Method for regulating high-density composite fluid for pumped storage system

By establishing an objective function and constraint library, a time-varying dynamic operating envelope of three control variables is generated for online estimation and inverse optimization. This solves the problem of decoupling medium design and system operation in high-density composite fluid pumped storage systems, realizes the synergistic optimization of fluid properties and unit operation, and improves the system's operating efficiency and safety.

CN121093862BActive Publication Date: 2026-02-03NANJING HYDRAULIC RES INST
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202511642394.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-11-11
Publication Date
2026-02-03
Estimated Expiration
2045-11-11

AI Technical Summary

Technical Problem

Existing technologies for applying high-density composite fluids to pumped storage suffer from the problem of decoupling between medium design and system operation, resulting in uncontrollable fluid properties, reduced system operating efficiency and safety, and an inability to dynamically respond to grid demands.

Method used

By establishing a system objective function and constraint library, a time-varying dynamic operating envelope of three control variables is generated. Sensor data is received for online estimation, reverse optimization is performed, and a joint schedule is generated to achieve coordinated optimization of fluid properties and unit operation.

Benefits of technology

It improves the system's economy, safety, and dynamic response to grid demand, and ensures the stability and controllability of fluid properties in the pumped storage system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121093862B_ABST
    Figure CN121093862B_ABST
Patent Text Reader

Abstract

The application discloses a high-density composite fluid regulation method for pumped storage system, comprising: based on basic data, establishing system target function and constraint library containing machine stability index and shear-sedimentation coupling stability window constraint; generating time-varying three-control dynamic operation envelope according to unit and pipe network parameter library; receiving sensor data stream, analyzing combined characteristics such as machine side current harmonic and main loop pressure pulsation, and obtaining state vector containing apparent viscosity, density and other parameters through non-invasive online estimation; under the constraint of dynamic operation envelope, reverse optimization is carried out to generate joint schedule containing unit operation trajectory and fluid property target trajectory. The application puts fluid properties into active regulation category, realizes collaborative optimization of high-density energy storage medium and conversion equipment, and improves the economy, safety and dynamic response ability to power grid demand of system operation.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fluid simulation and regulation, and particularly to a high-density composite fluid regulation method for pumped storage systems. BACKGROUND

[0002] As a large-scale energy storage method, pumped storage is a key support for ensuring the safety and stability of the power system and promoting the consumption of new energy. To further solve the problem that traditional pumped storage is limited by terrain elevation and has limited potential for energy density improvement, high-density composite fluid can be used as an energy storage medium. By uniformly dispersing high-density and low-cost natural mineral particles in a water-based medium, the system's energy storage capacity can be increased several times under the same reservoir capacity and water head conditions, providing the possibility of building efficient pumped storage power stations in plain areas or low water head scenarios, while expanding the application boundaries of energy storage technology and promoting the construction of new power systems.

[0003] Currently, the preparation and transportation technology of high-density composite fluids (such as ore slurry, drilling fluid, and high-concentration concrete) has been widely used in mining, oil, and construction fields. Existing technologies focus on the offline characterization of fluid formulation design and rheological properties, for example, by measuring viscosity, yield stress, and other parameters under different proportions using a laboratory rheometer, and selecting a formulation that meets the requirements of specific working conditions (such as long-distance pipeline transportation). On the other hand, at the operation control level, the fluid is considered as a controlled object with basically constant physical properties, and the control strategy mainly focuses on the adjustment of pumping, valves, and other transportation equipment to ensure the stability of the transportation process by controlling flow, pressure, and other macroscopic parameters. Overall, after determining the fluid formulation, the design of the transportation system and control strategy adapts to this formulation.

[0004] However, when applying existing technologies directly to pumped storage, there is a decoupling problem between medium design and system operation, which fixes the fluid properties, and the lack of state observation reduces the overall efficiency, safety, and response capability of the system to the grid demand. Therefore, further research and innovation are needed to solve the above problems existing in the prior art. SUMMARY

[0005] The present application relates to the field of fluid simulation and regulation, and particularly to a high-density composite fluid regulation method for pumped storage systems.

[0006] Technical solution: According to one aspect of the present application, a high-density composite fluid regulation method for pumped storage systems includes:

[0007] Based on the input basic data, a system objective function and a constraint library are established; the basic data includes grid service demand, unit and pipe network parameter library, fluid candidate property library, and economic and maintenance parameters, and the constraint library includes constraints of passable machine stability index and shear-sedimentation coupling stability window.

[0008] According to the unit and pipe network parameter library, a time-varying three-control variable dynamic operation envelope is generated.

[0009] A sensor data stream is received, and a state vector is obtained online according to the dynamic operation envelope, the state vector including an apparent viscosity estimated value and a solid phase volume fraction estimated value.

[0010] A system objective function, a constraint library, a dynamic operation envelope and a state vector are received, and a joint schedule is generated by reverse optimization under the constraint of the dynamic operation envelope; the joint schedule includes a unit operation trajectory and a fluid property target trajectory.

[0011] Beneficial effects: The fluid property is included in the active regulation category, the collaborative optimization of the high-density composite energy storage medium and the conversion equipment is realized, and the economic efficiency, safety and dynamic response ability to the power grid demand of the system operation are improved. The related technical effects will be described in detail below in combination with specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0012] Fig. 1 A flowchart of a high-density composite fluid regulation method for a pumped storage system provided by the embodiments of the present application.

[0013] Fig. 2 A flowchart of reverse optimization provided by the embodiments of the present application.

[0014] Fig. 3 A flowchart of generating a time-varying three-control variable dynamic operation envelope provided by the embodiments of the present application.

[0015] Fig. 4 A flowchart of performing control provided by the embodiments of the present application. DETAILED DESCRIPTION

[0016] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0017] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] To address the aforementioned issues, the applicant conducted in-depth searches and analyses, and discovered:

[0019] Existing technologies generally suffer from a deep decoupling between medium design and system operation. This results in fluid properties being fixed during the design phase and becoming uncontrollable boundary conditions during operation. The system can only passively adapt and cannot proactively and collaboratively adjust fluid properties and unit operating status according to dynamic changes in grid service demands (such as switching from stable power generation to rapid frequency regulation), thus missing the optimal solution at the system level.

[0020] Furthermore, the decoupling problem is exacerbated by the lack of state observation. Deploying insertion rheometers for real-time property monitoring within closed, high-pressure, high-velocity pumped storage networks is neither practical nor safe, leaving operators unable to perceive critical states of the fluid such as its true apparent viscosity, solid phase distribution, and gas content. This makes it impossible to formulate a reasonable control strategy, forcing the adoption of a passive, reactive control mode based on a coarse model and a large safety margin.

[0021] To solve these problems, combined with Figs. 1 to 4 The present invention will be specifically described through the following embodiments.

[0022] Example 1 describes optional high-density composite fluid screening, evaluation and preparation methods, providing an implementable object for subsequent control methods, namely high-density composite fluid, and providing a technical basis for the establishment of a fluid candidate property library.

[0023] In this embodiment, the method for obtaining the high-density composite fluid for the pumped storage system includes screening and evaluating raw materials, theoretically calculating the proportions, and a specific preparation process. Specifically, it includes:

[0024] Step S1.1: Establish a fluid composition evaluation system and raw material screening. Accordingly, a multi-dimensional evaluation index system is established for systematically screening solid raw materials; preferably, the system takes screening the optimal high-density composite fluid material as the target layer; five aspects, namely density, cost, environmental friendliness, safety, and stability, are selected as the criterion layer; the index layer includes specific sub-indicators corresponding to each dimension of the criterion layer.

[0025] Among them, the density index (ρ) is the parameter that determines the overall density of the fluid; the cost index (C) can be further decomposed into sub-costs such as extraction, transportation, processing, storage, and maintenance; the environmental index (E) can consider carbon emissions, waste concentration, and reuse rate in the production process; the safety index (A) addresses personnel health and production safety, such as toxicity, leakage risk, and dust explosion; and the stability index (S) integrates physical stability (such as settling velocity), chemical stability (such as corrosion resistance), and thermal stability (such as the coefficient of thermal expansion). As an example, the evaluation method uses the Analytic Hierarchy Process (AHP) to construct a judgment matrix to determine the weights of the above five indicators.

[0026] Step S1.2: Determine the initial fluid ratio based on theoretical calculations. Further, after selecting the solid raw materials, theoretical calculations can be performed to initially obtain a ratio that meets the basic performance requirements. For example, based on the target density of the composite fluid, the required volume fraction φ of the solid components is calculated using the following formula. s =(ρ k -ρ1) / (ρ s -ρ1);

[0027] Where, ρ k The target density of the composite fluid, in kg / m³ 3 Depending on the energy storage requirements of different pumped storage systems, its value can generally be set to 2000 kg / m³. 3 Up to 4000 kg / m 3 ;ρ l The density of the liquid medium is used; when water is used as the base liquid, the density of water is usually taken as the density, and the unit is kg / m³. 3 ;ρ s The density of the selected solid particles, in kg / m³ 3 .

[0028] Based on this, the mixing viscosity at this initial ratio can be further predicted, and the calculation formula can be used as follows:

[0029] η k =η l *(1+2.5*φ s +6.2*φ s 2 ); where η kThe predicted mixing viscosity of the composite fluid is given in Pa·s; η l The viscosity of the liquid after the additive is mixed with the water-based medium is expressed in Pa·s; φ s η represents the volume fraction of solid components calculated in the previous step, and is a dimensionless value. To ensure good pumpability of the fluid, η... k The value of η needs to satisfy certain constraints, such as η k It should not exceed 10 Pa·s.

[0030] Step S1.3: Fluid Preparation. In this step, the preparation process can employ different technical approaches depending on whether the selected mineral is metallic or non-metallic. For example, for a system using barite (non-metallic) as raw material, the process may include crushing, surface coating with silane coupling agent KH550, ultrasonic dispersion, and aging. For a system using hematite (metallic) as raw material, the process may include magnetic separation purification, surface coating with stearic acid, high-speed shear dispersion, and magnetic separation recovery.

[0031] Example 2 describes a method for determining fluid formulations based on machine learning and multi-objective optimization, used to solve the proportions of each fluid component, providing algorithmic support for the construction of a fluid candidate property library.

[0032] Specifically, this method simultaneously incorporates multiple interdependent performance indicators such as fluid density, flowability, and stability into a multi-objective coupled optimization framework for solution. A joint objective function is established, and an exemplary mathematical expression can be:

[0033] F=w1*(ρ 实际 / ρ 目标 )+w2*(η 阈值 / η 预测 )+w3*(u 阈值 / u 预测 );

[0034] Where F is the comprehensive objective function to be optimized, and the smaller its value, the better the proportioning scheme; ρ 实际 The actual density calculated for the current mix proportions, ρ 目标 To design the target density; η 预测 η is the predicted viscosity for the current mix ratio. 阈值 The maximum permissible viscosity of the system; u 预测 For the predicted settlement rate of the current mix ratio, u 阈值 The maximum allowable settlement rate; w1, w2, and w3 are the weighting coefficients for each performance parameter, which can be adjusted according to specific engineering requirements.

[0035] In this embodiment, a hybrid strategy combining theoretical calculation and machine learning can be adopted to solve the above optimization problem. Accordingly, a multi-objective coupled theoretical model based on Navier-Stokes equation and particle dynamics theory is established to perform theoretical calculation of the initial formulation and predict the fluidity and stability level.

[0036] Next, when the initial formulation obtained by theoretical calculation does not meet the constraint condition, or in order to find a better solution in a wider design space, a machine learning surrogate model is introduced for accelerated calculation. Specifically, a simulation data set covering the range of key parameters is generated by CFD numerical simulation or a small number of sample experiments, for example, 100 groups of data. The input features of the data set can include solid volume fraction, particle size distribution parameters, additive concentration, etc., and the output targets are density, viscosity, sedimentation rate, etc.

[0037] Further, a machine learning model such as random forest or support vector regression (SVR) is trained. For example, the number of trees of the random forest can be set to 100, and the maximum depth is 10. On this basis, the trained machine learning surrogate model is embedded into a multi-objective optimization algorithm, such as multi-objective particle swarm optimization (MOPSO). In each iteration of MOPSO, the parameter combination corresponding to each candidate solution (particle) is input into the surrogate model to quickly predict its performance, instead of performing time-consuming CFD calculation, thereby improving the optimization efficiency. Exemplarily, the number of particles can be set to 50, and the number of iterations is 100. The algorithm finally outputs a set of Pareto optimal solutions, from which the decision maker can select the final fluid formulation according to the specific engineering preference.

[0038] It should be understood that traditional composite fluid preparation methods rely on repeated experiments after setting target values by humans, which has problems of high cost, long cycle, and large uncertainty. The method provided in this embodiment can provide a formulation based on performance indicators through multi-physical field coupled modeling and optimization.

[0039] Embodiment Three, the specific process of the preferred preparation process of the preferred high-density composite fluid is provided, which provides a basis for the realization of fluid performance.

[0040] Step S3.1, natural high-density ore material screening and solid particle size distribution determination. Preferably, a mineral with a density of not less than 4.0 g / cm 3 , good chemical inertness (such as acid and alkali loss rate less than 3%), and high natural enrichment degree is selected. For example, barite (BaSO4, density 4.3-4.7 g / cm 3 ) or hematite (Fe2O3, density about 5.3 g / cm 3 ) are two suitable candidate materials.

[0041] After the selected mineral is chosen, it is crushed and ground to optimize the particle size distribution. In order to achieve close packing of the particles, reduce the porosity and increase the fluid density, a multi-modal strategy based on the Fuller curve model is preferred. The mathematical expression of the model is: P(d) = (d / D _max ) n ;

[0042] where P(d) is the cumulative mass percentage of particles with a size less than d; d is the particle size in pm; D _max is the maximum particle size, which can be taken as 100 pm for example; and n is the Fuller index, which can be taken as between 0.35 and 0.45. Through this model, a combination of particles with different sizes (e.g., D 10 = 5 pm, D 50 = 25 pm, and D 90 = 80 pm) can be designed such that the final packing porosity is no more than 15%.

[0043] Step S3.2, surface modification treatment of the mineral particles is performed to improve the dispersibility of the particles in the liquid medium and reduce the wear of the pipeline. One preferred way is to use a silane coupling agent for coating. Specifically, the screened mineral particles are soaked in a KH550 ethanol solution with a concentration of 0.7wt% to 1.5wt%, stirred at a speed of about 2500 rpm for 30 minutes, and then centrifuged and dried. The KH550 molecules can react with the hydroxyl groups on the surface of the mineral through -Si-O- bonds to form a nanoscale coating layer, which can reduce the friction coefficient between the particles.

[0044] Further, in order to enhance the long-term stability of the fluid, a dispersant can be compounded. For example, 0.3wt% sucrose fatty acid ester and 0.1wt% carboxymethyl chitosan, two bio-based dispersants, can inhibit the agglomeration and sedimentation of particles through electrostatic repulsion and steric hindrance effects.

[0045] Step S3.3, the mixture of the composite is mixed and stabilized. The surface-modified particles (e.g., accounting for 55wt% of the total mass) are mixed with water and other additives such as xanthan gum, carboxymethyl cellulose sodium and other flow modifiers. In order to mix the components uniformly and exclude the tiny bubbles in the slurry, the mixture is placed in an ultrasonic tank with a power of not less than 500W and a frequency of about 30 kHz for 30 minutes, while vacuum degassing (e.g., vacuum pressure -0.1 MPa) is performed until the target viscosity error is no more than 5%. On this basis, aging can be performed at about 45°C for 4 hours to obtain a high-density composite fluid with stable performance.

[0046] Example 4. The adaptive retrofitting and shear-enhancing implementation of the optional pumped storage system is provided to make it capable of handling the high-density composite fluid involved in the present application. It provides the physical basis for the implementation of the regulation method.

[0047] One possible retrofitting approach is to add an online sensor network at key nodes of the circulating pipeline, such as the inlet and outlet of the pump / turbine, long straight pipe sections, elbow sections, etc. Preferably, the network can include non-invasive capacitance particle counters and ultrasonic rheometers. The capacitance particle counter is used to monitor the volume fraction of solid particles and its distribution uniformity in the cross-section of the pipeline in real time, with a monitoring accuracy of ±0.1%; the ultrasonic rheometer analyzes the attenuation and Doppler shift of the ultrasonic signals propagating in the fluid by transmitting and receiving ultrasonic signals, and inversely calculates the real-time apparent viscosity and flow velocity profile of the fluid, with a sampling rate of up to 10 kHz. By deploying the sensor network, the state parameters required for subsequent regulation decisions can be obtained.

[0048] Another possible retrofitting approach is to add or optimize shear force application devices for the fluid, actively maintaining or restoring the fluidity of the fluid to prevent irreversible settling or flocculation when the fluid is at low flow rate or at rest. Since the fluid used in the present application has shear-thinning properties, i.e., its apparent viscosity decreases at high shear rates. The shear-enhancing implementation can use static shear-enhancing units integrated into the pipeline system. For example, static mixers can be installed in specific sections of the pipeline, and the helical blade or baffle structure inside the static mixers can force the fluid to generate turbulent flow and enhance shear. Alternatively, a Venturi structure can be provided in the pipeline, and when the fluid flows through the throat section, the flow rate increases sharply (e.g., more than 5 m / s), generating extremely high local shear rates (e.g., up to 5000 s -1 above) to break up particle agglomeration.

[0049] Further, dynamic shear-enhancing devices can also be used. For example, a high-pressure homogenizer can be introduced to process the fluid in a bypass or main loop. When the fluid passes through the micro-porous channel under a pressure of up to 50 MPa, a local shear rate of more than 10 5 s -1 can be generated to force the destruction of particle agglomeration. In addition, a variable frequency pump can be used to apply periodic pressure fluctuations to the system to form a pulse flow. The instantaneous high shear rate (e.g., pulse peak value greater than 1000 s -1 ) of such pulse flow can maintain the fluid in the entire system in a low viscosity state at a relatively low average energy consumption.

[0050] Further, multi-stage pump groups are also used to cooperate with shear. In large pumped storage systems, multi-stage centrifugal pump groups can be used in series to achieve step-by-step increase in fluid shear rate (e.g., 100 s -1, the second stage 500s -1 , the third stage 1000s -1 ), both to achieve the overall shear requirements, but also to avoid a single pump due to excessive load and damage.

[0051] In some embodiments, the above solutions can be combined. For example, static mixers are installed in long straight pipe sections to provide basic shear, high-pressure homogenizers are provided at the outlet of the liquid storage tank for cyclic treatment, and the frequency conversion function of the main circulating pump is used to apply periodic pulse flow, forming a multi-level, full-process shear guarantee system; in other words, a multi-link collaborative shear guarantee system is formed.

[0052] In this embodiment, the hardware of the conventional pumped storage system is adapted or added to enable the high-density composite fluid to operate stably in the pumped storage system for a long time, and the state of the high-density composite fluid can be monitored and controlled.

[0053] Embodiment five describes the overall framework of the high-density composite fluid and the pumped storage system, and provides a high-density composite fluid control method for a pumped storage system, which specifically comprises:

[0054] Step S5.1, based on the input power grid service demand, unit and pipe network parameter library, fluid candidate property library, and economic and maintenance parameters, a system objective function and a constraint library are established, and the constraint library includes constraints of the passable machine stability index and the shear-settlement coupling stability window.

[0055] In other embodiments, based on the input basic data, a system objective function and a constraint library are established; the basic data includes power grid service demand, unit and pipe network parameter library, fluid candidate property library, and economic and maintenance parameters, and the constraint library includes constraints of the passable machine stability index and the shear-settlement coupling stability window.

[0056] Specifically, the input power grid service demand refers to the functional requirements of the power grid on the pumped storage power station, such as rapid climbing ability, primary frequency modulation response performance, daily start-stop timing, etc. The unit and pipe network parameter library is a database describing the characteristics of physical equipment, including head, guide vane profile, speed range, pipe diameter and bend information, material abrasion characteristics, etc. The fluid candidate property library stores physical and chemical properties of different fluid formulations, such as the relationship curve of density, apparent viscosity and shear rate, solid volume fraction, particle size distribution, etc. The system objective function is a mathematical expression for evaluating the comprehensive benefits of system operation, which can maximize the energy density, minimize the operation loss and maintenance cost while meeting the power grid demand. The constraint library defines the boundary of safe operation of the system, among which the passable machine stability index and the shear-settlement coupling stability window are constraint conditions for characterizing the safe operation boundary of the special fluid of the application, and the specific construction method will be described in subsequent embodiments.

[0057] Step S5.2, generating time-varying three-control-quantity dynamic operation envelope according to the unit and pipe network parameter library. In other words, generating time-varying three-control-quantity dynamic operation envelope according to the unit and pipe network parameter library in the basic data.

[0058] Wherein, the three control quantities refer to three control dimensions of regulating system state. In the preferred embodiment of the present application, it is specifically the guide vane-speed coordinated trajectory, bypass degassing setting and bypass local shear power. Correspondingly, the dynamic operation envelope can be understood as a safe flight corridor defined in the multi-dimensional control space, which will be dynamically adjusted with time and system state (such as fluid temperature, gas holdup). The control instruction at any time must be located inside the envelope, so that the system remains stable. In other words, the dynamic operation envelope is a range constraint of safe system operation dynamically defined in the multi-dimensional control space according to time and system state, and the control instruction must satisfy this constraint.

[0059] Step S5.3, receiving sensor data stream, and on-line estimating state vector according to the dynamic operation envelope, the state vector containing apparent viscosity estimate and solid phase volume fraction estimate.

[0060] In some embodiments, receiving sensor data stream, and on-line estimating state vector containing apparent viscosity estimate based on the dynamic operation envelope. Further, the state vector also contains solid phase volume fraction estimate for representing density.

[0061] Wherein, the sensor data stream is mainly derived from the on-line monitoring network and the monitoring system of the unit itself (such as current, voltage, vibration, pressure signals, etc.). The present application adopts a preferred non-invasive estimation method, which does not directly measure the fluid properties, but inversely estimates the state vector containing apparent viscosity estimate, solid phase volume fraction estimate, gas holdup estimate, cavitation margin estimate and other key internal states by analyzing external and easily accessible signal characteristics.

[0062] Step S5.4, receiving system objective function, constraint library, dynamic operation envelope and state vector, and performing reverse optimization under the constraint of the dynamic operation envelope, to generate a joint schedule, which contains both unit operation trajectory and fluid property target trajectory.

[0063] Wherein, the reverse optimization is not to first give the unit operation and then passively adapt to the fluid state, but to also take the fluid properties as an optimizable target. The joint schedule generated thereby is a time series plan, which not only specifies how the guide vane should be opened and the speed should be adjusted in the future (i.e. the unit operation trajectory), but also specifies what target level the apparent viscosity and solid phase volume fraction of the fluid should be maintained at (i.e. the fluid property target trajectory), achieving the coordinated optimization of energy medium and energy conversion equipment.

[0064] Embodiment six, the construction of the passability stability index and the coupling stability window in system modeling, as an optional technical solution in the process of constraint library construction.

[0065] Step S6.1, constructing the passability stability index. The passability stability index is composed of at least four sub-items weighted: cavitation margin index, erosion life ratio, key clearance to particle size ratio, and population settling risk index; it is a comprehensive safety evaluation index for quantifying the overall risk of high-density composite fluid passing through the water pump water turbine. A preferred calculation method is:

[0066] G _pass = _cav *Cavitation _margin + _ero *Erosionlife _ratio + _clr *Clearance _ratio + set *(1-Settling _risk );

[0067] Wherein, G _pass is the final score of the passability stability index; w _cav , w _ero , w _clr , w set are the weight coefficients of each sub-item, which can be determined according to expert experience or historical data in the system modeling stage, and are adjusted adaptively in operation; Cavitation _margin is the cavitation margin index, which is used to evaluate the risk of cavitation of the fluid in the low pressure area such as the impeller inlet; Erosionlife _ratio is the erosion life ratio, which is based on the characteristics of solid particles in the fluid and the material of the unit, and predicts the wear rate of key components (such as guide vanes, impellers); Clearance ratio is the key clearance to particle size ratio, which evaluates the risk of particles with the maximum particle size blocking or jamming the narrow gaps such as guide vanes and seals; Settling _risk is the population settling risk index, which evaluates the risk of large-area settling of particles and causes starting difficulty when the fluid is stopped or running at low speed. Such sub-items are calculated online or offline through measurable proxy quantities (such as pressure, flow, particle concentration, particle size D 90 value, etc.).

[0068] Step S6.2, constructing the shear-settling coupled stability window. Exemplarily, the shear-settling coupled stability window is defined by both pumpability and resuspendability conditions; and the boundary of the window is dynamically reconfigured according to the shear setting map. The limitation of traditional fluids using only a single viscosity threshold as the operating constraint is overcome.

[0069] Specifically, the pumpability condition defines that, under a given shear force, the fluid must exhibit a sufficiently low apparent viscosity to ensure smooth delivery by the water pump turbine without overloading. The resuspendability condition defines that, after the fluid is at rest or flows at a low speed for a period of time, the system must have the ability to apply sufficient shear force (e.g. through a shear reinforcement device) to resuspend the particles that have undergone preliminary settling, restoring the uniformity of the fluid.

[0070] Therefore, the stability window is a feasible operating region defined in the multi-dimensional space of apparent viscosity-shear rate-solid phase volume fraction-gas holdup, etc.

[0071] Further, the boundary of the window is not fixed. Its boundary is dynamically reconfigured according to the shear setting map. The shear setting map is obtained by calibrating the shear reinforcement device during the system debugging stage (e.g. calibrating the relationship between different bypass shear power and the amount of apparent viscosity reduction of the fluid). In actual operation, the system can adjust the boundaries of pumpability and resuspendability in real time and dynamically according to the currently available shear capacity (i.e. the shear setting map), so that the operating window is more flexible and closer to the actual working condition.

[0072] Embodiment seven, an optional generation method of time-varying three-control-quantity dynamic operating envelope is provided, which provides a premise for the subsequent reverse optimization. In this embodiment, the three control dimensions are coupled and constrained in a unified framework to form a multi-dimensional safe operating domain. The three control quantities are: guide vane-speed coordinated trajectory; bypass degassing setting; bypass localized shear power.

[0073] Further, the generation of the time-varying three-control-quantity dynamic operating envelope includes: generating a degassing setting map and a shear setting map; generating a guide vane-speed initial envelope for the first second of mode switching in view of the transient characteristics of the mixed-phase fluid; and synthesizing the degassing setting map, the shear setting map, and the guide vane-speed initial envelope for the first second of mode switching to obtain the time-varying three-control-quantity dynamic operating envelope. In other words, the generation process of the envelope is generated by synthesizing the following items: the degassing setting map; the shear setting map; and the guide vane-speed initial envelope for the first second of mode switching in view of the transient characteristics of the mixed-phase fluid.

[0074] Optionally, the generation process can be decomposed into the following steps:

[0075] Step S7.1, generating degassing setting mapping and shearing setting mapping. Wherein, the generation of degassing setting mapping is used to associate the executable control quantity (degassing intensity) with the system safety constraint (cavitation margin). Specifically, the function relationship or lookup table between the operating power / intensity of the bypass degassing device, the concentration of fluid dissolved gas, and the unit inlet cavitation margin can be fitted by a series of short-time tests on the bypass system or by analyzing historical data. In subsequent optimization, the constraint that the cavitation margin is not lower than a certain threshold can be directly converted into a lower limit constraint on the bypass degassing setting, making the optimization problem easier to solve.

[0076] Further, the generation of shearing setting mapping is used to quantify and incorporate the fluid property maintenance means with local strong shearing and low energy consumption into the control and optimization framework. Preferably, a local shearing device composed of a short section static mixer and a small flow pulsating micro-jet is installed at a position about 1 to 2 times the hydraulic diameter in front of the unit inlet. By conducting power scanning calibration experiments on the device, i.e. systematically changing its shearing power, pulsating frequency and other parameters under different main flow rates, the apparent viscosity response of the fluid is estimated in real time using a state observer, and a local response mapping table of shearing power-apparent viscosity-solid phase volume fraction is established. The mapping table is the shearing setting mapping, which quantifies how much shearing power can be used to achieve how much viscosity reduction.

[0077] Step S7.2, generating the initial envelope of guide vane-speed in the first second of mode switching. Optionally, during the initial few seconds of the working condition switching (such as from power generation to pumping) of the pumped storage unit, the hydraulic system will undergo a severe dynamic process, and for the composite fluid containing solid phase and gas phase, the risk is much higher than that of clear water. Therefore, the mode switching rule of the clear water model cannot be directly applied. This step designs the envelope for the key time window of the first second of mode switching. Preferably, a simplified approximate model is established to represent the transient characteristics of the mixed phase fluid, in which the mixed density takes into account the influence of gas content, and the equivalent bulk modulus and time constant are used to represent the compressibility and dissolution / precipitation effect of the gas. Next, based on the model, the coordinated trajectory of the guide vane opening and the speed is parameterized with low dimension (for example, the guide vane opening is described by a three-section function, and the speed is described by an S-shaped curve), and within the time window of 0 to 3 seconds, the hard constraints of cavitation margin not lower than a threshold and pressure fluctuation peak not exceeding the limit are used, and the optimization objective of minimizing the pressure fluctuation and shortening the stabilization time is used to perform limit search simulation. Through the simulation, a feasible guide vane-speed coordinated trajectory family is obtained, and the upper and lower boundaries of the guide vane opening and the speed changing with time in the first second of mode switching are solidified, forming the initial envelope.

[0078] Step S7.3, perform the envelope synthesis and online updating rule solidification. The degassing setting map, shearing setting map, and the initial envelope of the first second of the guide vane-speed are synthesized and projected in the feasible region on a unified time axis. Specifically, taking the initial envelope of the guide vane-speed as the main axis, the reachable region of the degassing and shearing is superimposed as an additional dimension on each time slice, and the boundary of the synthesized polyhedron is cropped according to global constraints such as the passability machine stability index, to form a parameterized overall dynamic operating envelope (E _env ).

[0079] In addition, the application also includes an online updating rule for the envelope; according to the state vector, the online updating rule for the envelope is applied to automatically adjust the boundary of the time-varying three-control-quantity dynamic operating envelope. For example, the sensitivity of the envelope boundary to the state vector (such as the gas holdup estimate) and the updating step are defined in advance. When it is observed in operation that the gas holdup continuously rises and the cavitation margin approaches the boundary, the rule is triggered to automatically narrow the allowed upper limit of the guide vane opening and raise the minimum value of the bypass degassing setting, realizing adaptive adjustment of the envelope and keeping the system safe.

[0080] Embodiment eight describes a non-invasive state vector estimation method based on multi-modal perception, as a technical decomposition of the state vector online estimation process. It avoids the use of plug-in flow variation probes that are prone to damage and difficult to maintain.

[0081] Among them, the online estimation contains the state vector of the fluid property, specifically including: analyzing the combined characteristics of the machine-side current harmonics and the main circuit pressure pulsation, and inversely calculating the apparent viscosity estimate and the solid volume fraction estimate. In other words, the combined characteristics of the machine-side current harmonics and the main circuit pressure pulsation in the sensor data stream are analyzed, and the apparent viscosity estimate and the solid volume fraction estimate are inversely calculated.

[0082] Specifically, the process performs synchronization and feature construction from the sensor raw data stream. For example, the three-phase current signals of the machine-side motor are subjected to Fourier transform to extract the energy of specific harmonics (such as 5, 7, 11, 13 times) and their energy ratio relative to the fundamental wave; the pressure sensor signals of the main circuit (such as the volute, the draft tube) are subjected to time-frequency analysis to extract statistical features such as the mean, variance, and kurtosis of the pressure pulsation envelope. Such features together constitute a fingerprint reflecting the internal state of the fluid.

[0083] Next, a nonlinear mapping model from the multi-dimensional feature set to the target physical quantity (apparent viscosity, solid volume fraction) is established. A preferred implementation is to use a Gaussian process regression (GPR) or kernel regression model. The training data for this model is constructed using a strong and weak label mixing strategy. Among them, the strong label data comes from the rheological measurement of fluid samples in the offline laboratory; the weak label data is the apparent viscosity value indirectly estimated by the easily measured pressure drop, flow rate, temperature, etc. data in the actual operation of the system through the known pipe geometry parameters and the basic fluid dynamics equation (such as the generalized Darcy-Weisbach formula). By combining the two labels for training, the generalization ability and robustness of the model under real working conditions can be improved.

[0084] Further, according to the operating condition, the model domain is discriminated; based on the discriminated model domain, the gas-containing correction sub-model or the non-Newtonian main model is selectively applied to perform inversion; for dealing with the dramatic change of fluid characteristics of the pumping and storage system under different working conditions.

[0085] Specifically, the system can automatically distinguish whether the current is in the severe gas-containing disturbance domain of the first second of mold switching or in the non-Newtonian characteristic dominant domain of stable operation according to the mold switching state flag, the significance of the high-frequency component in the pressure fluctuation spectrum, etc. If it is discriminated as the former, the gas-containing correction sub-model for compensating the effects of gas phase on the equivalent density and viscosity of the fluid is superimposed on the basis of the above-mentioned inversion model; if it is the latter, the non-Newtonian main model is mainly relied on for estimation. The explicit domain switching mechanism solves the problem that a single model cannot accurately describe both transient and steady-state behaviors.

[0086] In addition, the application also includes a consistency regularization method, wherein the round-trip efficiency or pressure fluctuation stability is used as a regularization term to constrain the inversion process; for improving the physical consistency and long-term stability of the inversion.

[0087] In the training target (loss function) of the machine learning model, in addition to the conventional prediction error term, a consistency regularization term is also added. The regularization term is used to punish the prediction results that although well fitted on the instantaneous data, but will lead to the macroscopic system performance indicators (such as the calculated unit round-trip efficiency, pressure fluctuation stability index) inconsistent with the actual situation or physical laws. For example, if the model predicts a very low viscosity value, but the actual measured shaft power and hydraulic loss are very high under this working condition, the regularization term will impose a greater penalty on it. In this way, the macroscopic system knowledge is integrated into the microscopic state estimation, making the estimation result more reliable.

[0088] As an optional implementation manner, when the system is configured with the ultrasonic sensor, the estimation precision of the gas holdup can also be improved through a multi-source fusion estimator. The estimator can weightedly fuse three pieces of information: an acoustic channel based on amplitude attenuation and velocity profile distortion of ultrasonic echoes; a pressure vibration channel based on energy change of a pressure pulsation characteristic frequency band; and an electric parameter channel based on responses of current harmonic energy and power angle swing to load disturbance.

[0089] Embodiment Nine, describes a reverse optimization-based operation-recipe joint schedule synthesis method as a technical decomposition of reverse optimization. Used for simultaneously optimizing how a unit operates and what state a fluid should be in.

[0090] Step S9.1, perform reverse optimization, specifically implemented by constructing a decision vector and solving it, the decision vector containing at least two types of variables: the first type is a unit operation control variable, used to determine a unit operation trajectory, which contains a guide vane opening trajectory and a rotating speed trajectory; the second type is a fluid property target control variable, used to determine a fluid property target trajectory, which contains a target apparent viscosity trajectory, a target solid volume fraction trajectory and a target gas holdup trajectory.

[0091] In other words, within a given optimization time domain (for example, 15-60 minutes in the future), the solver needs to give a series of time series curves at the same time, including unit operation instructions such as guide vane opening g(t) and rotating speed n(t), and fluid state expectation values such as target apparent viscosity Visc _target (t), target solid volume fraction Solid _target (t), etc. The above two types of variables are coupled in a unified decision vector and serve as a solution to the optimization problem together.

[0092] Step S9.2, further, according to grid service requirements, decompose a regulation period into a plurality of service segments to form a service segment set; perform optimization solving based on the service segment set, and apply cross-segment splicing constraints at boundaries of adjacent service segments to ensure that each trajectory in the decision vector meets preset limits in continuity and change rate. Used for processing long-period operation scheduling problems, so that the optimization result has smoothness and executability,

[0093] Specifically, the 24-hour operation schedule can be decomposed into several service segments with different objectives, such as morning climbing segment, midday frequency modulation segment, night pumping segment, etc. The optimization solver optimizes each segment for its main objective. At the boundary of two segments, a cross-segment stitching constraint must be imposed, which requires that the values of all state variables (e.g. guide vane opening, speed, target viscosity) at the end of the former segment must be equal to the values at the beginning of the latter segment (zero-order continuity), and their rates of change (slopes) must also be smoothly transitioned (first-order continuity) without exceeding the maximum allowable rate of change of the equipment. This strategy reduces the complexity of the overall cycle optimization and ensures the physical realizability of the final schedule.

[0094] Step S9.3. When optimizing any service segment, adaptively calculate multiple weights according to the grid importance, equipment state, or prediction uncertainty of the service segment, form a weight set and apply it to the system objective function. Or, when optimizing any service segment, adaptively calculate multiple weights according to the grid importance, equipment state, or prediction uncertainty of the service segment, form a weight set; apply the weight set to the system objective function. This allows the optimization decision to adapt to changing grid conditions and equipment states.

[0095] For example, in the objective function J _total , there are energy density weight w _energy , loss term weight w _loss , risk term weight w _risk , etc.; where the weights are not fixed. When handling grid emergency support service segments, the system will automatically increase the weight of w _energy (corresponding to energy density or power response), and appropriately reduce the weight of w _loss (corresponding to energy consumption); when the state observer reports a high risk of equipment wear, the system will automatically increase the weight of w _risk ; when short-term prediction shows that future load uncertainty is large, the system will also increase the risk weight, making the optimization result more conservative and robust.

[0096] As a preferred solution strategy, the idea of hierarchical optimization can be used to improve computational efficiency. Accordingly, a rough trajectory search is performed on the two-dimensional variables of guide vane and speed, which have the greatest impact on power, to quickly lock in a relatively optimal operating range. Next, within this range, optimization is performed on multiple variables that affect fluid properties, such as bypass shear power, degassing target, and dosing amount, to minimize energy consumption and risk.

[0097] Embodiment 10 describes an optional implementation of execution control and closed-loop learning, as a technical decomposition of the joint schedule being executed, monitored, and the system self-optimizing based on the execution results. In this embodiment, the execution of the joint schedule is not simply a open-loop instruction issuance, but an intelligent process containing collaborative control, risk monitoring, and closed-loop learning.

[0098] Step S10.1, correspondingly, the method further comprises execution control, specifically: the joint schedule is executed collaboratively by the main loop control unit and the bypass fluid regulation unit; the bypass fluid regulation unit dispatches the shearing, degassing, and dosing equipment to make the actual fluid properties track the fluid property target trajectory according to the principle of minimum energy consumption.

[0099] Specifically, after receiving the joint schedule, the upper-level interpretation module will perform setpoint interpretation and interlock check. For example, before issuing the instructions, the system will check the maximum change rate of the guide vane opening, the upper and lower limits of the rotational speed, and the mutual dependency between instructions. For example, if the state observer reports that the current gas holdup is higher than the target gas holdup trajectory in the joint schedule, the system will preferentially increase the power level of the bypass degassing unit, and actively limit the rising rate of the guide vane opening in the main loop, so as to maintain sufficient cavitation safety margin even if the degassing is not up to standard. After the instructions are issued, the main loop control unit is responsible for tracking the target trajectory of the guide vane opening and the rotational speed. The bypass fluid regulation unit, i.e., the bypass micro-factory, is responsible for tracking the target trajectory of the fluid properties with minimum additional energy consumption. Correspondingly, when the actual apparent viscosity fed back by the state observer is higher than the target value, start the low-energy local shearing device (such as a static mixer or a pulsating micro-jet); if the deviation is still large or the response speed does not meet the requirements, then gradually enable the higher-energy global shearing equipment (such as a high-pressure homogenizer) or the dosing unit of the thickening agent / diluent. This scheduling logic based on priority and minimum energy consumption makes the regulation of fluid properties in a more economical way; or in other words, this logic can achieve economic regulation of fluid properties.

[0100] Step S10.2, during the execution process, the system continuously performs risk monitoring and emergency handling. The execution control further comprises: determining whether the running state is close to or breaks through the boundary of the time-varying three-control-quantity dynamic running envelope in real time; as a basis for detecting risks; when a risk is detected, triggering an emergency degradation strategy. In other words, whether the running state is close to or breaks through the boundary of the time-varying three-control-quantity dynamic running envelope is monitored in real time, if the running state is close to or breaks through the boundary, it is determined that a risk is detected, and an emergency degradation strategy is triggered.

[0101] Accordingly, the process can be regarded as a package line guardian mechanism. The system compares the current operating point (consisting of state vector and measured control variables) with the dynamic operating package line in real time. Once it detects that the operating point has entered the warning zone of the package line boundary (e.g., less than 5% of the preset safety margin from the boundary) or has actually broken through the boundary, the system will immediately trigger an emergency degradation strategy. This strategy is independent of the upper-level optimization schedule and is executed with the highest priority. Specific degradation actions can include: forced compression of the allowable operating domain of the guide vane-speed, immediate increase of the lower limit of the bypass degassing setting, temporary increase of the intensity and frequency of the shear pulse; pulling the system state back to the inside of the safe package line at the fastest speed, and recording this risk event as an important learning sample.

[0102] Step S10.3, on this basis, the system realizes long-term self-optimization through closed-loop learning. Accordingly, based on the deviation between the execution record and the prediction, the boundaries of the time-varying three-control variable dynamic operating package line, the threshold values in the constraint library, and the cost model are re-calibrated.

[0103] Specifically, after the end of a running cycle (e.g., a day or a week), the closed-loop learning module is started. Key performance indicators (KPIs) such as actual round-trip efficiency, unit energy cost, cavitation event occurrence rate, etc. are calculated. Next, the actual KPIs are compared with the KPIs predicted by the optimization module when planning, and the prediction-actual deviation is calculated. Based on this deviation, the system re-calibrates its multiple models and parameters: using the accumulated real running data, the state observer model and the surrogate model are incrementally trained or parameter-regressed. Based on the records of risk events and boundary approaches actually occurring, the dynamic operating package line boundaries are corrected (e.g., if it is found that cavitation events increase when the water temperature rises in summer, the package line boundary under this working condition is automatically tightened), and the threshold values of the constraint library (such as G _pass

[0104] According to the actual measurement of bypass device energy consumption and reagent consumption, the cost model is updated, so that the subsequent optimization decisions are closer to the economic reality. Through the complete closed loop of design-operation-learning, the system can continuously adapt to the slow drift of fluid properties, the aging of equipment, and changes in the external environment, and realize long-term robust and optimal operation.

[0105] Embodiment eleven, provide specific reproducible numerical calculation cases.

[0106] According to one aspect of the present application, a numerical calculation case of comprehensive scoring of mineral raw materials based on the analytic hierarchy process (AHP) is provided to illustrate the implementation process of the comprehensive scoring system of mineral raw materials. The following methods can be used:

[0107] ​Assuming that the selection needs to be made from two candidate minerals, an evaluation system including five core indexes of safety index (A), environmental protection index (E), stability index (S), density index (p) and cost index (C) has been established.

[0108] Step S11.1, according to the pairwise comparison of the importance of each index by experts, the following 5-order judgment matrix A is set:

[0109] A = [[1, 3, 5, 7, 9], [1 / 3, 1, 3, 5, 7], [1 / 5, 1 / 3, 1, 3, 5], [1 / 7, 1 / 5, 1 / 3, 1, 3], [1 / 9, 1 / 7, 1 / 5, 1 / 3, 1]];

[0110] Step S11.2, the weight of each index is calculated by using the geometric mean method.

[0111] Accordingly, the geometric mean value w of each row element is calculated _i_geom : the geometric mean value w of safety index _A_geom = (1*3*5*7*9) 1 / 5 ≈ 3.920; the geometric mean value w of environmental protection index _E_geom = (1 / 3*1*3*5*7) 1 / 5 ≈ 2.036; the geometric mean value w of stability index _S_geom = (1 / 5*1 / 3*1*3*5) 1 / 5 = 1.000; the geometric mean value w of density index _ρ_geom = (1 / 7*1 / 5*1 / 3*1*3) 1 / 5 ≈ 0.528; the geometric mean value w of cost index _C_geom = (1 / 9*1 / 7*1 / 5*1 / 3*1) 1 / 5 ≈ 0.247.

[0112] Next, the above geometric mean values are normalized to obtain the final weight vector W. The sum Sum = 3.920 + 2.036 + 1.000 + 0.528 + 0.247 = 7.731; the corresponding weight of each item is: w _A = 3.920 / 7.731 ≈ 0.507; w _E = 2.036 / 7.731 ≈ 0.263; w _S = 1.000 / 7.731 ≈ 0.129; w _ρ = 0.528 / 7.731 ≈ 0.068; w _C = 0.247 / 7.731 ≈ 0.032; therefore, the weight vector W ≈ [0.507, 0.263, 0.129, 0.068, 0.032] T .

[0113] Step S11.3, correspondingly, the product A*W of the matrix A and the weight vector W is calculated, A*W≈[2.710, 1.351, 0.664, 0.350, 0.163] T ; further, the maximum eigenvalue λ _max =(1 / 5)*[(2.710 / 0.507)+(1.351 / 0.263)+(0.664 / 0.129)+(0.350 / 0.068)+(0.163 / 0.032)]≈5.236.

[0114] Next, the consistency index CI=(λ _max -n) / (n-1)=(5.236-5) / (5-1)≈0.059; wherein n is the order of the matrix, and here n=5. On this basis, the consistency ratio CR is calculated. According to the table (random consistency index table), for a 5-order matrix, the random consistency index RI is 1.12. CR=CI / RI=0.059 / 1.12≈0.053 Since CR≈0.053<0.1, the consistency test is passed, indicating that the weight distribution determined by the judgment matrix is reasonable and acceptable.

[0115] Step S11.4, assuming that the scores of each item of a certain natural hematite particle X are evaluated (on a 10-point scale): safety index 8.0, environmental protection index 10.0, stability index 10.0, density index 10.0, and cost index 9.5. Then the comprehensive score F _X =0.507*8.0+0.263*10.0+0.129*10.0+0.068*10.0+0.032*9.5≈9.03 points. Through this method, a quantitative comparison can be made on a variety of candidate materials, providing a scientific basis for the establishment of a fluid candidate property library.

[0116] According to another aspect of the present application, an example of a joint schedule generation calculation process can be generated using reverse optimization. The implementation is as follows: assuming that a service segment is a 15-minute rapid ramping task, requiring the power of the pumped storage unit to be linearly ramped from the current 100 MW to 300 MW. To simplify the calculation, the 15-minute segment is discretized into 4 time points t=0, t=5, t=10, t=15 minutes. The initial state (t=0), the current power P(0)=100 MW, the apparent viscosity estimate Visc _est (0)=8.0 mPa·s, and the solid volume fraction estimate Solid _est (0)=35%. The guide vane opening rate |Δg / Δt|≤5% / min; the rotational speed n is constant; the bypass shear power P _shear∈ [0, 10] kW. The apparent viscosity Visc must be maintained within the interval [5.0, 12.0] mPa-s. The power P ~ k _g * g(k _g = 10 MW / %); the head loss H _loss ~ k _v * Visc * Q 2 (Q is the flow rate, positively correlated with P; k _v is the head loss coefficient).

[0117] Solve Max J = w _p *∑[-(P(t) - P _target (t)) 2 ] - w l *∑[H _loss (t)]; where J is the total objective function; P _target (t) is the target power trajectory (100, 166.7, 233.3, 300 MW); w _p is the power tracking weight, set to 1.0; w l is the loss weight, set to 0.2. The decision vector at each time point t (t = 5, 10, 15) contains: {g(t), Visc _target (t), P _shear (t)}. The optimizer needs to solve 3*3 = 9 variables, maximizing the objective function J while satisfying all constraints.

[0118] One possible optimization result is shown below: {time (t, min), target power (MW), [unit operation trajectory] guide vane opening g(t) (%), [fluid property target] target viscosity Visc _target (t) (mPa-s), [fluid regulation trajectory] bypass shear power P _shear (t) (kW)} = {(0, 100, 10.0, 8.0, 0.0), (5, 166.7, 16.7, 7.0, 2.5), (10, 233.3, 23.3, 6.5, 3.8), (15, 300, 30.0, 6.0, 5.0)};

[0119] It can be seen that in order to achieve a fast power ramp-up, the joint schedule given by the optimizer is not simply a linear increase in guide vane opening g(t). It also plans a trajectory of actively reducing the target apparent viscosity Visc _target (t) (from 8.0 mPa-s to 6.0 mPa-s). Because at higher power and flow rate, higher viscosity will cause a disproportionately large increase in head loss (H _loss is proportional to Q 2 ), affecting the actual output power and response speed.

[0120] Further, the schedule also gives the bypass shear power P _shear (t) to achieve this viscosity target trajectory. At t=5min, 2.5kW of shear power is needed to reduce the fluid viscosity from 8.0 to 7.0mPa·s. As the power continues to rise, the required shear power also increases accordingly. In summary, the joint schedule embodies the synergistic optimization of operation and formulation: by actively regulating the fluid property (reducing viscosity) during the ramp-up process, the economic (though shear power is consumed, more hydraulic loss is reduced) and response performance are achieved to meet the grid service requirements.

[0121] According to an aspect of the present application, for the inversion of the apparent viscosity process, the specific algorithm can be: a Gaussian process regression (GPR) model is constructed with a multi-dimensional feature vector as input and apparent viscosity as output. Wherein, the multi-dimensional feature vector X _feat Specifically, it can include: current fundamental amplitude, 5 / 7 / 11 / 13 harmonic energy ratio to fundamental energy, main circuit pressure signal energy spectrum density integration in 0-10Hz frequency band, kurtosis and kurtosis coefficient of pressure signal envelope. The GPR model maps the input features to a high-dimensional space for nonlinear regression through a kernel function (for example, a radial basis function (RBF) kernel is preferably used to determine the automatic correlation), and its output not only contains the predicted mean value of apparent viscosity, but also contains the confidence interval (variance) of the prediction result, which can be used to evaluate the reliability of the estimation result.

[0122] According to an aspect of the present application, for the dynamic reconstruction process of the boundary of the window according to the shear setting mapping, the specific method of dynamic reconstruction can also be: correspondingly, under the condition of no active shear application, the shear-settlement coupling stable window Wss0 of the reference is determined, and the pumpable boundary is Visc _max defined. The resuspendable boundary corresponds to the viscosity Visc _resus_base defined. Before operation, the bypass shear power P _shear and the function relationship between the apparent viscosity reduction amount ΔVisc of the fluid, that is, the shear setting mapping ΔVisc=f(P _shear ).

[0123] In real-time operation, the controller calculates the potential viscosity reduction amount ΔVisc _shear_avail that can be achieved by mapping according to the maximum bypass shear power P _potential currently available.

[0124] The controller expands the boundary of the reference window Wss0 to generate the dynamically reconstructed window Wss _newSpecifically, the new upper limit of pumpable viscosity can remain unchanged or be appropriately relaxed, while the new resuspension boundary corresponds to the viscosity Visc. _resus_new Then update to Visc _resus_base +ΔVisc _potential Thanks to the assistance of active shearing, the system can now allow the fluid to reach higher viscosity when at rest, as it has the ability to reactivate it.

[0125] According to another aspect of this application, the physical basis, mathematical construction, and information fusion method of the state inversion model are provided. The rheological properties of high-density composite fluids modulate the vibration characteristics of hydraulic machinery systems, which in turn couple to the electromagnetic system of the drive motor. Specifically, an increase in the apparent viscosity of the fluid increases the damping between the impeller and the fluid, thus stronger suppressing the high-frequency torsional vibration of the pump-turbine rotor system. Torsional vibration, through magnetic field coupling, excites harmonic components of specific frequencies in the stator current. Therefore, the amplitude and energy distribution of the current harmonics indirectly reflect the magnitude of the fluid's viscous damping, and a complex nonlinear transfer function relationship exists between the two.

[0126] Furthermore, the low-frequency components of the pressure pulsation signal in the main loop are mainly related to macroscopic hydraulic phenomena such as water hammer waves and impeller rotation frequency, while the high-frequency components are related to microscopic phenomena such as turbulence, vortex shedding, and cavitation bubble collapse within the flow channel. When the volume fraction of the fluid solid phase or the gas content changes, it alters the fluid's compressibility and density, affecting the propagation speed and attenuation characteristics of the pressure wave, which is reflected in the spectral structure and statistical characteristics of the pressure pulsation signal.

[0127] Based on the above physical foundation, this embodiment preferably uses a Gaussian process regression (GPR) model as the state observer model H. _obs The specific structure. Used for inverting apparent viscosity Visc. _est The GPR model can be formally defined as:

[0128] Visc _est =f(X _feat )+ε;

[0129] Among them, Visc _est X represents the apparent viscosity to be estimated. _feat ε is a multidimensional feature vector extracted from sensor data; ε is a noise term that follows a Gaussian distribution, ε ~ N(0, σ). n 2 f is a Gaussian process, f ~ GP(m(X)); _feat ), k(X _feat X _feat Specifically, m(X) _feat k(X) is the mean function, which can usually be set to zero. _feat, X _feat ') is a kernel function (or covariance function) that defines the similarity between any two input feature vectors. One preferred kernel function is the squared exponential (SE) kernel with automatic relevance determination (ARD):

[0130] k(x _i , x _j )=σ f 2 *exp(-0.5*Σ[(x _i , d-x _j , d) 2 / l d 2 ]);

[0131] where x _i and x _j are two different feature vectors; d is the dimension index of the feature vector; σ f 2 is the signal variance, representing the average deviation of the model output; l d is the length scale on dimension d, which is a hyperparameter indicating the degree of function smoothness on that dimension, and the smaller l d , the more drastic the change of the function on that dimension. The ARD mechanism allows learning an independent length scale l d for each feature dimension, which can automatically assess the relative importance of different input features (such as current harmonics, pressure pulsations) on the inversion results, and it belongs to the data-driven weight determination method.

[0132] Embodiment twelve, describes the specific implementation of the closed-loop learning and rolling re-calibration algorithm, taking the rolling re-calibration of the observer model H _obs as an example. Specifically as follows:

[0133] Step S12.1, the system continuously records the feature vectors X _feat generated by the feature extraction module and the estimated values y _est (output by the observer model, for example, apparent viscosity) in the background. Accordingly, the system will label a confidence score for each sample according to the stability of the operating conditions (for example, the power, flow rate, and speed fluctuate less than 2% within 10 minutes) and the consistency of the physical model (for example, the deviation between the weak label calculated based on the pressure drop and flow rate and the model output is less than 5%).

[0134] Step S12.2, set the re-calibration trigger condition. Optionally, the condition can be a combination of time-driven (for example, start once every 168 hours) and event-driven (for example, when the average prediction variance of the observer model is continuously higher than the threshold, or when the KPI predicted-actual deviation exceeds 10%).

[0135] Step S12.3, when re-calibration is triggered, the system filters out the N samples with the highest confidence scores (e.g., N=1000) from the recent accumulated data. The high-confidence samples are used as new training data to update the GPR model. Preferably, an online sparse Gaussian process regression algorithm is employed, which can incorporate new information into the existing model without storing all the historical data, update and maintain the basis vector set, and achieve incremental learning of the model.

[0136] Step S12.4, the updated model H _obs_new is evaluated on an independent validation set. Only when its performance (e.g., root mean square error) is better than the old model H _obs_old , it is officially deployed into the online system. Accordingly, the system records the version number and update log of the model.

[0137] Step S12.5, for recipe drift correction, its implementation can be linked with the above process. Specifically, if after multiple model re-calibrations, it is found that the hyperparameters of the GPR model, especially σ f 2 or σ n 2 representing the base noise, show a persistent and one-way drift, indicating that the baseline physical properties of the fluid itself have undergone irreversible changes (i.e., recipe drift). At this time, the system triggers a recipe drift alarm, prompting the need for fluid sampling and offline laboratory re-calibration, and updating the baseline parameters in the fluid candidate property library according to the new measurement results.

[0138] Embodiment XIII, an optional technical solution describing the formal mathematical modeling of the inverse optimization problem, which can be formalized as a nonlinear optimal control problem.

[0139] Exemplarily, for a given service segment, whose time domain is [t _0 , t _f ], the objective to be solved is to maximize (or minimize) the comprehensive performance index J, which generally has the form:

[0140] min J=∫ t_0 t _f L(x(t), u(t), t)dt + M(x(t _f ));

[0141] Where t is time; x(t) is the state vector of the system, including power, flow, and estimated apparent viscosity, solid content, etc.; u(t) is the decision vector (i.e., joint schedule), u(t) = [g(t), n(t),..., Visc _target(t),...]; L(·) is the Lagrangian term, representing the instantaneous cost or benefit during the operation, for example, L = w _loss *P _loss (x, u) - w _track *(P(x) - P _target (t)) 2 , where P _loss is the total power loss, P(x) is the actual power, P _target is the target power; M(·) is the terminal cost term, used to evaluate the state at the end of the segment, w _track is the weight coefficient of the power tracking target.

[0142] Further, the decision vector u(t) is a function of time, whose components include the unit operation control variables and the fluid property target control variables. The solution must be carried out under the following multi-type constraint conditions: the state evolution of the system must follow its physical law, which can be expressed as a differential algebraic equation: dx / dt = f(x(t), u(t)), which describes how the state vector changes with the change of the decision vector; the decision vector and the state vector must always be within the safe and allowed range within the entire time domain [t _0 , t _f ].

[0143] According to one aspect of the present application, the physical stability in the stability index can also be solved by the following formula:

[0144] S _P =(β _1 ⋅se+β _2 ⋅zi+β _3 ⋅dsi)×10;

[0145] In the formula, β _1 , β _2 , β _3 are the influence degrees of the particle settling index, the Zeta potential index, and the dynamic suspension index in the physical stability, which can generally be 40%, 30%, 30%, or determined by experts; se is the solid particle settling coefficient, se = (1-u _t ) / u _max , in which u _max is the maximum allowable settling velocity, in m / s, and u _t is the target solid particle settling velocity, in m / s, which can be solved by the following formula u _t =(gd p 2 (ρ _p -ρ _f) ) / 18μ, in which ρ _p is the target mineral density, in kg / m3, and ρ_f d is the density of the liquid medium, generally in the density of water, unit kg / m3 p μ is the viscosity of the liquid medium, unit Pa·s, g is the acceleration of gravity, 9.81 m / s 2 ; zi is the Zeta potential coefficient (generally take the reference value 30) calculated according to the formula: zi=(|ζ|) / 30, wherein ζ is the average potential value of the target mineral in the PH=7 neutral solution; dsi is the dynamic suspension coefficient, calculated according to the formula: dsi=V _离心后 / V _悬浮 ×100%. Correspondingly, the physical stability index S _P The calculation formula can be further rewritten as:

[0146] S _P =(β _1 ⋅((1-u _t ) / u _max )+β _2 ⋅|ζ| / 30+β _3 ⋅dsi)×10。

[0147] In some embodiments, for the deep decoupling problem of medium design and system operation, the application provides a solution in the generation mechanism of reverse optimization and joint schedule. It is described how to construct a unified decision vector containing unit operation control variables and fluid property target control variables, and through a numerical case it is described how the optimizer simultaneously solves the guide vane opening trajectory and the target viscosity trajectory. The paradigm shift of changing the fluid property from a fixed boundary condition to an optimizable decision variable breaks down the barriers between design and operation, enabling deep coupling and collaborative optimization of the two.

[0148] Further, for the problem of lack of state observation, the application is based on a non-intrusive state observer of multi-modal perception. The technical details of the observer are analyzed in depth. It does not rely on any plug-in sensor, but analyzes the machine-side current harmonics and the main circuit pressure pulsation, two easily accessible external signals, combined with a machine learning model trained with strong and weak labels, to invert the apparent viscosity, solid volume fraction and other key states inside the fluid; provides real-time and accurate state visibility.

[0149] On this basis, based on the solution of the first two problems, the application changes the passive and inefficient reactive control mode. The dynamic operating envelope plans a safety domain for the system, while the collaborative execution and closed-loop learning enable the landing of the joint schedule and the continuous self-improvement of the system. The entire framework changes the control mode from a passive mode of responding to problems to an intelligent and proactive control mode of predicting the future, actively planning, collaborative execution, and closed-loop learning, so that the system always runs in an optimal state under varying operating conditions.

[0150] The preferred embodiments of the present application are described in detail above, but the present application is not limited to the specific details of the above-described embodiments. Within the technical concept of the present application, various equivalent transformations of the technical solutions of the present application can be made, and these equivalent transformations all belong to the protection scope of the present application.

Claims

1. A method for regulating high-density composite fluids in a pumped storage system, characterized in that, include: Based on the input basic data, a system objective function and constraint library are established. The basic data includes power grid service requirements, unit and pipeline parameter libraries, fluid candidate property libraries, and economic and maintenance parameters. The constraint library includes constraints on the machine stability index and shear-settlement coupled stability window. Based on the unit and pipeline parameter library in the basic data, a time-varying dynamic operating envelope of three control variables is generated; The sensor data stream is received, and the state vector is estimated online based on the dynamic operating envelope. The state vector includes the estimated value of apparent viscosity and the estimated value of solid volume fraction. The system receives the objective function, constraint library, dynamic operating envelope, and state vector. Under the constraints of the dynamic operating envelope, it performs inverse optimization to generate a joint schedule. The joint schedule includes both the unit operating trajectory and the fluid property objective trajectory. Among them, the machine stability index is used to quantify the overall risk when high-density composite fluid passes through a water pump turbine; it consists of at least four weighted sub-items, including: cavitation margin index, erosion life ratio, critical gap to particle size ratio, and mass settling risk index. The three control variables are: guide vane-rotation speed coordinated trajectory; bypass degassing setting; and bypass localized shear power. The process of generating a time-varying dynamic operating envelope of the three control variables includes: generating a degassing setting map and a shearing setting map; generating an initial guide vane-rotation speed envelope for the first second of mold change, taking into account the transient characteristics of the mixed-phase fluid; and synthesizing the degassing setting map, the shearing setting map, and the initial guide vane-rotation speed envelope for the first second of mold change to obtain the time-varying dynamic operating envelope of the three control variables.

2. The method according to claim 1, characterized in that, Backward optimization is performed by constructing and solving for a decision vector, which contains at least two types of variables: The first category is the unit operation control variables, which include the guide vane opening trajectory and the rotational speed trajectory; The second category consists of fluid property target control variables, which include the target apparent viscosity trajectory, the target solid volume fraction trajectory, and the target gas content trajectory.

3. The method according to claim 2, characterized in that, Perform reverse optimization, including: Based on the power grid service demand, the control cycle is decomposed into multiple service segments to form a service segment set; The optimization solution is based on the service segment set. At the boundary of adjacent service segments, cross-segment splicing constraints are applied to ensure that the continuity and rate of change of each trajectory in the decision vector meet the preset constraints.

4. The method according to claim 3, characterized in that, The system objective function includes multiple weights, and also includes: When performing optimization for any service segment, multiple weights are adaptively calculated based on the grid importance, equipment status, or prediction uncertainty of that service segment, forming a weight set; Apply the weight set to the system objective function.

5. The method according to any one of claims 1 or 4, characterized in that, The method also includes: Fixed online update rules for the cable wrapping; Based on the state vector, the online envelope update rule is applied to automatically adjust the boundary of the dynamic operating envelope of the time-varying three control variables.

6. The method according to claim 1, characterized in that, Online estimation yields a state vector containing fluid properties, specifically including: By analyzing the combined characteristics of machine-side current harmonics and main circuit pressure pulsations in the sensor data stream, the estimated values ​​of apparent viscosity and solid volume fraction are derived.

7. The method according to claim 6, characterized in that, The method also includes: Determine the model domain based on the operating conditions; Based on the identified model domain, either the gas-containing correction sub-model or the non-Newtonian principal model is selectively applied to perform the inversion.

8. The method according to claim 1, characterized in that, The method also includes execution control, specifically: The joint schedule is executed collaboratively by the main loop control unit and the bypass fluid control unit; The bypass fluid control unit schedules the shearing, degassing, and dosing equipment based on the principle of minimizing energy consumption.

Citation Information

Patent Citations

  • Day-ahead scheduling method for flexible direct-current power transmission system containing pumped storage and new energy power generation

    CN111525627A

  • Drainage basin water volume and water quality combined scheduling method based on constraint reinforcement learning

    CN116187208A