Power plant boiler water replenishing scheduling optimization method and system based on artificial intelligence

Abstract

The invention discloses a power plant boiler water replenishing scheduling optimization method and system based on artificial intelligence, and relates to the technical field of industrial process optimization control, multi-source operation data of a boiler side and a pump set side are collected and processed, a system operation state matrix is constructed, high-order dynamic characterization is performed on a steam drum pressure state on the basis, and a power plant boiler water replenishing scheduling optimization system is constructed. A false water level component caused by transient change of steam drum pressure is decoupled, and a real water level signal reflecting the real working medium stock in the boiler is reconstructed; modeling is carried out on operation data of a water feeding pump, and a dynamic power consumption characteristic model reflecting the change of the energy efficiency of a pump set along with time is constructed; and in the water replenishing scheduling control process, solving a minimum total power consumption load distribution scheme under multi-pump parallel operation, and generating an optimal control instruction. Under the working conditions of boiler load fluctuation and equipment performance change, collaborative optimization control over steam drum water level safety and water supply system operation energy consumption is achieved, and the safety, stability and economical efficiency of boiler water supplementing dispatching are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of industrial process optimization and control technology, specifically to an artificial intelligence-based method and system for optimizing boiler feedwater scheduling in power plants. Background Technology

[0002] The AI-based boiler feedwater scheduling optimization method and system for power plants refers to a class of methods that utilize artificial intelligence technology to analyze and model multi-source time-series data generated during boiler operation. Based on a full consideration of boiler thermodynamic characteristics, changes in operating status, and the performance characteristics of feedwater equipment, this method intelligently schedules and optimizes the boiler feedwater process. This type of method typically constructs a data-driven model or combines a physical model with a data model to comprehensively analyze the drum water level status, pressure change characteristics, and feedwater pump operating efficiency, thereby enabling rational decisions on feedwater flow and pump load allocation under complex operating conditions. By introducing artificial intelligence technology, boiler feedwater scheduling no longer relies solely on fixed rules or static parameters but can adaptively adjust according to real-time changes in operating status. Its role is to improve the accuracy and safety of drum water level control, reduce the overall energy consumption of the feedwater system, and enhance the economy and stability of power plant boiler operation.

[0003] However, existing optimization methods for boiler feedwater scheduling in power plants have technical problems such as relying solely on apparent water level or static rules for feedwater control, making it difficult to distinguish false water level changes caused by pressure transients, and ignoring the changes in feedwater pump energy efficiency characteristics over time, resulting in insufficient safety and high energy consumption in feedwater regulation.

[0004] Existing pressure state characterization methods often rely solely on a single variable, such as steam drum water level or pressure, making it difficult to accurately characterize the spurious water level effect caused by transient pressure changes. This can lead to misjudgments during start-up, shutdown, or rapid load changes.

[0005] Existing pump set characteristic modeling methods generally use fixed efficiency curves or offline calibration models, which are difficult to reflect the energy efficiency degradation of pump sets caused by factors such as wear and scaling during long-term operation, and are easily affected by abnormal samples.

[0006] Existing water replenishment optimization control methods often fail to incorporate the dynamic energy efficiency characteristics and pressure state risks of pump sets into a unified optimization framework, resulting in unreasonable load distribution and difficulty in reducing the overall energy consumption of the system when multiple pumps are running in parallel. Summary of the Invention

[0007] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an artificial intelligence-based method and system for optimizing boiler feedwater scheduling in power plants. The technical solution adopted by this invention is as follows: The artificial intelligence-based method for optimizing boiler feedwater scheduling in power plants provided by this invention includes the following steps:

[0008] Step S1: Data Acquisition;

[0009] Step S2: Pressure state characterization;

[0010] Step S3: Three-dimensional feature modeling of the pump unit;

[0011] Step S4: Boiler water makeup optimization control.

[0012] Further, in step S1, the data acquisition is used to construct a high signal-to-noise ratio system standard operating state. Specifically, it involves acquiring multi-source time-series data from the boiler side and the pump group side, and performing outlier removal, interpolation alignment, and smoothing filtering on the multi-source time-series data based on multivariate statistical criteria to obtain a system operating state matrix that is time-synchronized and retains higher-order derivative features.

[0013] The multi-source time-series data specifically includes steam drum pressure data characterizing the thermal state of the steam drum, steam drum water level data characterizing the change in the liquid phase in the steam drum, main steam pressure data characterizing the boiler steam output condition, load command data characterizing the boiler operating target, feedwater flow rate data characterizing the supply capacity of the feedwater system, pump outlet pressure data characterizing the hydraulic output state of the feedwater pump, and pump motor power data characterizing the energy consumption characteristics of the feedwater pump.

[0014] Further, in step S2, the pressure state characterization is used to decouple the transient false water level and the true mass water level of the steam drum. Specifically, based on the system operating state matrix, the first-order rate of change and second-order acceleration characteristics of the steam drum pressure data are calculated. Combined with the steam drum geometric parameters, steam-water ratio and historical operating conditions, a thermodynamic nonlinear expansion compensation model is constructed, which includes physical prior terms based on the thermal expansion of the steam drum metal wall, steam-water interface disturbance and steam bubble generation and coalescence process. Then, based on historical start-up and shutdown data, a small-scale neural network correction term is trained. The thermodynamic nonlinear expansion compensation model and the small-scale neural network correction term are superimposed to obtain an estimation function for calculating the transient false water level component, and the false water level component is calculated.

[0015] Based on this, the false water level component is removed from the actual measured water level in real time to obtain the true water level signal that represents the actual working fluid content inside the boiler. The amplitude and duration of the false water level component are normalized to obtain the false water level intensity index, which is then combined with the steam drum pressure and load change rate to obtain the pressure state feature vector.

[0016] The pressure state characterization specifically includes the following steps:

[0017] Step S21: Calculation of high-order dynamic features of pressure, used to construct the dynamic state features of steam drum pressure. Specifically, based on the system operating state matrix, the first-order rate of change and second-order acceleration features of steam drum pressure time series data are calculated to obtain the high-order dynamic features of pressure characterizing the rate of change and intensity of abrupt change of steam drum pressure.

[0018] Step S22: Modeling the source of false water level, used to estimate the physical false water level component caused by transient changes in steam drum pressure. Specifically, by combining the steam drum geometric parameters, steam-water ratio and the higher-order dynamic characteristics of the pressure, a thermodynamic nonlinear expansion compensation model is constructed, including the thermal expansion of the steam drum metal wall, the disturbance of the steam-water interface and the process of steam bubble generation and coalescence, to obtain the false water level estimation component based on physical priors.

[0019] Step S23: Construct a small-scale neural network correction term to correct nonlinear false water level errors that are difficult to cover by the physical model. Specifically, a small-scale neural network is trained using historical start-up and shutdown data, with steam drum pressure, high-order dynamic features of pressure, and load change information as inputs, and the corresponding false water level correction component is output.

[0020] Step S24: False water level component estimation, used to obtain the comprehensive estimation result of the transient false water level of the steam drum. Specifically, the false water level component output by the thermodynamic nonlinear expansion compensation model is superimposed with the small-scale neural network correction component to obtain the transient false water level component of the steam drum.

[0021] Step S25: Reconstruction of the true mass water level signal, used to reconstruct the true water level signal representing the true working fluid inventory inside the boiler. Specifically, the transient false water level component of the steam drum is removed in real time from the actual measured water level to obtain the true mass water level signal.

[0022] Step S26: Pressure state vector construction, used to construct a pressure state feature vector for water replenishment scheduling control. Specifically, the amplitude and duration of the transient false water level component of the steam drum are normalized to form a false water level intensity index, and the false water level intensity index is combined with the steam drum pressure and load change rate to obtain the pressure state feature vector.

[0023] Further, in step S3, the three-dimensional feature modeling of the pump group is used to construct the time-varying energy efficiency characteristics of the pump group reflecting the aging characteristics of the equipment. Specifically, based on the system operating state matrix, the operating data samples of the feedwater pumps under stable operating conditions are screened using a sliding time window. A weighted least squares three-dimensional surface fitting method for flow rate, head, and power data with the introduction of a time forgetting factor is used to perform three-dimensional feature modeling of the pump group, obtaining the current dynamic power consumption characteristic model of each feedwater pump. This includes the following steps:

[0024] Step S31: Steady-state sample screening, used to construct an effective sample set for pump group three-dimensional feature modeling. Specifically, based on the system operation state matrix, the flow rate, head and power data of the water pump are judged for stable operating conditions within the sliding time window, and steady-state operation data samples that meet the condition that the fluctuations of flow rate, head and power are all less than the preset threshold are screened.

[0025] Step S32: Construction of time forgetting factor weights, which is used to introduce a time weighting mechanism for the impact of equipment aging. Specifically, time forgetting factor weights are constructed based on the timestamps of steady-state operation data samples, so that data samples closer to the current time have a higher weight in the modeling process, resulting in a weighted sample set of data that reflects the changes in pump group characteristics over time.

[0026] Step S33: Abnormal pump group optimization, used to suppress the interference of abnormal samples on the modeling results of pump group characteristics. Specifically, it calculates the sample power residual based on the existing power consumption characteristic model, and constructs a robust weight function based on the residual amplitude. It reduces the weight of abnormal deviation samples in the modeling and obtains a comprehensive weighted sample that integrates time forgetting weight and robust weight.

[0027] Step S34: Constructing the three-dimensional power consumption of the pump set, which is used to construct a three-dimensional power consumption surface model structure of the pump set that conforms to the physical characteristics. Specifically, a polynomial basis function with flow rate and head as independent variables is selected to establish a three-dimensional surface expression of pump power with respect to flow rate and head. The physical constraint condition that the power does not decrease monotonically with flow rate and head is introduced to limit the feasible range of model parameters.

[0028] Step S35: Power consumption feature fitting, used to solve the parameters of the time-varying power consumption feature model of the pump group. Specifically, under the conditions of the comprehensive weighted sample set and physical constraints, the weighted least squares method with the introduction of time forgetting factor is used to solve the parameters of the three-dimensional power consumption surface model to obtain the dynamic power consumption feature model of each water pump at the current moment.

[0029] Step S36: Pump group feature modeling, used to extract the pump group operating domain and energy efficiency degradation features. Specifically, based on the dynamic power consumption feature model, the effective operating domain boundary of the water pump is determined, and the relative power consumption change index under standard operating conditions is calculated to obtain time-varying energy efficiency features used to characterize the degree of energy efficiency degradation of the pump group.

[0030] Furthermore, in step S4, the boiler feedwater optimization control is used to achieve coordinated control of water level safety and operating energy consumption. Specifically, based on the dynamic power consumption characteristic model and the pressure state characteristic vector, the total feedwater flow demand is calculated in combination with the actual water level signal. The total feedwater flow demand is used as a constraint condition. Based on the improved dynamic power consumption function model, the minimum total power consumption load allocation ratio under the parallel operation of multiple pumps is solved to obtain the optimal actuator control command for each feedwater pump.

[0031] The artificial intelligence-based boiler water supply scheduling optimization system for power plants provided by this invention includes a data acquisition module, a pressure status characterization module, a pump group feature modeling module, and a water supply optimization control module.

[0032] The data acquisition module is used for data acquisition. Through data acquisition, it obtains the system operating status matrix and sends the system operating status matrix to the pressure status characterization module and the pump group feature modeling module.

[0033] The pressure state characterization module is used for pressure state characterization. Through pressure state characterization, a pressure state feature vector is obtained, and the pressure state feature vector is sent to the water replenishment optimization control module.

[0034] The pump set feature modeling module is used for three-dimensional feature modeling of the pump set. Through three-dimensional feature modeling of the pump set, a dynamic power consumption feature model is obtained, and the dynamic power consumption feature model is sent to the water replenishment optimization control module.

[0035] The water replenishment optimization control module is used for boiler water replenishment optimization control. Through boiler water replenishment optimization control, the optimal actuator control command is obtained.

[0036] The beneficial effects achieved by the present invention using the above solution are as follows:

[0037] (1) In view of the technical problems in the existing boiler water supply scheduling optimization methods for power plants, which are based solely on apparent water level or static rules for water supply control, it is difficult to distinguish false water level changes caused by pressure transients and ignore the time-varying energy efficiency characteristics of feedwater pumps, resulting in insufficient safety of water supply regulation and high energy consumption. This solution creatively adopts a comprehensive boiler water supply scheduling optimization method for power plants that combines pressure state characterization, pump group characteristic modeling and water supply optimization control. In the water supply scheduling decision-making process, the actual quality water level, pressure state risk information and pump group time-varying energy efficiency model are introduced simultaneously, realizing coordinated water supply scheduling control that takes into account both the safety of steam drum water level and the optimization of system operation energy consumption under boiler load fluctuation and equipment performance change conditions.

[0038] (2) In view of the technical problem that existing pressure state characterization methods usually only use the single variable of steam drum water level or pressure for judgment, it is difficult to accurately characterize the false water level effect caused by transient pressure changes, and it is easy to make misjudgments under start-up or rapid load change conditions. This solution creatively adopts a steam drum false water level state characterization method based on high-order pressure characteristics. By introducing high-order dynamic characteristics such as the first-order rate of change and second-order acceleration of steam drum pressure, combined with the steam drum thermodynamic physical prior model and small-scale neural network correction mechanism, the dynamic estimation and real-time elimination of false water level components are realized, thereby accurately reconstructing the real water level signal that reflects the real working fluid inventory inside the boiler, and improving the accuracy and robustness of water level state identification.

[0039] (3) In view of the technical problems in the existing pump group characteristic modeling methods, which generally use fixed efficiency curves or offline calibration models, it is difficult to reflect the energy efficiency degradation caused by wear, scaling and other factors during long-term operation of the pump group, and it is easily affected by abnormal samples. This solution creatively adopts the pump group characteristic modeling method based on the forgetting factor robust constraint surface fitting. By introducing the time forgetting factor and robust weight mechanism in the three-dimensional flow-head-power space, the adaptive modeling of the pump group power consumption characteristics changing with time is realized. Under the premise of satisfying the physical monotonicity constraint, the influence of abnormal samples on the model parameters is effectively suppressed, thereby realizing the accurate characterization of the time-varying energy efficiency characteristics of the pump group.

[0040] (4) In view of the technical problem that the existing water replenishment optimization control methods often fail to incorporate the dynamic energy efficiency characteristics and pressure state risks of pump sets into a unified optimization framework, resulting in unreasonable load distribution and difficulty in reducing the overall energy consumption of the system when multiple pumps are running in parallel, this solution creatively adopts an improved dynamic power consumption function model to solve the minimum total power consumption load distribution ratio under the operation of multiple pumps in parallel. Under the premise of meeting the total water supply flow requirements and operation safety constraints, the solution comprehensively considers the current dynamic power consumption characteristic model and operation constraints of each water supply pump to obtain the load distribution ratio corresponding to the minimum total system power consumption, thus realizing adaptive energy consumption optimization control under different pump set energy efficiency levels and operating conditions. Attached Figure Description

[0041] Figure 1 A flowchart illustrating the AI-based boiler feedwater scheduling optimization method for power plants provided by this invention.

[0042] Figure 2 A schematic diagram of the power plant boiler water supply scheduling optimization system based on artificial intelligence provided by the present invention;

[0043] Figure 3 This is a schematic diagram of the process for characterizing the pressure state in step S2;

[0044] Figure 4A schematic diagram of the process for modeling the three-dimensional features of the pump unit in step S3.

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0047] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0048] Example 1, see Figure 1 The present invention provides an artificial intelligence-based optimization method for boiler feedwater scheduling in power plants, which includes the following steps:

[0049] Step S1: Data Acquisition;

[0050] Step S2: Pressure state characterization;

[0051] Step S3: Three-dimensional feature modeling of the pump unit;

[0052] Step S4: Boiler water makeup optimization control.

[0053] By performing the above operations, this solution addresses the technical problems in existing power plant boiler feedwater scheduling optimization methods. These methods rely solely on apparent water levels or static rules for feedwater control, making it difficult to distinguish false water level changes caused by pressure transients and neglecting the time-varying energy efficiency characteristics of feedwater pumps, resulting in insufficient safety and high energy consumption in feedwater regulation. This solution creatively adopts a comprehensive power plant boiler feedwater scheduling optimization method that combines pressure state characterization, pump set characteristic modeling, and feedwater optimization control. In the feedwater scheduling decision-making process, it simultaneously introduces real quality water level, pressure state risk information, and pump set time-varying energy efficiency models, achieving coordinated feedwater scheduling control that balances boiler load fluctuations and equipment performance changes while considering the safety of the steam drum water level and the optimization of system operating energy consumption.

[0054] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the data acquisition is used to construct a high signal-to-noise ratio system standard operating state. Specifically, it involves acquiring multi-source time-series data from the boiler side and the pump group side. Based on multivariate statistical criteria, outlier removal, interpolation alignment, and smoothing filtering are performed on the multi-source time-series data to obtain a system operating state matrix that is time-synchronized and retains higher-order derivative features.

[0055] The multi-source time-series data specifically includes steam drum pressure data characterizing the thermal state of the steam drum, steam drum water level data characterizing the change in the liquid phase in the steam drum, main steam pressure data characterizing the boiler steam output condition, load command data characterizing the boiler operating target, feedwater flow rate data characterizing the supply capacity of the feedwater system, pump outlet pressure data characterizing the hydraulic output state of the feedwater pump, and pump motor power data characterizing the energy consumption characteristics of the feedwater pump.

[0056] Preferably, to ensure that data from different sources and with different sampling periods can be effectively correlated within the same analytical framework, the multi-source time-series data is first subjected to time consistency preprocessing. Specifically, using the timestamp of the boiler main control system as a unified time reference, data sequences with different sampling frequencies are resampled and interpolated for alignment, so that the drum pressure, drum water level, load command, and pump-side operating data form a synchronized state vector on a unified time axis. Linear interpolation is used for continuously changing data, while a hold-type interpolation is used for step or piecewise constant data to avoid introducing non-physical abrupt changes.

[0057] More preferably, to eliminate the impact of measurement noise, communication jitter, and occasional anomalies on system state identification, outlier detection and removal are performed on the aligned multi-source time-series data based on multivariate statistical criteria. Specifically, this includes calculating the mean, variance, and covariance matrix of each variable within a sliding time window, and using Mahalanobis distance to determine the consistency of multidimensional state samples. When a multidimensional state sample at a certain moment deviates significantly from the historical statistical distribution and does not meet the physical constraints of the boiler feedwater system, the sample is identified as an outlier and replaced by the interpolation result of adjacent valid samples.

[0058] After outlier handling, key pressure and flow signals are smoothed and filtered to suppress high-frequency random noise while retaining high-order variation features for dynamic analysis. Specifically, for steam drum pressure and pump outlet pressure signals, the Savitsky-Gore filtering method is used to smooth the original signals by polynomial fitting within a preset time window, so that the pressure signals can still be stably calculated for first-order rate of change and second-order acceleration features after filtering.

[0059] For the water supply flow and pump motor power signals, a low-pass filter is used to suppress high-frequency disturbances caused by sensor jitter, thereby ensuring the stability of the pump set energy efficiency characteristic analysis;

[0060] Through the above-mentioned time alignment, anomaly removal, and feature preservation smoothing processes, a system operating state matrix composed of multiple physical quantities is constructed. The state vector at each moment includes at least pressure features, water level features, load features, and pump operation features. Furthermore, the state matrix maintains continuity in the time dimension and physical consistency in the feature dimension, thereby providing a reliable data foundation for subsequent decoupling modeling of spurious water levels based on higher-order pressure features and modeling of time-varying energy efficiency characteristics of pump units.

[0061] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the pressure state characterization is used to decouple the transient false water level and the true mass water level of the steam drum. Specifically, based on the system operating state matrix, the first-order rate of change and second-order acceleration characteristics of the steam drum pressure data are calculated. Combined with the steam drum geometric parameters, steam-water ratio and historical operating conditions, a thermodynamic nonlinear expansion compensation model is constructed, which includes physical prior terms based on the thermal expansion of the steam drum metal wall, steam-water interface disturbance and steam bubble generation and coalescence process. Then, based on historical start-stop data, a small-scale neural network correction term is trained. The thermodynamic nonlinear expansion compensation model and the small-scale neural network correction term are superimposed to obtain an estimation function for calculating the transient false water level component, and the false water level component is calculated.

[0062] Based on this, the false water level component is removed from the actual measured water level in real time to obtain the true water level signal that represents the actual working fluid content inside the boiler. The amplitude and duration of the false water level component are normalized to obtain the false water level intensity index, which is then combined with the steam drum pressure and load change rate to obtain the pressure state feature vector.

[0063] The pressure state characterization specifically includes the following steps:

[0064] Step S21: Calculation of high-order dynamic features of pressure, used to construct the dynamic state features of steam drum pressure. Specifically, based on the system operating state matrix, the first-order rate of change and second-order acceleration features of steam drum pressure time series data are calculated to obtain the high-order dynamic features of pressure characterizing the rate of change and intensity of abrupt change of steam drum pressure.

[0065] The formula for calculating the higher-order dynamic characteristics of the pressure is as follows:

[0066] ;

[0067] In the formula, P(t) is the first-order rate of change of the steam drum pressure time series data, used to characterize the rate of change of steam drum pressure. P(t) represents the steam drum pressure time series data, and t is the time index. It is the sampling time interval. It is the second-order acceleration feature of the steam drum pressure time series data, used to characterize the intensity of abrupt changes in steam drum pressure;

[0068] Step S22: Modeling the source of false water level, used to estimate the physical false water level component caused by transient changes in steam drum pressure. Specifically, by combining the steam drum geometric parameters, steam-water ratio and the higher-order dynamic characteristics of the pressure, a thermodynamic nonlinear expansion compensation model is constructed, including the thermal expansion of the steam drum metal wall, the disturbance of the steam-water interface and the process of steam bubble generation and coalescence, to obtain the false water level estimation component based on physical priors.

[0069] The thermal expansion of the steam drum metal wall is used to model the equivalent volume change caused by the thermo-mechanical coupling deformation of the steam drum metal wall when the pressure changes rapidly. The calculation formula is as follows:

[0070] ;

[0071] In the formula, This is the spurious water level estimation component caused by the thermal expansion effect of the steam drum metal wall, where k1 is the thermal expansion coefficient of the steam drum metal wall, and V drum It is the equivalent volume of the steam drum, and T is the temperature parameter;

[0072] The gas-water interface disturbance is used to model the apparent liquid level movement caused by the change in density difference at the gas-water interface due to sudden pressure changes. The calculation formula is as follows:

[0073] ;

[0074] In the formula, This is the spurious water level estimation component caused by the gas-water interface disturbance effect, where k2 is the gas-water interface disturbance coefficient. It is the density of the liquid working fluid inside the steam drum. It is the density of the vapor phase working fluid inside the steam drum;

[0075] The bubble formation and coalescence process is used to model the artificially high or low liquid level caused by changes in bubble volume fraction during pressurization or depressurization. The calculation formula is as follows:

[0076] ;

[0077] In the formula, The component representing the spurious water level estimate is the bubble generation and coalescence process, and k3 is the bubble dynamics influence coefficient. It is a nonlinear response function of bubble dynamics, used to comprehensively describe the relationship between the first and second rates of change of steam drum pressure on the change of bubble volume fraction.

[0078] The calculation formula for the thermodynamic nonlinear expansion compensation model is as follows:

[0079] ;

[0080] In the formula, It is the spurious water level estimate component based on physical priors output by the thermodynamic nonlinear expansion compensation model;

[0081] Step S23: Construct a small-scale neural network correction term to correct nonlinear false water level errors that are difficult to cover by the physical model. Specifically, a small-scale neural network is trained using historical start-up and shutdown data, with steam drum pressure, high-order dynamic features of pressure, and load change information as inputs, and the corresponding false water level correction component is output.

[0082] Preferably, the small-scale neural network specifically includes an input layer, a hidden layer, and an output layer. The input layer is a four-dimensional input layer that receives data input including steam drum pressure, high-order dynamic features of pressure, and load change information. The hidden layer includes a first hidden layer and a second hidden layer. The first hidden layer has 8 neurons activated by the ReLU function, and the second hidden layer has 4 neurons activated by the ReLU function. The output layer has one neuron that outputs a linear output and a spurious water level correction component.

[0083] More preferably, the small-scale neural network minimizes the following loss function by utilizing historical start-stop data and corresponding water level measurement deviations:

[0084] ;

[0085] In the formula, H is the loss function for a small-scale neural network. meas (t) is the actual measured value of the water level in the steam drum, H ref (t) is the reference mass water level value under stable operating conditions, used to represent the baseline level of the actual working fluid inventory inside the boiler. It is the spurious water level correction component output by a small-scale neural network.

[0086] Step S24: False water level component estimation, used to obtain the comprehensive estimation result of the transient false water level of the steam drum. Specifically, the false water level component output by the thermodynamic nonlinear expansion compensation model is superimposed with the small-scale neural network correction component to obtain the transient false water level component of the steam drum.

[0087] The formula for calculating the transient spurious water level component of the steam drum is as follows:

[0088] ;

[0089] In the formula, It is the transient spurious water level component of the steam drum obtained from the physical model and neural network correction results, which is used to uniformly characterize the non-real components in water level measurement under dynamic pressure change conditions;

[0090] Step S25: Reconstruction of the true mass water level signal, used to reconstruct the true water level signal representing the true working fluid inventory inside the boiler. Specifically, the transient false water level component of the steam drum is removed in real time from the actual measured water level to obtain the true mass water level signal.

[0091] The formula for calculating the actual mass water level signal is as follows:

[0092] ;

[0093] In the formula, H true (t) is the true mass water level signal obtained after removing false water level components. It is used to characterize the change in the true working fluid inventory inside the boiler and serves as the core input for subsequent water replenishment scheduling control.

[0094] Step S26: Pressure state vector construction, used to construct a pressure state feature vector for water replenishment scheduling control. Specifically, the amplitude and duration of the transient false water level component of the steam drum are normalized to form a false water level intensity index, and the false water level intensity index is combined with the steam drum pressure and load change rate to obtain the pressure state feature vector.

[0095] Preferably, the formula for calculating the false water level intensity index is:

[0096] ;

[0097] In the formula, I false (t) is a false water level intensity index, used to comprehensively characterize the magnitude and duration of the false water level component. It provides guidance for the adjustment range and safety constraints of water replenishment scheduling strategies. H nom This refers to the rated water level range of the steam drum. It is the false water level duration factor, which reflects the duration of the false water level component in the time dimension. It can be calculated based on the length of time that the false water level continuously exceeds the threshold.

[0098] The formula for calculating the pressure state feature vector is as follows:

[0099] ;

[0100] In the formula, S p (t) is the pressure state feature vector.

[0101] By performing the above operations, this solution addresses the technical problem in existing pressure state characterization methods that typically rely solely on a single variable such as drum water level or pressure, making it difficult to accurately characterize the spurious water level effect caused by transient pressure changes and prone to misjudgment during start-up, shutdown, or rapid load changes. This solution creatively adopts a drum spurious water level characterization method based on higher-order pressure characteristics. By introducing higher-order dynamic features such as the first-order rate of change and second-order acceleration of drum pressure, combined with a priori thermodynamic and physical model of the drum and a small-scale neural network correction mechanism, it achieves dynamic estimation and real-time removal of spurious water level components. This enables accurate reconstruction of the true water level signal reflecting the actual working fluid quantity inside the boiler, improving the accuracy and robustness of water level state identification.

[0102] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the three-dimensional feature modeling of the pump group is used to construct the time-varying energy efficiency characteristics of the pump group that reflect the aging characteristics of the equipment. Specifically, based on the system operating state matrix, the operating data samples of the feedwater pump under stable operating conditions are screened using a sliding time window. A weighted least squares three-dimensional surface fitting method for flow rate, head, and power data with the introduction of a time forgetting factor is used to perform three-dimensional feature modeling of the pump group, thereby obtaining the current dynamic power consumption characteristic model of each feedwater pump. The steps include:

[0103] Step S31: Steady-state sample screening, used to construct an effective sample set for pump group three-dimensional feature modeling. Specifically, based on the system operation state matrix, the flow rate, head and power data of the water pump are judged for stable operating conditions within the sliding time window, and steady-state operation data samples that meet the condition that the fluctuations of flow rate, head and power are all less than the preset threshold are screened.

[0104] Preferably, the steady-state consistency judgment is achieved through a multivariate statistical consistency criterion; specifically, the mean and standard deviation of the feedwater pump flow rate, head, and power data are calculated respectively within the sliding time window, and steady-state judgment is performed based on the following conditions: within the preset time window, when the relative fluctuation rates of flow rate, head, and power are all less than the corresponding thresholds, and the absolute values ​​of their first-order rates of change are all less than the rate of change thresholds, the operating data within the time window is determined to be a steady-state operating data sample;

[0105] Among them, relative volatility is used to characterize the stability of the operating quantity in a statistical sense, and first-order rate of change is used to exclude non-steady-state transient conditions such as start-up and shutdown, valve impact, etc., so as to ensure that the data samples entering subsequent modeling are identifiable and repeatable.

[0106] Step S32: Construction of time forgetting factor weights, which is used to introduce a time weighting mechanism for the impact of equipment aging. Specifically, time forgetting factor weights are constructed based on the timestamps of steady-state operation data samples, so that data samples closer to the current time have a higher weight in the modeling process, resulting in a weighted sample set of data that reflects the changes in pump group characteristics over time.

[0107] Preferably, in step S32, the time forgetting factor weight is constructed using an exponential decay method. Specifically, for any steady-state operating data sample, a weight coefficient that monotonically decreases with increasing time interval is constructed based on the time interval between the timestamp of the sample and the current modeling time, so that recent samples have a higher weight in the model fitting process, while the influence of historical samples gradually weakens. By introducing the time forgetting factor weight, the three-dimensional power consumption characteristic model of the pump group can adaptively migrate with aging factors such as equipment wear and scaling, thereby avoiding the distortion of energy efficiency assessment caused by using fixed model parameters.

[0108] Step S33: Abnormal pump group optimization, used to suppress the interference of abnormal samples on the modeling results of pump group characteristics. Specifically, it calculates the sample power residual based on the existing power consumption characteristic model, and constructs a robust weight function based on the residual amplitude. It reduces the weight of abnormal deviation samples in the modeling and obtains a comprehensive weighted sample that integrates time forgetting weight and robust weight.

[0109] The robust weighting function is constructed based on the power residual. Specifically, it uses the pump group power consumption characteristic model obtained in the previous time step or the previous sliding time window to predict the power value of the current steady-state sample and calculate the residual between the predicted power and the actual measured power. When the absolute value of the residual exceeds a preset threshold, the sample is determined to be an abnormal deviation sample and its weight in the weighted least squares fitting process is reduced.

[0110] By superimposing robust weights with time-forgetting factor weights, it is possible to maintain the model's sensitivity to recent valid data while effectively suppressing the interference of abnormal samples introduced by sensor noise, transient disturbances, or local abnormal conditions on the modeling results.

[0111] Step S34: Constructing the three-dimensional power consumption of the pump set, which is used to construct a three-dimensional power consumption surface model structure of the pump set that conforms to the physical characteristics. Specifically, a polynomial basis function with flow rate and head as independent variables is selected to establish a three-dimensional surface expression of pump power with respect to flow rate and head. The physical constraint condition that the power does not decrease monotonically with flow rate and head is introduced to limit the feasible range of model parameters.

[0112] The three-dimensional power consumption surface model of the pump unit is constructed using low-order polynomial basis functions, with flow rate and head as independent variables and power as the dependent variable, forming a continuously differentiable three-dimensional surface expression. As a further optimization of this embodiment, physical consistency constraints are introduced during the model construction process to ensure that the relationship between power and flow rate and head satisfies the monotonically non-decreasing characteristic, thereby avoiding anomalous surface shapes that violate the basic physical mechanism of the pump unit during the fitting process. By applying the physical constraints to the model parameters, the stability and engineering reliability of the power consumption characteristic model under extrapolation conditions can be effectively improved.

[0113] The basic form of the continuously differentiable three-dimensional surface representation is as follows:

[0114] ;

[0115] In the formula, P(Q,H) represents the power output value of the feedwater pump under given flow rate Q and head H conditions. It is the output of the three-dimensional power consumption surface model of the pump unit, used to characterize the operating energy consumption characteristics of the feedwater pump. These are the constant parameters of the power consumption surface model. and These are the coefficient parameters of the first-order term, used to characterize the linear response characteristics of power as a function of flow rate and head. , and These are quadratic coefficient parameters, used successively to characterize the nonlinear characteristics of power changing with the square of flow rate, the influence of the coupling effect between flow rate and head on power, and the nonlinear characteristics of power changing with the square of head.

[0116] The specific calculation formula for the physical consistency constraint is as follows:

[0117] ;

[0118] In the formula, It is the partial derivative of power with respect to flow rate, used to describe the instantaneous effect of flow rate changes on pump power under constant head conditions. It is the partial derivative of power with respect to head, used to describe the instantaneous effect of head changes on pump power under constant flow conditions;

[0119] Step S35: Power consumption feature fitting, used to solve the parameters of the time-varying power consumption feature model of the pump group. Specifically, under the conditions of the comprehensive weighted sample set and physical constraints, the weighted least squares method with the introduction of time forgetting factor is used to solve the parameters of the three-dimensional power consumption surface model to obtain the dynamic power consumption feature model of each water pump at the current moment.

[0120] Preferably, the solution of the pump group time-varying power consumption characteristic model parameters is based on a weighted least squares optimization framework, where the sample weights are composed of time forgetting factor weights and robust weights. By minimizing the weighted residual sum of squares, the model parameters are solved under the premise of satisfying physical constraints, so that the dynamic power consumption characteristic model has good numerical stability and online update capability while ensuring fitting accuracy. This solution method is suitable for repeated modeling processes under sliding time window conditions, and can periodically update the pump group power consumption characteristic model to reflect the trend of equipment performance changes over time.

[0121] More preferably, the objective function can be constructed based on the weighted least squares optimization framework:

[0122] ;

[0123] In the formula, These are the parameter vectors of the three-dimensional power consumption surface model of the pump unit, derived from... to Composition, where N is the number of steady-state operating data samples, i is the index of the steady-state operating data sample, and w t This is the time forgetting factor weight, used to characterize the temporality of the i-th steady-state data sample relative to the current modeling moment, giving more recent samples a higher weight in the model parameter solving process. w is the time interval between the acquisition time of the i-th steady-state operating data sample and the current modeling time. r It is a robust weighting function used to adjust sample weights based on the magnitude of sample residuals, suppressing the interference of outliers on the model fitting results. i P is the power residual of the i-th steady-state operating data sample. i It is the actual measured power value corresponding to the i-th steady-state operating data sample. It is the transpose of the parameter vector of the three-dimensional power consumption surface model of the pump unit. It is a polynomial basis function vector used to map flow rate and head to the feature space of the power consumption surface model. Its elements consist of constant terms, first-order terms and second-order terms.

[0124] By solving the above objective function, the time-varying update of the pump group power consumption characteristic parameters can be achieved while ensuring numerical stability.

[0125] Step S36: Pump group feature modeling, used to extract the pump group operating domain and energy efficiency degradation features. Specifically, based on the dynamic power consumption feature model, the effective operating domain boundary of the water pump is determined, and the relative power consumption change index under standard operating conditions is calculated to obtain the time-varying energy efficiency features used to characterize the degree of energy efficiency degradation of the pump group.

[0126] Preferably, in step S36, the effective operating domain boundary is determined based on the distribution range of steady-state operating data samples in the dimensions of flow rate and head, which is used to limit the effective application range of the pump group's three-dimensional power consumption model;

[0127] More preferably, by selecting a standard operating point or a representative operating point, the model power consumption output at the current time and the reference time are compared, and the relative power consumption change index is calculated as a time-varying energy efficiency characteristic characterizing the degree of energy efficiency degradation of the pump group; the time-varying energy efficiency characteristic can be used to prioritize or constrain the load of different feedwater pumps in the subsequent boiler feedwater optimization control process, thereby achieving coordinated optimization of energy efficiency and equipment life at the system level.

[0128] The relative power consumption change index can be defined as an energy efficiency degradation index as follows:

[0129] ;

[0130] In the formula, I deg (t) is the energy efficiency degradation index, which is used to quantitatively characterize the degree of power consumption change of the water pump at time t relative to the reference time, reflecting the degradation trend of pump energy efficiency over time. Q0 is the flow rate at the standard operating point, H0 is the head at the standard operating point, t0 is the reference time, which is used as a reference time point for energy efficiency degradation assessment, and P(·) is the power consumption characteristic model identifier function.

[0131] The above indicators can be used to quantitatively characterize the trend of pump group energy efficiency changes over time, and provide constraints or weight information for subsequent water replenishment scheduling optimization.

[0132] By performing the above operations, this solution addresses the technical problems of existing pump set characteristic modeling methods, which commonly use fixed efficiency curves or offline calibration models, making it difficult to reflect the energy efficiency degradation of pump sets due to factors such as wear and scaling during long-term operation, and are easily affected by abnormal samples. This solution creatively adopts a pump set characteristic modeling method based on robust constraint surface fitting with a forgetting factor. By introducing a time forgetting factor and a robust weighting mechanism into the three-dimensional flow-head-power space, it achieves adaptive modeling of the pump set's power consumption characteristics as a function of time. Under the premise of satisfying the physical monotonicity constraint, it effectively suppresses the influence of abnormal samples on the model parameters, thereby achieving an accurate characterization of the pump set's time-varying energy efficiency characteristics.

[0133] Example 5, see Figure 1 , Figure 2This embodiment is based on the above embodiment. In step S4, the boiler feedwater optimization control is used to achieve coordinated control of water level safety and operating energy consumption. Specifically, based on the dynamic power consumption characteristic model and the pressure state characteristic vector, the total feedwater flow demand is calculated in combination with the real water level signal. The total feedwater flow demand is used as a constraint condition. Based on the improved dynamic power consumption function model, the minimum total power consumption load allocation ratio under the parallel operation of multiple pumps is solved to obtain the optimal actuator control command for each feedwater pump.

[0134] As a further optimization of this embodiment, in this embodiment, the boiler feedwater optimization control in step S4 is used to achieve coordinated control of the boiler drum water level safety and the energy consumption of the feedwater system under the conditions of boiler load fluctuation and feedwater equipment characteristics changing over time.

[0135] The control system first receives the real water level signal and pressure state feature vector output in step S2. The real water level signal is used to characterize the change in the actual working fluid inventory inside the boiler, and the pressure state feature vector is used to reflect the pressure change in the steam drum and the degree of risk of false water level under the current operating conditions. At the same time, the control system obtains the dynamic power consumption feature model of each feedwater pump constructed in step S3, which is used to characterize the correspondence between flow rate, head and power consumption of different feedwater pumps at the current moment.

[0136] Based on this, the control system calculates the total water supply flow requirement at the current moment by combining the deviation between the actual water level signal and the preset target water level with the load command information.

[0137] To suppress spurious water level interference caused by transient pressure changes, a spurious water level intensity index from the pressure state feature vector is further introduced to adaptively correct the adjustment range of the total water supply flow demand. Specifically, an adaptive suppression gain function is constructed that monotonically decreases as the spurious water level intensity index increases, and the calculation formula is as follows:

[0138] ;

[0139] In the formula, It is an adaptive suppression gain function. It is the suppression intensity coefficient. It is a sensitivity index.

[0140] Therefore, the revised formula for calculating the total water supply flow demand is as follows:

[0141] ;

[0142] In the formula, Q req (t) is the corrected total water supply flow demand, Q steam(t) is the main steam flow rate, which serves as a feedforward control signal to characterize the current steam load output of the boiler. K p K is the proportional gain of the PID controller. i K is the integral coefficient of the PID controller. d These are the derivative coefficients of the PID controller, and e(t) is the water level deviation signal, specifically the difference between the preset steam drum water level setpoint and the actual mass water level output in step S2. It is the integral term of water level deviation over time. It is the differential term of the water level deviation;

[0143] This makes the change in total water supply flow more gradual when the risk of false water levels is high, thereby avoiding excessive water replenishment or pumping due to transient measurement distortion.

[0144] After obtaining the total water supply flow demand, it is used as a constraint for the parallel operation of multiple pumps to construct a load allocation model with the goal of minimizing the total operating power consumption of the water supply system.

[0145] The load allocation model uses the dynamic power consumption characteristic model of each water supply pump as the basis for energy consumption calculation, and constructs an energy consumption objective function:

[0146] ;

[0147] In the formula, J is the total operating power consumption of the water supply system, which serves as the optimization objective function of the load allocation model. This refers to the number of water pumps currently in operation. It is the index number of the water pump. It is the first Dynamic power consumption characteristic model of the water pump. It is the first The distribution flow rate of the water pump, H sys This is the total head required by the system;

[0148] Simultaneously, considering the available operating range, operational stability constraints, and flow rate change limits of each water supply pump at the current moment, a multi-pump collaborative optimization problem is formed that meets the requirements of safety and stability. Among them, the constraints include: the sum of the flow rates of each pump is equal to the total water supply flow demand, and the flow rate allocated to each pump must be within its efficient operating range, and the flow rate change rate between adjacent moments is limited to prevent hydraulic shock.

[0149] The control system solves the load distribution model. Given the nonlinear characteristics of the objective function and constraints, it is preferable to use the sequential quadratic programming method (SQP) or the interior point method for iterative optimization to obtain the optimal water flow distribution result for each water pump under the premise of meeting the total water flow demand.

[0150] This allocation result can automatically distribute more load to the operating range of pump sets with higher energy efficiency when the energy efficiency characteristics of different water supply pumps vary and change over time, thereby reducing the overall operating power consumption.

[0151] Finally, the control system generates corresponding actuator control commands based on the optimal water flow distribution results of each water pump.

[0152] Specifically, for feedwater pumps using variable frequency speed control, the optimal feedwater flow rate is converted into a corresponding target speed control command; for feedwater pumps using valve regulation, the optimal feedwater flow rate is converted into a corresponding valve opening control command; the control system sends the above control commands to each feedwater pump actuator to complete the closed-loop optimization control of the boiler water replenishment process.

[0153] Through the above implementation methods, this embodiment achieves energy consumption optimization for multi-pump parallel operation while ensuring the safety and stability of steam drum water level regulation. It can also adaptively track the characteristics of feedwater pump performance changes over time, effectively reducing the overall operating energy consumption of the boiler feedwater system.

[0154] By performing the above operations, this solution addresses the technical problem in existing water replenishment optimization control methods that often fail to incorporate the dynamic energy efficiency characteristics and pressure state risks of pump sets into a unified optimization framework, leading to unreasonable load distribution and difficulty in reducing overall system energy consumption during multi-pump parallel operation. This solution creatively adopts an improved dynamic power consumption function model to solve for the minimum total power consumption load distribution ratio under multi-pump parallel operation. Under the premise of meeting the total water supply flow demand and operational safety constraints, it comprehensively considers the current dynamic power consumption characteristics and operational constraints of each water supply pump to obtain the load distribution ratio corresponding to the minimum total system power consumption, thus achieving adaptive energy consumption optimization control under different pump set energy efficiency levels and operating conditions.

[0155] Example 6, see Figure 1 and Figure 2 Based on the above embodiments, the artificial intelligence-based power plant boiler water supply scheduling optimization system provided by the present invention includes a data acquisition module, a pressure state characterization module, a pump group feature modeling module, and a water supply optimization control module.

[0156] The data acquisition module is used for data acquisition. Through data acquisition, it obtains the system operating status matrix and sends the system operating status matrix to the pressure status characterization module and the pump group feature modeling module.

[0157] The pressure state characterization module is used for pressure state characterization. Through pressure state characterization, a pressure state feature vector is obtained, and the pressure state feature vector is sent to the water replenishment optimization control module.

[0158] The pump set feature modeling module is used for three-dimensional feature modeling of the pump set. Through three-dimensional feature modeling of the pump set, a dynamic power consumption feature model is obtained, and the dynamic power consumption feature model is sent to the water replenishment optimization control module.

[0159] The water replenishment optimization control module is used for boiler water replenishment optimization control. Through boiler water replenishment optimization control, the optimal actuator control command is obtained.

[0160] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0161] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0162] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An AI-based optimization method for boiler feedwater scheduling in power plants, characterized by: The method includes the following steps: Step S1: Data acquisition, collect multi-source time-series data from the boiler side and pump set side to obtain the system operation status matrix; Step S2: Pressure state characterization. Calculate the first-order rate of change and second-order acceleration characteristics of the steam drum pressure data. Combined with the steam drum geometric parameters, steam-water ratio, and historical operating conditions, construct a thermodynamic nonlinear expansion compensation model including physical prior terms based on the thermal expansion of the steam drum metal wall, steam-water interface disturbance, and the bubble generation and coalescence process. Then, based on historical start-up and shutdown data, train a small-scale neural network correction term. Superimpose the thermodynamic nonlinear expansion compensation model and the small-scale neural network correction term to obtain an estimation function for calculating the transient false water level component, and calculate the false water level component. Based on this, the false water level component is removed from the actual measured water level in real time to obtain the true water level signal that represents the actual working fluid content inside the boiler. The amplitude and duration of the false water level component are normalized to obtain the false water level intensity index, which is then combined with the steam drum pressure and load change rate to obtain the pressure state feature vector. Step S3: Three-dimensional feature modeling of the pump group. The sliding time window is used to filter the operating data samples of the water pump under stable operating conditions. The weighted least squares three-dimensional surface fitting method with the introduction of time forgetting factor for flow rate, head and power data is used to perform three-dimensional feature modeling of the pump group and obtain the current dynamic power consumption feature model of each water pump. Step S4: Boiler feedwater optimization control, solve for the minimum total power consumption load distribution ratio under the parallel operation of multiple pumps, and obtain the optimal actuator control command for each feedwater pump.

2. The method for optimizing boiler feedwater scheduling in power plants based on artificial intelligence according to claim 1, characterized in that: In step S1, the data acquisition is specifically based on multivariate statistical criteria, which involves outlier removal, interpolation alignment, and smoothing filtering of the multi-source time-series data to obtain a system operating state matrix that is time-synchronized and retains higher-order derivative features.

3. The method for optimizing boiler feedwater scheduling in power plants based on artificial intelligence according to claim 2, characterized in that: In step S1, the multi-source time-series data specifically includes steam drum pressure data characterizing the thermal state of the steam drum, steam drum water level data characterizing the change in the liquid phase in the steam drum, main steam pressure data characterizing the boiler steam output condition, load command data characterizing the boiler operating target, feedwater flow rate data characterizing the supply capacity of the feedwater system, pump outlet pressure data characterizing the hydraulic output state of the feedwater pump, and pump motor power data characterizing the energy consumption characteristics of the feedwater pump.

4. The method for optimizing boiler feedwater scheduling in power plants based on artificial intelligence according to claim 3, characterized in that: In step S2, the pressure state characterization specifically includes the following steps: Step S21: Calculation of high-order dynamic features of pressure, used to construct the dynamic state features of steam drum pressure. Specifically, based on the system operating state matrix, the first-order rate of change and second-order acceleration features of steam drum pressure time series data are calculated to obtain the high-order dynamic features of pressure that characterize the rate of change and intensity of abrupt change of steam drum pressure. Step S22: Modeling the source of false water level, used to estimate the physical false water level component caused by transient changes in steam drum pressure. Specifically, by combining the steam drum geometric parameters, steam-water ratio and the aforementioned high-order dynamic characteristics of pressure, a thermodynamic nonlinear expansion compensation model is constructed, including the thermal expansion of the steam drum metal wall, the disturbance of the steam-water interface and the process of steam bubble generation and coalescence, to obtain the false water level estimation component based on physical priors. Step S23: Construct a small-scale neural network correction term to correct nonlinear false water level errors that are difficult to cover by the physical model. Specifically, a small-scale neural network is trained using historical start-up and shutdown data, with steam drum pressure, high-order dynamic features of pressure, and load change information as inputs, and the corresponding false water level correction component is output. Step S24: False water level component estimation, used to obtain the comprehensive estimation result of the transient false water level of the steam drum. Specifically, the false water level component output by the thermodynamic nonlinear expansion compensation model is superimposed with the small-scale neural network correction component to obtain the transient false water level component of the steam drum. Step S25: Reconstruction of the true mass water level signal, used to reconstruct the true water level signal representing the true working fluid inventory inside the boiler, specifically by removing the transient false water level component of the steam drum from the actual measured water level in real time to obtain the true mass water level signal; Step S26: Pressure state vector construction, used to construct a pressure state feature vector for water replenishment scheduling control. Specifically, the amplitude and duration of the transient false water level component of the steam drum are normalized to form a false water level intensity index, and the false water level intensity index is combined with the steam drum pressure and load change rate to obtain the pressure state feature vector.

5. The method for optimizing boiler feedwater scheduling in power plants based on artificial intelligence according to claim 4, characterized in that: In step S3, the three-dimensional feature modeling of the pump group includes the following steps: Step S31: steady-state sample screening; Step S32: construction of time forgetting factor weights; Step S33: optimization of abnormal pump groups; Step S34: construction of three-dimensional power consumption of the pump group; Step S35: power consumption feature fitting; Step S36: pump group feature modeling.

6. The method for optimizing boiler feedwater scheduling in power plants based on artificial intelligence according to claim 5, characterized in that: In step S3, the steady-state sample screening is used to construct an effective sample set for pump group three-dimensional feature modeling. Specifically, based on the system operation state matrix, the flow rate, head and power data of the water pump are judged for stable operating conditions within a sliding time window, and steady-state operation data samples that meet the condition that the fluctuations of flow rate, head and power are all less than a preset threshold are screened. The construction of the time forgetting factor weight is used to introduce a time weighting mechanism for the impact of equipment aging. Specifically, it constructs the time forgetting factor weight based on the timestamp of the steady-state operation data sample, so that the data sample closer to the current time occupies a higher weight in the model modeling process, and obtains a weighted sample set of data that reflects the changes of pump group characteristics over time. The abnormal pump group optimization is used to suppress the interference of abnormal samples on the modeling results of pump group characteristics. Specifically, it calculates the sample power residual based on the existing power consumption characteristic model, and constructs a robust weight function based on the residual amplitude. It reduces the weight of abnormal deviation samples in the modeling and obtains a comprehensive weighted sample that integrates time forgetting weight and robust weight. The three-dimensional power consumption construction of the pump set is used to construct a three-dimensional power consumption surface model structure of the pump set that conforms to physical characteristics. Specifically, it selects a polynomial basis function with flow rate and head as independent variables, establishes a three-dimensional surface expression of pump power with respect to flow rate and head, and introduces the physical constraint that power does not decrease monotonically with flow rate and head to limit the feasible range of model parameters. The power consumption feature fitting is used to solve the parameters of the time-varying power consumption feature model of the pump group. Specifically, under the conditions of the comprehensive weighted sample set and physical constraints, the weighted least squares method with the introduction of a time forgetting factor is used to solve the parameters of the three-dimensional power consumption surface model to obtain the dynamic power consumption feature model of each water pump at the current moment. The pump set feature modeling is used to extract the pump set operating domain and energy efficiency degradation features. Specifically, it determines the effective operating domain boundary of the water pump based on the dynamic power consumption feature model, and calculates the relative power consumption change index under standard operating conditions to obtain time-varying energy efficiency features that characterize the degree of energy efficiency degradation of the pump set.

7. The method for optimizing boiler feedwater scheduling in power plants based on artificial intelligence according to claim 6, characterized in that: In step S4, the boiler feedwater optimization control is used to achieve coordinated control of water level safety and operating energy consumption. Specifically, based on the dynamic power consumption characteristic model and the pressure state characteristic vector, the total feedwater flow demand is calculated in combination with the real water level signal. The total feedwater flow demand is used as a constraint condition. Based on the improved dynamic power consumption function model, the minimum total power consumption load allocation ratio under the parallel operation of multiple pumps is solved to obtain the optimal actuator control command for each feedwater pump.

8. An AI-based power plant boiler feedwater scheduling optimization system, used to implement the AI-based power plant boiler feedwater scheduling optimization method as described in any one of claims 1-7, characterized in that: It includes a data acquisition module, a pressure status characterization module, a pump set characteristic modeling module, and a water replenishment optimization control module.

9. The artificial intelligence-based power plant boiler feedwater scheduling optimization system according to claim 8, characterized in that: The data acquisition module is used for data acquisition. Through data acquisition, it obtains the system operating status matrix and sends the system operating status matrix to the pressure status characterization module and the pump group feature modeling module. The pressure state characterization module is used for pressure state characterization. Through pressure state characterization, a pressure state feature vector is obtained, and the pressure state feature vector is sent to the water replenishment optimization control module. The pump set feature modeling module is used for three-dimensional feature modeling of the pump set. Through three-dimensional feature modeling of the pump set, a dynamic power consumption feature model is obtained, and the dynamic power consumption feature model is sent to the water replenishment optimization control module. The water replenishment optimization control module is used for boiler water replenishment optimization control. Through boiler water replenishment optimization control, the optimal actuator control command is obtained.

Images