target output by the system-level integrated optimization algorithm as part of the input to the unit-level operation optimization algorithm, and then using the unit combination and flow allocation output by the unit-level operation optimization algorithm as part of the pipeline system optimization algorithm. The system incorporates a portion of the inputs to the optimization algorithm, thereby achieving coordinated optimization of pump station resource allocation, unit energy consumption, and pipeline energy loss within the same framework. This effectively solves the technical problems in existing technologies, such as the disconnect between system scheduling, single-unit operation, and pipeline distribution, the tendency to fall into local optima, and the difficulty in balancing overall energy efficiency and hydraulic condition coordination. Simultaneously, by real-time sensing of the operating status and total flow of each pump at the pump station and inputting it into the pump station optimization scheduling model, the model adaptively outputs the optimal pump combination and the flow of each pump at the current moment. This allows the scheduling scheme to dynamically adapt to actual operating conditions, improving the accuracy of energy consumption assessment and flow allocation while reducing unnecessary start-ups and shutdowns and local overload operation. It significantly improves equipment loss distribution and hydraulic impact risk, enhancing the continuity of the water supply process and the executability of the scheduling results.

[0013] (2) When providing algorithmic support for the pump station optimization scheduling model, a hierarchical data sharing and input association mechanism is established between the system-level integrated optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm, so that the three types of algorithms can work together under a unified constraint system: the system-level integrated optimization algorithm provides a preliminary flow allocation scheme under the premise of satisfying the global constraints such as the total daily water lifting volume, the time-sharing electricity price, and the safety boundaries of water level and pressure, which serves as the input to the unit-level operation optimization algorithm. This ensures that the unit level is no longer isolated from the system objective when making start-up and shutdown combinations and operating point selections, thus avoiding the deviation from the global energy consumption optimization due to the pursuit of local high efficiency. The flow demand and pipeline section output by the system-level integrated optimization algorithm are used as input to the unit-level operation optimization algorithm. The water supply pressure at the point is transmitted to the pipeline system optimization algorithm, which ensures that the flow distribution, hydraulic loss, and pressure constraint solutions on the pipeline side are based on the actual achievable unit operating conditions, avoiding the problems of unachievable flow targets or hydraulic condition mismatch. At the same time, all three types of algorithms introduce additional constraints on the real-time operating status and historical characteristics of each pump during the data sharing process, so that the global quota, single-unit operation, and pipeline distribution are consistent in terms of time sequence and physical boundaries. This overcomes the technical problems of existing technologies, such as the independent operation of system scheduling, unit operation, and pipeline distribution, inconsistent constraint standards, easy getting trapped in local optima, and insufficient coordination of hydraulic conditions. It significantly improves the accuracy and reliability of scheduling results in terms of energy consumption assessment, pressure control, and scheme feasibility.

[0014] (3) This invention also continuously monitors and diagnoses the multi-level data sharing links, and promptly implements self-consistency deviation suppression measures when it is determined that there is an input inconsistency anomaly caused by pseudo deadlock. This can reduce the interference of inconsistent timeliness, version mismatch and inconsistent statistical standards on the multi-level optimization scheduling results from the data source. On the one hand, this process can proactively identify abnormal scenarios where upstream data is waited for a long time but the state has deviated significantly from the actual situation before the pseudo deadlock trend is formed. This avoids the system from repeatedly staying in the waiting state in the same window, causing the optimization algorithms at the system level, unit level and pipeline level to solve based on half new and half old inputs, thereby causing end pressure estimation deviation, energy consumption assessment distortion and feasible region determination error. On the other hand, by triggering self-consistency deviation suppression measures for abnormal windows, the data flow is limited, snapshot rollback or weight adjustment is performed, so that the algorithms at each level can prioritize optimization calculation based on self-consistent and interpretable state snapshots in the current window, effectively reducing the risk of non-convergence caused by livelock oscillation and long tail delay. Therefore, the multi-level optimization scheduling process, while maintaining continuous operation, can stably produce scheduling schemes that are more consistent with the on-site working conditions, significantly improving the accuracy and reliability of pump station operation energy consumption assessment, pressure control, and scheme execution effectiveness. Attached Figure Description

[0015] Figure 1This is a flowchart of a multi-level optimized scheduling method for pump station operation status perception provided by an embodiment of the present invention;

[0016] Figure 2 This is a schematic diagram of a multi-level optimized scheduling system for pump station operation status perception provided in an embodiment of the present invention;

[0017] Figure 3 This is a detailed flowchart of an optimized scheduling model for pumping stations provided in an embodiment of the present invention;

[0018] Figure 4 This is a schematic diagram of the pump station optimization scheduling model provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0020] In this application, the term "at least one" means one or more, and the term "multiple" means two or more; for example, multiple devices means two or more devices. "At least two" means two or more. "At least three" means three or more.

[0021] It should also be understood that the term “comprising” (also referred to as “includes”, “including”, “comprises” and / or “comprising”) as used in this specification specifies the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0022] This invention provides a multi-level optimized scheduling method for pump station operation status perception, such as... Figure 1A flowchart of a multi-level optimized scheduling method for pump station operation status awareness is provided. This method may include the following steps: using historical operation status records of the pump station as training data, constructing a pump station optimized scheduling model that can achieve optimal pump station operation energy consumption, balanced equipment (such as pump units and auxiliary equipment such as valves) losses (such as mechanical losses, electrical losses, and wear losses), and improved water supply stability; providing algorithmic support for the pump station optimized scheduling model, which includes a system-level comprehensive optimization algorithm for optimizing pump station resource allocation, a unit-level operation optimization algorithm for reducing pump unit energy consumption, and a pipeline system algorithm for reducing pump station energy loss. The optimization algorithms work together to achieve global scheduling and management of the pumping station. The algorithms adopt an input correlation mechanism. The output of the system-level comprehensive optimization algorithm serves as part of the input to the unit-level operation optimization algorithm. The output of the system-level comprehensive optimization algorithm also serves as part of the input to the pipeline system optimization algorithm. The system-level comprehensive optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm all need to additionally obtain the operating status of each pump in the pumping station. The system senses the operating status and total flow of each pump in the pumping station and inputs them into the pumping station optimization scheduling model. The pumping station optimization scheduling model outputs the optimal pump combination and pump flow, realizing multi-level optimization scheduling of the pumping station.

[0023] The operating status of each pump in the pumping station includes at least the start / stop status, operating frequency, current and voltage, and the total outflow of water from the pumping station.

[0024] The input association mechanism refers to the upstream output to downstream input data interaction rules formed between algorithms at various levels (system level, unit level, and pipeline system level) in the pump station optimization scheduling model, based on functional division of labor and logical dependencies.

[0025] In one example embodiment, based on a year's worth of preprocessed historical operating records, machine learning methods are first used to perform data-driven modeling of the pump station's operating data, resulting in a characteristic model that reflects the true performance of each pump. Specifically, the start-stop status, operating frequency, corresponding outlet flow rate, main pipe outlet pressure, current, voltage, and active power of one or more pumps recorded at each time point are used as training samples. Nonlinear fitting is performed on the flow-head-frequency relationship and the flow-power-efficiency relationship to obtain the actual head output capacity and energy consumption level of each pump under different frequency and flow conditions, providing accurate physical constraints and energy consumption evaluation basis for subsequent scheduling optimization.

[0026] The input to the pump station optimization scheduling model is the current pump operating status and total flow rate, and the output is the optimal pump combination and pump flow rate. The constraints include the water supply flow and pressure requirements that must be met, the safe operating area of ​​each pump, etc.

[0027] The system-level integrated optimization algorithm is a key component of the pump station optimization scheduling model. It is responsible for global optimization based on the overall flow demand of the pump station and changes in the external environment. By comprehensively considering various complex factors within the pump station's operating cycle, this algorithm aims to calculate the optimal scheduling scheme with the lowest energy consumption and cost at different time periods. Its main objective is to optimize the allocation of pump station resources through flexible scheduling strategies to maximize energy efficiency and minimize operating costs. Specifically, it involves inputting the total daily water output, electricity price fluctuations, water demand fluctuations, and pump station operating constraints into the system-level integrated optimization algorithm.

[0028] Pump station operation constraints refer to the series of physical limitations that typically restrict the operation of a pump station. These constraints include the maximum operating capacity of the equipment, the reasonable range of flow rate, safe water level limits, and equipment maintenance cycles. System-level optimization algorithms must be optimized while adhering to these constraints to ensure that the pump station achieves not only the lowest energy consumption and cost during scheduling but also safe and stable operation. The inclusion of these constraints effectively prevents the pump station from operating under overload or beyond its safe operating range, avoiding equipment damage and system failures.

[0029] The objective function is to minimize the total operating cost. Where C represents total operating cost, G j Let Q be the electricity price for time period j. j For the traffic demand of time period j, T j Let j be the duration of time period j. This objective function clarifies the direction of optimization: rationally allocating traffic within each time period to reduce operating costs. j is the number of each time period, j = 1, 2, 3, ..., N, where N is the total time period number.

[0030] The constraints are flow rate constraints, water level constraints, and equipment operation limitations.

[0031] Flow constraint means that the total flow of the pumping station in all time periods is equal to the preset total demand flow; water level constraint means ensuring that the water level of the pumping station is kept within a safe range during operation to prevent overflow or drying up; equipment operation limit means that the flow rate and head of each unit must be operated within the limits of its technical parameters to avoid equipment damage caused by overload.

[0032] The system-level integrated optimization algorithm is based on dynamic programming. It first divides the daily operation time into several time periods, and uses the water pumping volume, flow distribution, and equipment operating status of each time period as decision variables. It uses the state transition equation to clarify the changing relationship and continuity of parameters such as flow and water storage between time periods. It iterative adjustments are made by combining the actual demand of each time period, historical data and real-time feedback. Under the global optimal framework, the local optimum of each time period is achieved. Finally, the optimal combination of decision variables is solved to achieve the optimal allocation of pump station resources.

[0033] The system-level integrated optimization algorithm outputs a preliminary flow allocation scheme, which includes at least the optimal water extraction volume and flow allocation.

[0034] The system-level integrated optimization algorithm not only considers multiple factors such as flow demand, electricity price fluctuations, and equipment status on a global scale, but also maximizes resource utilization efficiency and reduces operating costs through flexible scheduling strategies. Using techniques such as dynamic programming and state transition equations, the system-level integrated optimization algorithm ensures the stability and continuity of the pump station scheduling process, providing strong technical support for the efficient and sustainable operation of pump stations. The synergistic effect of this algorithm with unit-level and pipeline system optimization algorithms enables pump stations to achieve optimized scheduling under different operating conditions, thereby improving the overall efficiency of the system.

[0035] Among them, the unit-level operation optimization algorithm is an important component of the pump station optimization scheduling model. Its core objective is to ensure that the units operate within their optimal efficiency range by rationally allocating their workloads while meeting the pump station's flow demand, thereby minimizing energy consumption and improving the overall energy efficiency of the system. This algorithm mainly relies on accurate modeling of the performance characteristics of different units. It achieves the optimal solution for the coordinated operation of multiple units through optimization algorithms. Specifically, it involves inputting the unit performance characteristics, pump station operating status, expected flow, and the preliminary flow allocation scheme output by the system-level comprehensive optimization algorithm into the unit-level operation optimization algorithm.

[0036] Among these, the performance characteristics of each pump unit, specifically its performance curves, are crucial for unit-level optimization. These performance curves typically include flow-head and efficiency curves, describing the unit's operation and energy efficiency under different flow and head conditions. These curves reflect the dynamic characteristics of the pumping station equipment and accurately represent the unit's performance under specific conditions. Therefore, the primary task of unit-level optimization is to accurately acquire and comprehensively model the performance characteristics of each unit, including its power requirements and operating efficiency at different flow and head conditions. By conducting a detailed analysis of each unit's performance, scientific decisions can be made based on accurate data during the optimization process, avoiding ineffective scheduling and thereby improving the overall energy efficiency of the system.

[0037] The pump station's operational status, referring to its real-time operating condition, directly impacts the implementation of unit-level optimization algorithms. Real-time data on each unit's flow rate, head, speed, power consumption, and operating efficiency must be monitored and fed back in real time. This data not only reflects the unit's operating status at a specific moment but also provides reliable real-time data input for the optimization algorithm. During the optimization process, the operational data of each unit must be dynamically updated to ensure that optimization decisions align with actual operating conditions. By acquiring real-time unit status information, the algorithm can adjust unit scheduling and load allocation in real time, improving the flexibility and adaptability of unit operation and ensuring a smooth transition of equipment during load changes.

[0038] Based on the results of the system-level integrated optimization algorithm, the expected flow rate will determine the flow rate demand of the pumping station in each time period.

[0039] The objective function is to minimize the total energy consumption of the generating units. E represents the total energy consumption of the unit, and P represents the total energy consumption of the unit. k Let η be the input power of the k-th unit. k Let be the operating efficiency of the k-th unit, where k is the unit number (k=1, 2, 3, ...), and M is the number of units. By minimizing the energy consumption target, the algorithm can rationally select and allocate units to ensure optimal overall system energy efficiency. Flow constraints, head constraints, efficiency constraints, and operating time constraints are used as constraints.

[0040] The flow constraint here refers to the total flow of the pump station over all time periods being equal to the preset total demand flow.

[0041] Head constraint means that the head of each unit should be consistent with the working head of the system to ensure smooth fluid transport and that the pumping station can maintain normal operation; efficiency constraint means that the operating efficiency of the unit under different flow and head conditions should be kept within its rated range to avoid long-term inefficient operation of the unit; operating time constraint means that according to the daily operation strategy of the pumping station, some units may only operate during specific periods to cope with different peak and off-peak water use periods.

[0042] The genetic algorithm is used to iterate and solve the problem multiple times until the convergence criterion or the maximum number of generations is reached. The convergence criterion refers to finding the optimal fitness solution, and the maximum number of generations refers to reaching the preset number of iterations.

[0043] The unit-level operation optimization algorithm uses evolutionary algorithms such as genetic algorithms to solve the problem. First, an initial population representing the unit operation combination and flow allocation scheme is randomly generated. The fitness function is constructed with total energy consumption as the core to evaluate the quality of candidate solutions. Excellent parent individuals are selected through tournament selection, roulette wheel selection and other methods. Then, the parent genes are exchanged through crossover operation to increase the diversity of solutions. Combined with mutation operation, the individual genes are randomly fine-tuned to deepen the exploration of the solution space. Finally, the above operations are repeated through multiple generations of iteration until the optimal fitness solution is found or the preset number of iterations is reached, and finally the unit operation scheme with the best energy efficiency is obtained.

[0044] The unit-level operation optimization algorithm outputs the unit combination and allocation scheme with the lowest total energy consumption, which includes at least flow allocation, flow demand and water supply pressure of pipeline nodes.

[0045] Unit-level operation optimization algorithms are a crucial component of pump station optimization scheduling. Their core objective is to maximize overall pump station energy efficiency through rational scheduling of unit operating states. By combining precise modeling with optimization tools such as genetic algorithms, unit-level optimization algorithms can effectively improve pump station energy efficiency, reduce energy consumption, and lower operating costs. Through real-time data input, optimization decision-making, and feedback adjustments, unit-level optimization algorithms can be continuously improved and refined during actual pump station operation, ensuring long-term stable and efficient operation.

[0046] Pipeline system optimization algorithms play a crucial role in pump station optimization scheduling models. Their purpose is to minimize energy losses caused by factors such as pipeline friction, local resistance, and uneven flow during water transport by rationally allocating flow within the pipeline, thereby improving the overall operating efficiency of the pump station system. Since water flow within pipelines is affected by factors such as friction, bends, valves, and pipeline materials, leading to energy losses, optimizing the pipeline flow allocation scheme can not only significantly reduce energy consumption and save operating costs but also improve system stability, safety, and long-term operational reliability. By precisely controlling the flow state and flow allocation within the pipeline, it is possible to ensure that the flow requirements and water supply pressure of different nodes are met while minimizing the energy loss of the entire pump station system, thereby optimizing resource allocation and improving operating efficiency. Specifically, this involves inputting the pipeline geometric characteristics, fluid physical characteristics, flow requirements output from the system-level integrated optimization algorithm, and water supply pressure of the pipeline nodes into the pipeline system optimization algorithm.

[0047] Pipeline geometry refers to the characteristics of a pipeline, such as its length, diameter, material, and surface smoothness, which directly affect the flow resistance of water. The geometry of each pipeline is an important factor affecting friction loss and energy loss during water transport. These characteristics determine the flow state of the fluid in the pipeline, including flow velocity, flow mode (laminar or turbulent), and flow resistance.

[0048] Fluid physical properties refer to the physical characteristics of a fluid, such as density, viscosity, and temperature, which directly affect the flow state of water and the energy loss of the fluid within a pipe. Fluid viscosity affects flow resistance, especially at lower flow velocities, where its influence is more significant. Fluid density is closely related to the dynamic properties of water flow within a pipe. Therefore, accurately understanding the physical properties of fluids and combining this knowledge with fluid dynamics modeling can provide a more scientific basis for optimizing pipe flow distribution.

[0049] The flow demand and water supply pressure of pipeline nodes, output by the system-level integrated optimization algorithm, serve as the basic constraints for pipeline system optimization. Flow demand typically varies over time, while water supply pressure fluctuates with the pipeline's operating status and changes in the nodes.

[0050] The objective function is to minimize the total energy loss. Where LS is the total energy loss, f s Let L be the friction coefficient of the s-th pipe. s Let D be the length of the s-th pipe. s Let Q be the diameter of the s-th pipe. s Let be the flow rate of the s-th pipe, g be the acceleration due to gravity, and by minimizing energy loss, the optimization algorithm can find the optimal flow allocation scheme, thereby reducing the operating cost of the entire system. s is the number of each pipe, s=1,2,3,...,H, where H is the total number of pipes.

[0051] The constraints are flow rate constraint, pressure balance constraint, pipeline characteristic limitation, and local loss limitation.

[0052] Flow constraint means that the total flow of the pumping station in all time periods is equal to the preset total demand flow, thereby satisfying the flow requirements of the pumping station.

[0053] Pressure balance constraint means that the pressure distribution within the pipeline system must remain balanced, especially ensuring that the pressure at each pipeline node meets the water supply demand, and avoiding system instability due to insufficient or excessive pressure; pipeline characteristic limitation means that the maximum flow rate that each pipeline can withstand should be within its design specifications to prevent pipeline damage or bursting caused by overload operation; local loss limitation means that the local resistance of local components such as elbows and valves in the pipeline system must be kept within an acceptable range to avoid excessive energy loss to the system.

[0054] The pipeline system optimization algorithm employs linear programming or heuristic algorithms for solution. It utilizes heuristic algorithms such as linear programming, nonlinear programming, genetic algorithms, and particle swarm optimization. First, a pipeline flow model is established based on fluid dynamics equations such as the Bernoulli equation and the Darcy-Weisbach equation. An initial flow allocation scheme covering the solution space is randomly generated. A fitness function is constructed with total energy consumption as the core to evaluate the merits of candidate solutions. Simultaneously, a penalty function is used to handle constraints such as flow demand, pipeline capacity, and node water supply pressure. Then, the solution set is iteratively updated through heuristic algorithms such as crossover and mutation, comprehensively searching the solution space and gradually converging. Ultimately, the complex flow allocation problem is efficiently solved, minimizing system energy loss.

[0055] The pipeline system optimization algorithm outputs the optimal flow distribution for each pipeline, energy consumption analysis, and the total energy consumption of the entire system.

[0056] In one example embodiment, a schematic diagram of the pump station optimization scheduling model structure is shown below. Figure 4 As shown, this architecture visually presents the core logic link for data-driven fitting and optimization output. Based on one year of full-dimensional time-series operational data from the pumping station, the data undergoes preprocessing (including outlier identification, standardization, and missing value completion). The input layer receives two core parameters: the current pump operating status (start / stop / frequency) and the total pumping station flow rate. These parameters are combined with time-series derived features and constraint features (such as water supply pressure demand and pump safety current threshold) to form a high-dimensional input. The hidden layer has four neurons equipped with bias terms (nodes marked "1" in the diagram, used to introduce adjustable offsets to the model and improve fitting flexibility). The ReLU activation function is used to fit the nonlinear correlations in the pumping station's operational data. The output layer activates the optimal pump combination (start / stop status) through Softmax classification and outputs the single-pump flow allocation results through linear regression. This provides the foundational decision-making basis for the subsequent three-layer optimization algorithms: system-level integrated optimization, unit-level operation optimization, and pipeline system optimization. Ultimately, under constraints such as water supply flow / pressure demand and pump safety operating range, the architecture achieves the scheduling goal of maximizing pumping station energy efficiency and minimizing energy consumption and operating costs.

[0057] Furthermore, the multi-level optimization scheduling of pump stations also includes continuous monitoring and diagnosis of the multi-level data sharing links during the application of the pump station optimization scheduling model. When it is determined that there is an input inconsistency anomaly caused by a pseudo deadlock, self-consistency deviation suppression measures are implemented to optimize the multi-level optimization scheduling process of the pump station optimization scheduling model.

[0058] The input inconsistency anomaly caused by pseudo-deadlock refers to an abnormal state in the multi-level data sharing links of a pump station optimization scheduling model, where data transmission blockage, queue delays, and other pseudo-deadlock phenomena lead to inconsistencies, missing data, or timing mismatches in the input data between the system-level integrated optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm. This type of anomaly is not a complete standstill caused by a true deadlock, but rather a hidden blockage in the data interaction links that prevents the consistency and validity of the basic operational data (such as pump status, flow parameters, and output results of previous algorithms) required by each level of algorithm. This results in conflicting algorithm input conditions, chaotic optimization boundaries, and ultimately affects the accuracy and feasibility of the model's scheduling decisions. Targeted self-consistency deviation suppression measures are needed to resolve this issue.

[0059] Multi-level data sharing links refer to the cross-level data transmission and interaction channels in the pump station optimization scheduling model, including system-level comprehensive optimization algorithms, unit-level operation optimization algorithms, and pipeline system optimization algorithms.

[0060] The following embodiment uses the cross-level data transmission and interaction channel between the system-level integrated optimization algorithm and the unit-level operation optimization algorithm as an example for analysis.

[0061] Continuous monitoring and diagnosis of multi-level data sharing links are carried out. The specific monitoring and diagnosis process is as follows: identify the key monitoring dimensions of multi-level data sharing links, including at least the on-time output rate of optimization solutions, the maximum queue retention time, and the maximum waiting time for upstream data.

[0062] Extract the maximum queue dwell time of the multi-level data sharing link from key monitoring dimensions and compare it with the preset maximum allowed queue dwell time in the database. The maximum queue dwell time refers to the maximum time that data, tasks or requests wait to be processed in the queue.

[0063] If the maximum queue dwell time of the multi-level data sharing link is less than the maximum allowed queue dwell time, the multi-level data sharing link is diagnosed as normal. The key monitoring dimensions of the multi-level data sharing link are continuously monitored to determine whether there is an input inconsistency anomaly caused by a pseudo deadlock.

[0064] If the maximum queue dwell time of a multi-level data sharing link is greater than or equal to the maximum allowed queue dwell time, then congestion is diagnosed in the multi-level data sharing link. In this case, a rate limiting strategy is implemented for low-priority data in the multi-level data sharing link, that is, the transmission rate of low-priority data is reduced.

[0065] Low-priority data refers to data that is pre-defined based on the operational requirements of the pump station optimization scheduling model, has little impact on real-time scheduling decisions, and can be transmitted with delay or reduced speed. This includes, but is not limited to, non-real-time statistical data: such as historical energy consumption summary data, equipment runtime statistical reports, and past scheduling effect review data, which are only used for offline analysis or archiving and do not affect the real-time calculation of the optimization algorithm in the current period; duplicate and redundant data: such as basic parameters of water pumps (rated power, diameter, etc.) that have been transmitted multiple times and have not changed, and static attribute data of pipelines, etc. Repeated transmission does not provide additional benefits to the scheduling model; low-urgency alarm data: such as non-critical alarm information such as minor abnormal noises from equipment and minor fluctuations in parameters, which do not require immediate transmission and processing, and the reduced speed will not affect the safe operation of the system.

[0066] Reducing the transmission rate of low-priority data means retrieving the magnitude of a single transmission rate decrease from the database, subtracting this magnitude from the current transmission rate of low-priority data, thereby reducing the transmission rate.

[0067] The single transmission rate reduction refers to the fixed value of the transmission rate that needs to be reduced in a single rate limiting operation for low-priority data. It is the core parameter for quantifying the execution of rate limiting strategies. Its value needs to be pre-calibrated in combination with the bandwidth carrying capacity of multi-level data sharing links, the latency tolerance of low-priority data, and the data transmission requirements of high-priority data.

[0068] Reducing the transmission rate of low-priority data can decrease queue congestion caused by low-priority data occupying transmission resources, reduce input inconsistency anomalies caused by excessive queue congestion time, avoid interference of pseudo-deadlock on the operation of the optimization scheduling model from the source, reserve sufficient bandwidth for cross-level data interaction of core algorithms such as system-level and unit-level, solve the problems of algorithm input delay and data loss caused by link congestion, and ensure the stability and accuracy of the optimization solution process.

[0069] Specifically, to determine whether there is an input inconsistency anomaly caused by a pseudo-deadlock, the specific determination process is as follows: set preliminary conditions for anomaly identification.

[0070] Extract the on-time output rate of optimization solutions and the maximum waiting time for upstream level data from key monitoring dimensions; the on-time output rate of optimization solutions refers to the number of optimization solutions successfully produced within a specified time ÷ the total number of optimization solutions that should be produced in the same period × 100%. Its core function is to reflect the execution efficiency of the optimization mechanism. If this indicator is low, it indicates that the output of optimization solutions is lagging behind, which may be due to upstream data blockage, algorithm operation delay, etc., and directly affects the real-time performance of scheduling decisions.

[0071] The maximum waiting time for upstream data refers to the maximum waiting time consumed by a certain level algorithm (such as a pipeline system optimization algorithm) during its operation to obtain output data from an upstream level algorithm (system-level integrated optimization algorithm) or real-time status data of a pumping station. It directly reflects the transmission efficiency of the multi-level data sharing link. If the cumulative time is too long, it indicates that data interaction is blocked or delayed, which can easily lead to problems such as input data timing mismatch and excessive queue congestion, thereby inducing pseudo-deadlock and input inconsistency anomalies.

[0072] The system-level integrated optimization algorithm is the upstream level of the unit-level operation optimization algorithm and the pipeline system optimization algorithm, while the unit-level operation optimization algorithm and the pipeline system optimization algorithm are both downstream levels of the system-level integrated optimization algorithm.

[0073] The initial conditions for anomaly identification are as follows: the on-time output rate of the optimization scheme in the current optimization window is lower than the preset on-time output rate of the optimization scheme in the database, and the maximum waiting time for upstream data at a certain level exceeds the preset waiting time in the database. The preset on-time output rate of the optimization scheme refers to the minimum allowed value of the on-time output rate of the optimization scheme. The preset waiting time is used to determine whether the maximum waiting time for upstream data is abnormal. If the maximum waiting time for upstream data exceeds the preset waiting time in the database, an anomaly exists.

[0074] If the initial conditions for anomaly identification are not met, continuous monitoring and diagnosis of the multi-level data sharing link will be carried out; if the initial conditions for anomaly identification are met, the transmission rate of low-priority data will be reduced again, and anomaly auxiliary verification will be triggered.

[0075] After meeting the initial conditions for anomaly identification, the data transmission rate of low-priority data is reduced again before triggering anomaly auxiliary verification. The core principle is to precisely alleviate potential congestion in multi-level data sharing links through a pre-emptive bandwidth release action, without interrupting the model scheduling process, thus creating a stable data interaction environment for anomaly verification. This operation can quickly release link bandwidth resources, reduce the occupation of core transmission channels by low-priority data, resolve upstream data transmission delays caused by bandwidth constraints, and prevent further accumulation of queue delays. Furthermore, by observing the changes in link status after the speed reduction, it can help distinguish whether the root cause of the anomaly is simple link congestion or input inconsistency caused by pseudo-deadlock, avoiding direct anomaly judgments and excessive measures that could disturb the scheduling model. Simultaneously, it provides more accurate link operation data support for subsequent anomaly verification, ensuring the accuracy of anomaly diagnosis and reducing the risk of decreased optimization scheduling efficiency or decision-making bias due to misjudgments.

[0076] Anomaly verification refers to situations where the length of the data revision number queue for each level to be aligned within the current optimization window is greater than 0, or the residual of cross-level optimization iteration has not entered a downward trend and remains in an unsolved initial state.

[0077] The current optimization window has a data revision version number queue length greater than 0 at each level. Assuming the current optimization window for the pump station optimization scheduling model is the 10:00-11:00 time period for unit operation optimization, the unit-level operation optimization algorithm can only start optimization at this level after receiving the resource allocation revision data from the system-level integrated optimization algorithm. Due to real-time data updates, the system-level integrated optimization algorithm revises the total water supply of the pump station for the 10:00 time period to 8000 m³ (revision version number V2) and sends it to the alignment queue of the unit-level operation optimization algorithm. The water supply data of 7500m³ was used to replace the original version V1. If the queue status is checked at 10:05, it is found that the revised data of version V2 has not been received and aligned by the unit-level operation optimization algorithm (queue length = 1), which satisfies the condition that the queue length is greater than 0. This situation indicates that the data revision synchronization between the two levels is blocked. The unit-level algorithm may still be based on the old version data (V1) for calculation, which may easily lead to inconsistency between the input data and the upstream, thus causing deviation in the optimization results. It also provides a basis for subsequent verification of the input inconsistency caused by pseudo deadlock.

[0078] Cross-level optimization iteration residual refers to the deviation between the algorithm's calculation result at each level and the ideal target. If the ideal target of the algorithm at this level is ≤1000kWh of total energy consumption of the pump station unit in a certain period, and the total energy consumption calculated in the current iteration is 1200kWh, then the cross-level optimization iteration residual at this time = 1200kWh - 1000kWh = 200kWh. If the total energy consumption drops to 1100kWh after the next iteration, the residual is updated to 100kWh, which directly reflects the gap between the algorithm result and the ideal target. The downward trend refers to the state in which the cross-level optimization iteration residual gradually decreases and continuously approaches the ideal target as the number of algorithm iterations increases. Continuing with the example above: the residual is 200kWh in the first iteration, 100kWh in the second, 50kWh in the third, and 20kWh in the fourth. The residual decreases continuously with the number of iterations, showing a clear downward trend, indicating that the algorithm's input data is valid and the computational logic is normal. Conversely, if the residual remains around 200kWh during the iteration process (e.g., from 200kWh to 210kWh to 195kWh to 205kWh), without a clear decreasing pattern, or remains unchanged at the initial 200kWh, it is determined that the algorithm has not entered a downward trend, suggesting that the algorithm may not be able to converge to the ideal target due to inconsistencies in the input data (such as inconsistencies caused by pseudo-deadlock).

[0079] If the anomaly auxiliary verification is satisfied, it is determined that there is an input inconsistency anomaly caused by a pseudo-deadlock, and the current optimization window is marked as an abnormal window and the abnormal location is located; if the anomaly auxiliary verification is not satisfied, it is determined that there is no input inconsistency anomaly caused by a pseudo-deadlock.

[0080] Locating the anomaly location refers to accurately pinpointing the specific point in time where a pseudo-deadlock causes an input inconsistency anomaly. The core focus is on the data transmission nodes and cross-level interaction interfaces of the multi-level data sharing link. Specifically, this includes the interaction interface between the system-level integrated optimization algorithm and the unit-level operation optimization algorithm, the interaction interface between the unit-level operation optimization algorithm and the pipeline system optimization algorithm, as well as key components such as the data flow queues and data alignment modules corresponding to each interface.

[0081] Anomalies are precisely located through a collaborative approach of indicator tracing and link backtracking. Using anomaly-assisted verification trigger indicators as the core clues, the data flow path and algorithm operation status are checked in reverse. For example, during an optimization window (15:00-16:00), anomaly verification is triggered because the queue length of the data revision version number to be aligned in the unit-level algorithm is 3. During location, the source of the accumulated data in the queue is first traced back to resource allocation revision data output by the system-level algorithm. Then, the transmission link status is checked through link monitoring logs. It is found that the response time of the interaction interface between the system-level integrated optimization algorithm and the unit-level operation optimization algorithm reaches 45 seconds (far exceeding the normal threshold of 15 seconds), and the data verification date of this interface... The log showed a persistent error message indicating a version number format mismatch. The anomaly was ultimately located at the cross-level interaction interface between the system-level and unit-level algorithms and the associated data verification module. For example, if the pipeline system optimization algorithm's iteration residual remained at 300kWh (ideal threshold ≤50kWh) for an extended period without a downward trend triggering verification, tracing back its input data link revealed that the unit-level operation optimization algorithm had pushed the latest traffic allocation revision data, but the pipeline system optimization algorithm's version number identification unit had not triggered the data update instruction. This led to the anomaly being located in the pipeline system optimization algorithm's data receiving and alignment module, where a version number identification malfunction prevented upstream data from being properly connected.

[0082] Specifically, self-consistency deviation suppression measures include proactive self-consistency deviation control strategies and data interaction architecture adaptation and optimization strategies; after the self-consistency deviation suppression measures are implemented, the effectiveness of the self-consistency deviation suppression measures is verified.

[0083] The self-consistency deviation proactive control strategy specifically refers to: for the current abnormal window, using the historical reliable snapshot of the previous window to calculate the scheduling scheme at the abnormal position and prioritizing its execution; injecting a waiting cost item into the objective function corresponding to the abnormal position to forcibly prohibit continuous waiting for upstream data; and labeling the scheduling scheme generated in this rollback, specifying the rollback level and snapshot version validity period, so that the downstream level corresponding to the abnormal position can receive it and automatically adapt to the label with reduced weight.

[0084] Suppose that the 16:00-17:00 period in a certain pump station's optimized scheduling model is an abnormal window. After location confirmation, the abnormality is found at the interaction interface between the unit-level operation optimization algorithm and the pipeline system optimization algorithm (due to a fault in the interface data alignment module, the pipeline system optimization algorithm cannot receive the latest flow allocation revision data output by the unit-level operation optimization algorithm, causing a pseudo-deadlock and resulting in input inconsistency). The corresponding control strategy execution process is as follows:

[0085] The system generates a scheduling scheme based on historical reliable snapshots: It automatically retrieves a reliable operating snapshot of the pipeline system optimization algorithm from 15:00 to 16:00 (the previous normal window) (including valid data such as the flow allocation scheme, pipeline pressure threshold, and energy consumption optimization parameters that have been verified during this period). Based on this snapshot, it quickly calculates the pipeline system scheduling scheme for the period from 16:00 to 17:00 and prioritizes its distribution to the field for execution, avoiding scheduling interruptions or decision gaps due to anomalies.

[0086] Injecting a waiting cost term into the objective function: A waiting cost term is added to the optimization objective function of the pipeline system-level algorithm, as follows:

[0087] ;

[0088] Where pen(w) is the waiting cost item, w is the maximum waiting time for upstream level data, TM1 defines the waiting time, TM2 is the warning waiting time stored in the database, α is the first weight coefficient stored in the database, used to quantify the linear growth intensity of the waiting cost item w in the TM1 to TM2 interval, β is the second weight coefficient stored in the database, used to quantify the linear growth intensity of the waiting cost item w greater than TM2, α is less than β, and TM1 is less than TM2.

[0089] Warning waiting time refers to the maximum allowed waiting time for upstream level data.

[0090] It should also be explained that the units of α and β are taken as the penalty per unit of time, so the units cancel each other out when multiplied by w, and pen(w) is a dimensionless pure number.

[0091] Introducing a waiting cost term can suppress computational biases caused by inconsistent data arrival, buffer delays, and timing misalignments at the source, significantly improving the accuracy of indicators and decisions: First, by weighting the waiting time exceeding the tolerance and warning thresholds, the optimizer will proactively avoid using outdated / asynchronous data in the estimation of key indicators such as material balance and recovery rate and removal rate, reducing false trends and error accumulation caused by asynchronous splicing; Second, the two-slope design ensures smooth but sensitive performance when approaching the threshold, and rapidly amplifies the penalty after exceeding the limit, prompting task scheduling to maintain the time alignment and freshness of each data source, thereby reducing the variance and systematic bias of statistics; Third, the dimensionless design of the penalty facilitates synthesis with other objective terms at the same scale, avoiding weight imbalance caused by inconsistent units, and improving the stability and interpretability of the solution; Fourth, the single-merge and continuous piecewise configuration ensures stable numerical gradients and clear search direction, reducing local oscillations and false convergence. Overall, this waiting cost item explicitly incorporates the time series health status into the objective function, ensuring that estimation and control are based on time series consistency, unified standards, and fresh data, thereby improving the reliability and auditability of the monitoring and parameter adjustment results throughout the process.

[0092] Dispatch scheme labeling and downstream adaptation: The dispatch scheme generated based on snapshot rollback is labeled with the following rollback levels: Rollback level: Level 1 (temporary rollback due to interface anomaly), Snapshot version: V2025111315 (window snapshot at 15:00 on November 13, 2025), and validity period: valid only from 16:00 to 17:00. After receiving the scheme, the downstream pipeline operation monitoring module automatically adapts it by reducing its weight according to the label—it will no longer be used as a long-term optimization benchmark, but only for emergency dispatching in the current period. At the same time, it retains the key monitoring authority for the scheme's execution effect, and automatically switches to the normal optimization process after the anomaly is repaired.

[0093] Furthermore, the data interaction architecture adaptation and optimization strategy specifically refers to: reconstructing the cross-layer data interaction mechanism, adopting a publish-subscribe model to transmit snapshot data, with the upstream layer actively publishing the current window snapshot after generating it through copy-on-write; setting a maximum continuous rollback window threshold, and if a rollback anomaly is identified, terminating the cross-layer iteration and executing a preset conservative fixed running combination, such as the current running combination; relying on edge computing nodes to perform feature extraction on the original high-frequency data stream, retaining only the feature summary necessary for decision-making; and storing the original high-frequency data stream in local storage.

[0094] An abnormal rollback condition refers to a situation where the maximum number of consecutive rollback windows exceeds the maximum consecutive rollback window threshold.

[0095] In one example implementation, the system-level integrated optimization algorithm, acting as the upstream level, generates a snapshot of the current window every 15 minutes (one optimization window) using copy-on-write technology. This snapshot includes total water supply indicators and global energy consumption control thresholds. It is then proactively published to a dedicated data topic using a publish-subscribe model. The unit-level operation optimization algorithm and the pipeline system optimization algorithm, having subscribed in advance, can immediately access this snapshot, using it as supplementary data for their core inputs. Simultaneously, a maximum consecutive rollback window threshold of two is set. If the maximum consecutive rollback window for the unit-level operation optimization algorithm is three (meaning three consecutive windows result in scheduling plan rollback due to data mismatches), exceeding the maximum threshold... If the system continuously rolls back to the window threshold, it will immediately stop the cross-layer iteration process and automatically execute a preset conservative fixed operation combination of running the two main water pumps at 55% of their rated frequency to maintain the basic pressure of the pipeline network. In addition, the edge computing nodes deployed next to each water pump will perform feature extraction on the raw high-frequency data streams of 400 data points per second, such as vibration, outlet pressure, and real-time current. Only the feature summaries necessary for decision-making, such as whether the vibration peak exceeds the standard, the average current, and the pressure fluctuation range, will be transmitted to the algorithms at each level, while the complete raw high-frequency data stream will be stored in the local storage unit of the water pump. This ensures both the efficiency of algorithm input and provides data support for equipment operation traceability.

[0096] The publish-subscribe mechanism uses window snapshots generated by copy-on-write as the sole transmission unit, ensuring the atomicity and immutability of data within a single window and eliminating dirty reads / phantom reads and caliber drift caused by concurrent upstream writes and downstream reads. Abnormal rollbacks are truncated according to the maximum continuous rollback window threshold to avoid amplifying errors by iterating on expired data, and a conservative fixed-run combination is used as a fallback to suppress oscillations and misjudgments of indicators and control quantities. At the edge, high-frequency streams are locally feature extracted and only necessary summaries are sent up, reducing link noise and latency while ensuring consistency in timing and feature caliber between upstream and downstream for the same window, reducing estimation bias caused by packet loss, out-of-order delivery, and oversampling from the source. The original high-frequency stream is retained locally to provide a benchmark for offline recalculation, model backtracking, and auditing, forming a closed loop of lightweight online decision-making and rigorous offline verification. As a result, cross-layer data interaction has shifted from streaming splicing to a snapshot-driven, threshold-controlled, and traceable governance paradigm, significantly reducing systemic errors caused by temporal misalignment and data obsolescence, and ensuring that recovery / removal rates, alarms, and linkage decisions are based on temporal alignment, unified standards, and reproducible evidence.

[0097] Furthermore, verifying the effectiveness of self-consistency deviation suppression measures specifically refers to: determining whether the indicator recovery conditions are met within the verification window; the indicator recovery conditions refer to the following: the on-time output rate of the optimized scheme is not lower than the defined on-time output rate of the optimized scheme, the maximum waiting time for upstream level data does not exceed the preset defined waiting time, and the rollback trigger count is cleared. Among these, the verification window and the abnormal window have the same duration.

[0098] If the indicator recovery conditions are met within the verification window, the verification window will be marked as a valid verification window; otherwise, continuous monitoring and diagnosis will be implemented for the multi-level data sharing link.

[0099] The number of consecutive occurrences of valid windows is counted. If the number of consecutive occurrences of valid windows is less than the number of first windows, continuous monitoring and diagnosis are implemented for the multi-level data sharing link. If the number of consecutive occurrences of valid windows is greater than or equal to the number of first windows and less than the number of second windows, the value corresponding to the waiting cost item is reduced. If the number of consecutive occurrences of valid windows is greater than or equal to the number of second windows and less than the number of third windows, the waiting cost item is removed. If the number of consecutive occurrences of valid windows is greater than or equal to the number of third windows, the locally stored data is input into the pump station optimization scheduling model in a tiered manner.

[0100] Stepped input refers to importing locally stored data into the model in batches and according to priority, rather than inputting all data at once. Specifically, the first step is to select the core data in local storage that best matches the current scheduling scenario (such as recent equipment health characteristics and operational data under similar water supply) as the first-tier priority input, and after verifying the model's stable operation, import the next core data (such as historical energy consumption optimization cases and pipeline resistance change data) as the second tier; finally, import non-critical supplementary data (such as long-term environmental temperature and humidity trends) as the third tier as needed. After each tier of input, a short time window is reserved to observe the model's output effect, ensuring that the data import does not cause algorithm fluctuations, while making full use of local data to improve the model's optimization accuracy.

[0101] The number of first windows is less than the number of second windows, and the number of second windows is less than the number of third windows. All of them are stored in the database and are used as the threshold for the number of consecutive occurrences of effective windows for gradient execution strategies.

[0102] Reducing the value corresponding to the waiting cost item means retrieving a preset single data reduction amount from the database, thereby subtracting the single data reduction amount from the current value corresponding to the waiting cost item, thus completing the update of the value corresponding to the waiting cost item.

[0103] Embedding the mechanism from verification window to recovery conditions to tiered handling into the link governance can transform data freshness, time-series health, and production stability into measurable and decision-making thresholds, thereby significantly improving the accuracy of computation and scheduling. Specifically: simultaneously constraining on-time output rate, maximum waiting time, and resetting backtracking count within the verification window can suppress systematic biases caused by data staleness, cross-layer misalignment, and backtracking diffusion, ensuring that the data entering the estimation and optimization stages has consistent caliber and is up-to-date; setting a triple threshold for the consecutive number of valid verification windows is equivalent to adding a statistically significant confidence threshold / hysteresis to the recovery judgment, which reduces false recovery caused by a one-off accidental improvement (false relaxation of penalties, false parameter tuning), and adaptively weakens or even removes the waiting penalty term after continuous stabilization, avoiding long-term bias of the objective function due to penalty residue; when continuous stability reaches the highest threshold, the input intensity of local data to the pump station optimization scheduling model is restored in a stepwise manner to prevent solution oscillations and parameter backtracking caused by sudden increases in volume. This forms a closed loop from short-term monitoring to medium-term steady-state confirmation to long-term penalty unloading to smooth re-launch, ensuring that the estimation of key quantities such as recovery rate / removal rate and flow load is based on a continuous and verifiable healthy window, and allowing the optimizer to converge under real operating conditions, significantly reducing the risks of false alarms, misadjustments and overfitting.

[0104] A second aspect of the present invention provides a multi-level optimized scheduling system for pump station operation status perception, such as... Figure 2 A schematic diagram of a multi-level optimized scheduling system for pump station operation status perception is shown, including a model building module, an algorithm composition module, an optimized scheduling module, and a database.

[0105] The model building module is connected to the algorithm component module, the algorithm component module is connected to the optimization and scheduling module, and the model building module, algorithm component module, and optimization and scheduling module are all connected to the database.

[0106] The model building module is used to build an optimized scheduling model for pump stations, using historical operating status records as training data, which can achieve optimal energy consumption, balanced equipment losses, and improved water supply stability.

[0107] The algorithm module provides algorithmic support for the pump station optimization scheduling model. The pump station optimization scheduling model includes a system-level comprehensive optimization algorithm for optimizing pump station resource allocation, a unit-level operation optimization algorithm for reducing pump station unit energy consumption, and a pipeline system optimization algorithm for reducing pump station energy loss. These algorithms work together for the global scheduling and management of the pump station. An input correlation mechanism is used between the algorithms. The output of the system-level comprehensive optimization algorithm serves as part of the input to the unit-level operation optimization algorithm, and the output of the system-level comprehensive optimization algorithm also serves as part of the input to the pipeline system optimization algorithm. The system-level comprehensive optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm all require additional information on the operating status of each pump in the pump station.

[0108] The optimization scheduling module is used to sense the operating status and total flow of each pump in the pumping station and input them into the pumping station optimization scheduling model. The pumping station optimization scheduling model outputs the optimal pump combination and pump flow, realizing multi-level optimization scheduling of the pumping station.

[0109] The database is used to store parameters involved in a multi-level optimized scheduling system for pump station operation status awareness. Its formulation process is usually based on system functional requirements and scheduling accuracy requirements, and is structured according to the complete link from operation awareness to model building to optimization solution to result execution to process traceability.

[0110] Figure 3 This invention provides a detailed flowchart of a pump station optimization scheduling model. Using historical operating status records of the pump station as training data, a pump station optimization scheduling model is constructed to achieve optimal energy consumption, balanced equipment wear, and improved water supply stability. The model senses the operating status and total flow of each pump within the pump station and inputs this information into the model. The model outputs the optimal pump combination and pump flow rate, achieving multi-level optimization scheduling of the pump station. During the application of the pump station optimization scheduling model, continuous monitoring and diagnosis are implemented on the multi-level data sharing links. When a pseudo-deadlock-induced input inconsistency anomaly is detected, self-consistency deviation suppression measures are executed to optimize the multi-level optimization scheduling process of the pump station optimization scheduling model. If no pseudo-deadlock-induced input inconsistency anomaly is detected, continuous monitoring and diagnosis of the multi-level data sharing links are continued, and self-consistency deviation suppression measures are implemented. After the implementation of the deviation suppression measures, the effectiveness of the self-consistency deviation suppression measures is verified. Within the verification window, it is determined whether the index recovery conditions are met. If the index recovery conditions are met within the verification window, the verification window is marked as a valid verification window; otherwise, continuous monitoring and diagnosis are implemented on the multi-level data sharing link. The number of consecutive occurrences of valid verification windows is counted. If the number of consecutive occurrences of valid windows is less than the number of first windows, continuous monitoring and diagnosis are implemented on the multi-level data sharing link. If the number of consecutive occurrences of valid windows is greater than or equal to the number of first windows and less than the number of second windows, the value corresponding to the waiting cost item is reduced. If the number of consecutive occurrences of valid windows is greater than or equal to the number of second windows and less than the number of third windows, the waiting cost item is removed. If the number of consecutive occurrences of valid windows is greater than or equal to the number of third windows, the locally stored data is input into the pump station optimization scheduling model in a tiered manner.

[0111] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0112] The above description is only an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A multi-level optimized scheduling method for pump station operation status perception, characterized in that, The method includes: Using historical operating status records of pumping stations as training data, an optimized scheduling model for pumping stations is constructed, which can achieve optimal energy consumption, balanced equipment wear and tear, and improved water supply stability. This provides algorithmic support for the pump station optimization scheduling model, which includes a system-level comprehensive optimization algorithm for optimizing pump station resource allocation, a unit-level operation optimization algorithm for reducing pump station unit energy consumption, and a pipeline system optimization algorithm for reducing pump station energy loss. These algorithms work together to achieve global scheduling and management of the pump station. An input correlation mechanism is used between the algorithms. The output of the system-level comprehensive optimization algorithm serves as part of the input to the unit-level operation optimization algorithm, and the output of the system-level comprehensive optimization algorithm also serves as part of the input to the pipeline system optimization algorithm. The system-level comprehensive optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm all require additional information on the operating status of each pump in the pump station. The system senses the operating status and total flow of each pump in the pumping station and inputs this information into the pumping station optimization scheduling model. The model then outputs the optimal pump combination and pump flow rate, enabling multi-level optimization scheduling of the pumping station.

2. The multi-level optimized scheduling method for pump station operation status perception as described in claim 1, characterized in that, The system-level comprehensive optimization algorithm specifically refers to: The total daily water lifting volume, electricity price changes during different periods, water demand fluctuations, and pump station operation constraints are input into the system-level comprehensive optimization algorithm. The objective function is to minimize the total operating cost, and the constraints are flow constraints, water level constraints, and equipment operating limitations. Using a dynamic programming approach, the state transition equation is used to characterize the evolution of water storage and water level over time, and the solution is iteratively obtained around the decision variables. The system-level integrated optimization algorithm outputs a preliminary traffic allocation scheme.

3. The multi-level optimized scheduling method for pump station operation status perception as described in claim 1, characterized in that, The unit-level operation optimization algorithm specifically refers to: The unit performance characteristics, pump station operating status, expected flow, and preliminary flow allocation scheme output by the system-level integrated optimization algorithm are input into the unit-level operation optimization algorithm; The objective function is to minimize the total energy consumption of the unit, and the constraints are flow rate constraints, head constraints, efficiency constraints, and operating time constraints. The genetic algorithm is used to iterate and solve the problem multiple times until the convergence criterion or the maximum number of generations is reached. The unit-level operation optimization algorithm outputs the unit combination and allocation scheme with the lowest total energy consumption.

4. The multi-level optimized scheduling method for pump station operation status perception as described in claim 1, characterized in that, The pipeline system optimization algorithm specifically refers to: The pipeline geometry, fluid physics, flow requirements output by the system-level integrated optimization algorithm, and water supply pressure of the pipeline nodes are input into the pipeline system optimization algorithm. The objective function is to minimize total energy loss, and the constraints are flow constraints, pressure balance constraints, pipeline characteristic limitations, and local loss limitations. Solve using linear programming or heuristic algorithms; The pipeline system optimization algorithm outputs the optimal flow distribution for each pipeline, energy consumption analysis, and the total energy consumption of the entire system.

5. The multi-level optimized scheduling method for pump station operation status perception as described in claim 1, characterized in that, The multi-level optimization scheduling of pump stations also includes continuous monitoring and diagnosis of the multi-level data sharing links during the application of the pump station optimization scheduling model. When it is determined that there is an input inconsistency anomaly caused by a false deadlock, self-consistency deviation suppression measures are implemented to optimize the multi-level optimization scheduling process of the pump station optimization scheduling model. The continuous monitoring and diagnosis of the multi-level data sharing links is specifically carried out as follows: Identify the key monitoring dimensions for multi-level data sharing links; Extract the maximum queue dwell time of the multi-level data sharing link from key monitoring dimensions and compare it with the preset maximum allowed queue dwell time; If the maximum queue dwell time of the multi-level data sharing link is less than the maximum allowed queue dwell time, the multi-level data sharing link is diagnosed as normal. The key monitoring dimensions of the multi-level data sharing link are continuously monitored to determine whether there is an input inconsistency anomaly caused by a pseudo deadlock. If the maximum queue dwell time of a multi-level data sharing link is greater than or equal to the maximum allowed queue dwell time, then congestion is diagnosed in the multi-level data sharing link. In this case, a rate limiting strategy is implemented for low-priority data in the multi-level data sharing link, that is, the transmission rate of low-priority data is reduced.

6. The multi-level optimized scheduling method for pump station operation status perception as described in claim 5, characterized in that, The specific process for determining whether a false deadlock-induced input inconsistency anomaly exists is as follows: Set preliminary conditions for anomaly identification; Extract the on-time output rate of optimization solutions and the maximum waiting time for upstream data from key monitoring dimensions; The preliminary conditions for anomaly identification refer to the following: the on-time output rate of the optimization scheme in the current optimization window is lower than the on-time output rate of the defined optimization scheme, and the maximum waiting time for upstream data at a certain level exceeds the preset defined waiting time. If the initial conditions for anomaly identification are not met, continuous monitoring and diagnosis will be carried out on the multi-level data sharing links. If the initial conditions for anomaly identification are met, the transmission rate of low-priority data will be reduced again, and anomaly auxiliary verification will be triggered. The aforementioned abnormal auxiliary verification refers to the fact that the length of the data revision version number queue to be aligned at each level within the current optimization window is greater than 0, or the cross-level optimization iteration residual has not entered a downward trend and remains in an unsolved initial state. If the abnormal auxiliary verification is satisfied, it is determined that there is an input inconsistency anomaly caused by a pseudo deadlock. The current optimization window is then marked as an abnormal window, and the abnormal location is located. If the abnormal auxiliary verification is not met, it is determined that there is no input inconsistency anomaly caused by a pseudo deadlock, and continuous monitoring and diagnosis of the multi-level data sharing link will continue.

7. The multi-level optimized scheduling method for pump station operation status perception as described in claim 5, characterized in that, The self-consistency deviation suppression measures include a proactive self-consistency deviation control strategy and a data interaction architecture adaptation and optimization strategy. After the self-consistency deviation suppression measures are implemented, verify the effectiveness of the self-consistency deviation suppression measures. The self-consistency deviation proactive control strategy specifically refers to: For the current abnormal window, the historical reliable snapshot calculation scheduling scheme of the previous window is adopted at the abnormal location and executed with priority; Inject a waiting cost term into the objective function corresponding to the abnormal location to forcibly prohibit continuous waiting for upstream data; The scheduling scheme generated in this rollback is labeled to specify the rollback level and snapshot version validity period, so that the downstream level corresponding to the abnormal position can receive it and automatically adapt to the reduced weight according to the label.

8. The multi-level optimized scheduling method for pump station operation status perception as described in claim 7, characterized in that, The data interaction architecture adaptation and optimization strategy specifically refers to: The cross-layer data interaction mechanism was restructured, and a publish-subscribe model was adopted to transmit snapshot data. The upstream layer actively publishes the current window snapshot after generating it through copy-on-write. Set a maximum continuous rollback window threshold. If a rollback anomaly is detected, stop cross-layer iteration and execute a preset conservative fixed operation combination. The rollback anomaly refers to a maximum number of consecutive rollback windows exceeding the maximum consecutive rollback window threshold. By leveraging edge computing nodes, feature extraction is performed on the original high-frequency data stream, retaining only the feature summary necessary for decision-making; The original high-frequency data stream is stored locally.

9. The multi-level optimized scheduling method for pump station operation status perception as described in claim 7, characterized in that, The verification of the effectiveness of the self-consistency deviation suppression measures specifically refers to: The verification window determines whether the indicator recovery conditions are met. The recovery conditions for the indicators are: the on-time output rate of the optimization scheme is not lower than the defined on-time output rate of the optimization scheme, the maximum waiting time for upstream level data does not exceed the preset defined waiting time, and the rollback trigger count is cleared. If the indicator recovery conditions are met within the verification window, the verification window will be marked as a valid verification window; otherwise, continuous monitoring and diagnosis will be implemented for the multi-level data sharing link. The number of consecutive occurrences of valid verification windows is counted. If the number of consecutive occurrences of valid windows is less than the number of first windows, continuous monitoring and diagnosis are implemented for the multi-level data sharing link. If the number of consecutive occurrences of a valid window is greater than or equal to the number of the first window, but less than the number of the second window, reduce the value corresponding to the waiting cost item. If the number of consecutive occurrences of a valid window is greater than or equal to the number of the second window, but less than the number of the third window, the waiting cost item is removed. If the number of consecutive occurrences of a valid window is greater than or equal to the number of third windows, the locally stored data will be input into the pump station optimization scheduling model in a tiered manner.

10. A multi-level optimized scheduling system for pump station operation status perception, employing the multi-level optimized scheduling method for pump station operation status perception as described in any one of claims 1-9, characterized in that: include: The model building module is used to build an optimized scheduling model for pump stations, which can achieve optimal energy consumption, balanced equipment wear and tear, and improved water supply stability, using historical operating status records of pump stations as training data. The algorithm module provides algorithmic support for the pump station optimization scheduling model. The pump station optimization scheduling model includes a system-level comprehensive optimization algorithm for optimizing pump station resource allocation, a unit-level operation optimization algorithm for reducing pump station unit energy consumption, and a pipeline system optimization algorithm for reducing pump station energy loss. These algorithms work together for the global scheduling and management of the pump station. An input association mechanism is used between the algorithms. The output of the system-level comprehensive optimization algorithm serves as part of the input of the unit-level operation optimization algorithm. The output of the system-level comprehensive optimization algorithm also serves as part of the input of the pipeline system optimization algorithm. The system-level comprehensive optimization algorithm, the unit-level operation optimization algorithm, and the pipeline system optimization algorithm all require additional information on the operating status of each pump in the pump station. The optimization scheduling module is used to sense the operating status and total flow of each pump in the pumping station and input them into the pumping station optimization scheduling model. The pumping station optimization scheduling model outputs the optimal pump combination and pump flow, realizing multi-level optimization scheduling of the pumping station.

Images