Hydro-meteorological combined power output risk driven twin correction control method

Abstract

This invention discloses a risk-driven twin-based control method for combined hydropower, wind power, and solar power output, belonging to the field of new energy combined power output control and equipment health management technology. The method includes: receiving automatic power generation control commands from the power grid; establishing a joint prediction model of command adjustment and degradation of vulnerable core components, and performing online calibration of the joint prediction model based on historical operating data; constructing a digital twin model, updating the current operating state and health status of vulnerable core components of the hydropower, wind power, solar power, and pumped storage power station system using historical operating data; constructing a risk-driven energy consumption-lifetime coupled objective function, obtaining the optimal control command through rolling optimization; and triggering the reshaping and issuance of the optimal control command when the calibration is satisfied. This invention, by constructing a joint prediction model of system operating state and the health status of vulnerable core components and introducing a digital twin online calibration mechanism, effectively improves the state recognition accuracy of the model and ensures a high degree of consistency between the control results and actual operating conditions.

Description

Technical Field

[0001] This invention relates to the field of new energy combined output control and equipment health management technology, and in particular to a risk-driven twin correction control method for combined hydropower, wind power and solar power output. Background Technology

[0002] With the continuous expansion of new energy grid connection, the joint output control of hydropower, wind power, and photovoltaic power has become an important technical means to improve the grid's peak-shaving and frequency regulation capabilities and promote the consumption of new energy. Among them, hydropower has relatively strong regulation capabilities, while wind power and photovoltaic power have volatility and randomness. The coordinated operation of the three can achieve power complementarity to a certain extent and improve the stability of the overall grid-connected output. Most existing hydropower, wind power, and photovoltaic power joint output control methods mainly focus on grid-connected power tracking and short-term economic dispatch. They usually adopt fixed parameter allocation, empirical rule control, or optimization allocation methods based on static models, which do not adequately consider the dynamic coupling relationship between multiple energy units, the impact of operating condition switching, and changes in health status under long-term operation.

[0003] In actual operation, the system frequently responds to automatic power generation control commands, and vulnerable core components are under constant adjustment, rapid switching, and continuous load, which can easily lead to problems such as thermal stress accumulation, mechanical fatigue, performance degradation, and parameter drift. This degradation not only reduces power point tracking accuracy and operational economy but may also increase the risk of failure and shorten component lifespan.

[0004] However, existing technologies lack effective characterization of the degradation process of such components, making it difficult to form risk assessment criteria that match actual operation. At the same time, existing methods generally lack online digital twin correction mechanisms that incorporate real-time operational data; once a deviation occurs between the model and the real object, the control effect gradually declines. Existing operation and maintenance methods also largely rely on periodic maintenance or post-incident handling, making it difficult to form a collaborative closed loop with control decisions. Summary of the Invention

[0005] Purpose of the Invention: To address the above problems, the purpose of this invention is to provide a risk-driven twin correction control method for combined hydropower, wind power, and solar power output. Based on receiving AGC power commands, this method forms a state and risk characterization that can be used for operational decision-making, and outputs executable control commands under operational constraints to reduce unnecessary shocks and losses during the adjustment process. At the same time, it incorporates maintenance strategy generation and historical operation data updates into a unified process, improving the continuity and consistency of operation management, thereby meeting grid connection scheduling requirements and improving the long-term operational adaptability of the device.

[0006] Technical solution: The present invention provides a risk-driven twin correction control method for combined hydropower, wind power, and solar power output, comprising the following steps:

[0007] Step 1: Receive the automatic power generation control command from the power grid and form the decision input for this cycle based on the command;

[0008] Step 2: Establish a joint prediction model for instruction adjustment and deterioration of vulnerable core components, and perform online correction of the joint prediction model based on historical operating data to generate risk prediction sequences and health risk indicators.

[0009] Step 3: Construct a digital twin model, use historical operating data to update the current operating status and health status of vulnerable core components of the hydropower, wind power, solar power and pumped storage power station system, and predict the operating status of the system in the future preset time domain;

[0010] Step 4: Based on the current operating status and the health status of core components, construct a risk-driven energy consumption-lifetime coupled objective function, and obtain the optimal control command through rolling optimization;

[0011] Step 5: Under the action of the optimal control command, update the current operating status of the hydropower, wind power, solar power and pumped storage power station system, and predict the operating status of the system in the future preset time domain.

[0012] Step 6: Perform a feasibility check on the simulated future operating state and generate operating constraints. If the check is not met, trigger constraint update and return to the rolling optimization step to continue optimization; if the check is met, trigger the optimal control command shaping and issuance.

[0013] Step 7: Generate maintenance strategies based on the generated risk prediction sequences, and write back the maintenance information to update the historical operation data to achieve integrated closed-loop iteration of maintenance and control strategies.

[0014] Preferably, step 2 includes:

[0015] Step 21, in each sampling period Within, obtain the operating state vector of the wind-solar combined power system. Control command vector Health status vector of vulnerable core components And the digital twin model parameter vector available at the current cycle prediction start time. ;

[0016] Step 22, construct the simultaneous prediction model, expressed as:

[0017] ,

[0018] ,

[0019] ,

[0020] In the formula, Indicates time-based Information on the future Predicted values ​​of the operating state vector of the combined wind and solar power system; Indicates time-based Information on the future Candidate control command vector for each step; Indicates the future number The rate of health degradation of easily damaged core components; Indicates time-based Information on the future Predicted values ​​of the health status vector of vulnerable core components; Indicates time Available model parameter vectors; Indicates the sampling period; This represents the system state transition mapping function driven by control commands; This represents a health degradation mapping function that is jointly determined by the operating state, control commands, current health state, and model parameters.

[0021] Step 23: Obtain the system operating state prediction sequence covering the rolling prediction window by simultaneously establishing the prediction model. Health status prediction sequence ;

[0022] Step 24, Define the risk mapping function , used to represent the future 1 The health risk prediction value for each step is expressed as:

[0023] ,

[0024] In the formula, Indicates the future number The dimensionless health risk prediction value of the step; Represents a health state vector The Middle Predicted values ​​for each component; Indicates the first The minimum health boundary allowed for each health state component; when When, the corresponding component risk is 0; when At that time, the corresponding component risk increases monotonically as health status declines;

[0025] Step 25: To unify the risks within the rolling domain to a single risk level, define the health risk indicator for the current sampling period as follows:

[0026] ,

[0027] In the formula, This indicates the health risk indicators for the current sampling period; This indicates the rolling forecast length, i.e., the number of steps considered forward;

[0028] Step 26: To ensure that measurement consistency and risk consistency both participate in online correction, a reference risk level is defined, and an online correction residual vector is constructed, expressed as:

[0029] ,

[0030] In the formula, This represents the measurement residual weighting matrix. Indicates the digital twin model at time... The predicted output for the running measurement vector, , Represents the measurement output matrix; Indicates the measurement bias term; This represents the risk residual weighting coefficient; This indicates the health risk indicators for the current sampling period; Indicates the reference risk level; Indicates time The running measurement vector is represented as:

[0031] ,

[0032] In the formula, This indicates the measured value of the combined output power; This indicates the measured value of grid-connected power deviation; This indicates the measured value of the adjustment rate; This indicates the temperature measurement value of the vulnerable core component;

[0033] Step 27: Update the digital twin model parameter vector using a recursive approach. This yields the unconstrained update parameter vector for the next time step. , represented as:

[0034] ,

[0035] ,

[0036] In the formula, This represents the intermediate parameter vector at the next time step without any boundary constraints applied. Represents the online correction gain matrix; Represents the fundamental gain matrix; Indicates the level of risk A variable gain adjustment function is used to adaptively adjust the parameter update magnitude according to the current risk level;

[0037] Then, a preset boundary constraint is applied to the unconstrained update parameter vector to obtain the digital twin model parameter vector for the next time step. , represented as:

[0038] ,

[0039] in, Represents the lower bound vector of the parameters; Represents the upper bound vector of the parameters; This represents the element-wise limiting operator, used to constrain the parameter vector within a preset allowable range. The element-wise limiting operator satisfies:

[0040] ,

[0041] In the formula, The dimension of the parameter vector of the digital twin model; Represents unconstrained update of parameter vector The One component; Indicates the first The allowable lower bound of each model parameter; Indicates the first The upper bound of each model parameter; This represents the parameter vector of the digital twin model at the next time step after applying boundary constraints. The One component;

[0042] Step 28, after boundary constraints Feedback is fed into the joint prediction model for rolling prediction in the next sampling period.

[0043] Preferably, step 3 includes:

[0044] Step 31: Construct a digital twin model comprising three elements: device operating status, health status, and measurement mapping. The device operating status includes combined output power status, regulation rate status, operating mode status, and temperature status. The discrete-time state evolution relationship of the digital twin model is expressed as follows:

[0045] ,

[0046] In the formula, The extended state vector of the digital twin is used to characterize the operating status and health status of vulnerable core components of hydropower, wind power, solar power and pumped storage power station systems. Represents the state transition matrix; Represents the control input matrix; Indicates the bias term;

[0047] Step 32: To establish consistency constraints between the twin state and the measured data, define the running measurement vector. satisfy:

[0048] ,

[0049] In the formula, This is the measurement error term;

[0050] Step 33: Based on the posterior state estimate of the previous time step, the control command of the previous time step, and the historical operation data, construct the prior state estimate of the current time step. Reuse time Running measurement vector By fusing and correcting the prior state estimates, we obtain the posterior state estimate vector at the current time step. , represented as:

[0051] ,

[0052] In the formula, Indicates time Prior state estimation vector before measurement fusion; Indicates time The posterior state estimation vector after measurement fusion; Indicates time The posterior state estimation vector; Indicates time The control command vector; Indicates time The parameter vector of the digital twin model; Represents the state fusion gain matrix;

[0053] Step 34, using the posterior state estimation vector at the current time. As an initial condition for rolling prediction, in the future control command sequence Given the conditions, forward extrapolation is performed according to the state evolution relationship of the digital twin model to obtain the future augmented state prediction sequence and the corresponding future operational measurement prediction sequence within the rolling prediction window, which are expressed as follows:

[0054] ,

[0055] ;

[0056] In the formula, The state evolution function of the digital twin model is used to generate the prior state estimate at the current time step based on the posterior state estimate of the previous time step, control commands, and model parameters.

[0057] Step 35: Extract the healthy state sub-vector from the extended state vector, denoted as: ,in, Select a matrix for health status.

[0058] Preferably, step 4 includes:

[0059] Step 41, define the rolling optimization decision variables, represented as: ,in, The sequence of control commands to be optimized. For at any time Calculate and plan for future moments The vector of control instructions to be executed;

[0060] Step 42, construct the tracking performance cost term, represented as:

[0061] ,

[0062] In the formula, For reference target output, It is a positive semidefinite weight matrix;

[0063] The energy loss cost term is constructed and represented as follows:

[0064] ,

[0065] In the formula, For the predicted active power at future moments, To predict the power consumption of auxiliary equipment, For equivalent efficiency, Indicates the sampling period;

[0066] The construction lifetime-switching joint cost term is represented as:

[0067] ,

[0068] In the formula, These are dimensionless weighting coefficients. This is a start / stop switching indicator, with a value of 0 or 1. This is the predicted equivalent temperature value for vulnerable core components. It is a norm 2;

[0069] Step 43, construct the risk-driven twin correction energy consumption-lifetime coupling objective function, expressed as:

[0070] ,

[0071] In the formula, , , This is a dimensionless weighting coefficient.

[0072] Preferably, step 4 further includes:

[0073] Step 44, construct the set of vectors of inequality constraint functions. The joint modeling and risk assessment device outputs a rolling domain risk sequence and defines the risk level for the current period.

[0074] Construct a risk-tightened feasible region, denoted as:

[0075] ,

[0076] In the formula, This corresponds to the upper bound vector of the constraint. The mapping coefficient from risk to constraint margin;

[0077] Introducing risk smoothing value Risk of price limits , respectively represented as:

[0078] ,

[0079] ,

[0080] In the formula, For smoothing coefficients, For bounded operators, This is the upper limit threshold for risk.

[0081] Step 45, employ the rolling shift operator. The initial candidate control sequence is generated as follows:

[0082] ,

[0083] In the formula, The optimal control command sequence obtained in the previous sampling period is indicated by the superscript. Indicates the optimal solution;

[0084] Step 46, in By performing a local quadratic transformation on the objective function in the vicinity, a quadratic approximation is obtained, which is expressed as:

[0085] ,

[0086] In the formula, It is the control sequence increment, representing the adjustment amount relative to the operating point. It is the gradient vector of the objective function with respect to the control sequence. It is a symmetric positive semi-definite second-order curvature approximation matrix; superscript Represents the transpose of a vector;

[0087] Introducing the trust region radius By limiting the applicability of the quadratic approximation and regularizing the curvature matrix, the objective function is approximated as follows:

[0088] ,

[0089] In the formula, is the regularization coefficient, taking a non-negative value. It is the identity matrix;

[0090] Step 47: Solve the sequential quadratic programming subproblem within the risk-tightening feasible region to obtain the optimal control increment, expressed as:

[0091] ,

[0092] In the formula, It is the optimal solution to the subproblem;

[0093] When a subproblem becomes infeasible, a deterministic backoff strategy is implemented: the trust region radius is adjusted. Update it to its current value and trust region shrinkage coefficient The product of the regularization coefficients and the regularization coefficients Update it to its current value and regularization amplification factor. The product of these is then reconstructed and the subproblems are solved until a feasible solution is obtained.

[0094] Step 48: Transform the subproblem solution into a new candidate control sequence, and update the magnitude using risk level scheduling, expressed as:

[0095] ,

[0096] In the formula, It is the updated control sequence; It is the step size coefficient, which controls the step size used in each iteration. The proportion; Maximum base step size;

[0097] Step 49: Output the optimal instruction and proceed to the next time step. The closed loop will use the first control value of the updated sequence as the instruction to be issued, represented as:

[0098] ,

[0099] In the formula, It is a moment The optimal control command; This indicates the first step of taking the control sequence.

[0100] Preferably, step 7 includes:

[0101] Step 71, to predict the risk sequence As input, the maintenance trigger strength is calculated using the following formula:

[0102] ,

[0103] In the formula, To maintain the evaluation window length and satisfy , These are dimensionless weighting coefficients. Indicates the sampling sequence number The future number is obtained based on online correction parameters and digital twin predictions. Step risk prediction value, risk threshold To the maximum allowable risk level, For positive part operators, satisfying ;

[0104] Step 72, Define the set of candidate maintenance actions Each action within the set Given a defined maintenance strategy, the optimal maintenance action satisfies:

[0105] ,

[0106] In the formula, Indicates the action The maintenance cost This indicates the cost of the shutdown. This represents the residual risk sequence after the action is performed. This is the risk trade-off coefficient;

[0107] Step 73, Maintenance Action At any moment Once the maintenance is completed, the system will write the maintenance action type, execution timestamp, key state quantities before and after execution, and corresponding risk sequences into the historical operation data, and simultaneously update the current state and health status of the device, serving as the unified starting point data for the next sampling cycle of joint modeling and risk assessment, digital twin state prediction, constraint tightening, and rolling optimization.

[0108] Beneficial effects: Compared with the prior art, the significant advantages of this invention are:

[0109] 1. This invention constructs a joint prediction model of the system's operating status and the health status of vulnerable core components, and introduces a digital twin online correction mechanism, which effectively improves the accuracy of the model's status identification and ensures a high degree of consistency between the control results and the actual operating conditions.

[0110] 2. This invention integrates power point tracking, health degradation, risk level and adjustment cost into a rolling optimization framework, thereby achieving synergistic optimization of joint output control target and lifetime loss suppression target, and overcoming the limitation of traditional methods that emphasize efficiency over lifetime.

[0111] 3. This invention significantly enhances the system's constraint adaptability and control stability under high-risk conditions by aggregating, smoothing, and limiting risks, and constructs a risk-tightened feasible region accordingly, thereby reducing the risk of instability in extreme scenarios.

[0112] 4. This invention constructs a complete closed loop from state estimation, risk assessment, optimization solution to maintenance decision-making, systematically improving the safety, economy and engineering practicality of combined hydropower, wind and solar power operation, and has good value for promotion and application. Attached Figure Description

[0113] Figure 1 This is a flowchart of the present invention;

[0114] Figure 2 A flowchart of a quadratic programming rolling optimization algorithm for twin-corrected sequences with risk adaptive constraint tightening;

[0115] Figure 3 A statistical comparison chart of power regulation switching intensity;

[0116] Figure 4 A comparison chart of the lifespan loss evolution of vulnerable core components;

[0117] Figure 5 A graph showing how health risk indicators change over time;

[0118] Figure 6 A comparative chart of energy consumption statistics for the entire operating cycle;

[0119] Figure 7 A comparison chart of the overall operational costs under risk-driven twin correction. Detailed Implementation

[0120] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the embodiments of the present invention, and not all structures.

[0121] In the following description, specific details such as target system architecture and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0122] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0123] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0124] Furthermore, in the description of this application and the appended claims, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0125] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include the target features, structures, or characteristics described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0126] Combination Figure 1 As shown in this embodiment, the risk-driven twin correction control method for combined hydropower, wind power, and solar power output includes the following steps:

[0127] Step 1: Receive the automatic power generation control command from the power grid and form the decision input for this cycle based on the command.

[0128] The system receives power commands from Automatic Generation Control (AGC) and uses these commands as the unified driving force for operational decisions in the current cycle. Simultaneously, it calls upon historical operational data generated in the previous cycle, which includes at least historical AGC power command data, hydropower unit operational measurement data, wind power operational measurement data, photovoltaic operational measurement data, grid-connected side operational measurement data, device status baseline, vulnerable core component health status baseline, online calibration-related data, historical risk prediction information, and recorded maintenance execution information. Based on the historical operational data, it completes data consistency verification and status baseline preparation, forming a usable dataset for the current cycle. This usable dataset serves as the unified input basis for subsequent joint modeling and risk assessment, digital twin calibration prediction, constraint generation, and rolling optimization solutions, ensuring that subsequent modeling, prediction, verification, and optimization are all based on the same data context. The consistency verification results and aligned data are written into the historical operational data for use in the next cycle.

[0129] Step 2: Establish a joint prediction model for instruction adjustment and degradation of vulnerable core components, and perform online calibration of the joint prediction model based on historical operating data to generate risk prediction sequences and health risk indicators.

[0130] To ensure that control command optimization can simultaneously constrain the degradation trends of vulnerable core components, this example combines descriptions of device operating dynamics and component health evolution within a unified mathematical framework. Health risk indicators are used as online correction signals to guide adaptive updates of digital twin parameters. This maintains model consistency under fluctuating operating conditions and long-term drift, providing reliable input for subsequent loss assessment, operational constraint generation, and objective function optimization.

[0131] The mechanism by which joint output adjustment behavior affects the evolution of the plant's operating status is coupled with the mechanism by which vulnerable core components deteriorate within the same prediction framework, forming a joint prediction model. Based on this joint prediction model, the future operating status of the plant and the future health status of vulnerable core components within a rolling prediction window are recursively predicted. The predicted future health status is then converted into a risk prediction sequence for the rolling prediction window based on a pre-defined risk mapping mechanism. Based on this risk prediction sequence, an aggregation criterion is used to extract health risk indicators representing the most unfavorable risk level within the rolling prediction window, providing a unified risk input basis for subsequent constraint management and maintenance decisions.

[0132] Further innovations in risk-driven online twin calibration and health risk inference are implemented. Online calibration residuals are constructed by integrating measurement consistency information and risk consistency information. Measurement consistency information characterizes the deviation between model output and measured operational data, while risk consistency information characterizes the deviation between health risk indicators and reference risk levels. A weighting mechanism ensures the comparability and synergistic constraint of these two types of deviations on a numerical scale. Based on the online calibration residuals, the parameter representations and key degradation representations of the simultaneous prediction model are updated. An adaptive calibration gain associated with the health risk level is used to adjust the calibration strength, making the calibration more sensitive and maintaining stable convergence under high-risk conditions. The online calibration parameters, rolling prediction window risk prediction sequence, and health risk indicators are output as unified prior inputs for digital twin state estimation and prediction, risk adaptive constraint tightening, and rolling optimization solution, ensuring that subsequent control and maintenance closed-loop operations operate on the basis of prediction consistency.

[0133] Further, step 2 includes:

[0134] Step 21, in each sampling period Within, obtain the operating state vector of the wind-solar combined power system. Control command vector Health status vector of vulnerable core components And the digital twin model parameter vector available at the current cycle prediction start time. , Determined by historical operating data and the results of the previous online calibration cycle. At least include the combined output power state. It is a dimensionless normalized quantity, with a value range from 0 to 1. The larger the value, the higher the health level.

[0135] Step 22, construct the simultaneous prediction model, expressed as:

[0136] ,

[0137] ,

[0138] ,

[0139] In the formula, Indicates based on Moment information for the future Predicted values ​​of the operating state vector of the combined solar-wind power system of Bushui; Indicates based on Moment information for the future Candidate control command vector for each step; Indicates the future number The rate of health degradation of easily damaged core components; Indicates based on Moment information for the future Predicted values ​​of the health status vector of vulnerable core components; express The model parameter vector available at any given time; Indicates the sampling period; Represents the state transition matrix; Represents the control input matrix; Represents the state bias vector; This matrix represents the coefficients representing the influence of operating conditions on the rate of health degradation. This represents the matrix of coefficients representing the influence of control commands on the rate of health degradation. The autoregressive coefficient matrix represents the rate of health deterioration in relation to the current health status; Represents the health degradation bias vector; , , , , , and All are composed of model parameter vectors The characteristics are derived from historical operational data and are recursively updated through online correction during operation.

[0140] Step 23: Obtain the system operating state prediction sequence covering the rolling prediction window by simultaneously establishing the prediction model. Health status prediction sequence This serves as the basis for subsequent risk calculations and online parameter corrections;

[0141] Step 24, Define the risk mapping function , used to represent the future 1 The health risk prediction value for each step is expressed as:

[0142] ,

[0143] In the formula, Indicates the future number The dimensionless health risk prediction value of the step; Represents a health state vector The Middle Predicted values ​​for each component; Indicates the first The minimum health boundary allowed for each health state component; when When, the corresponding component risk is 0; when At that time, the corresponding component risk increases monotonically as health status declines.

[0144] Through risk mapping function Output the maximum risk value among all health status components, and use it as the unified health risk prediction value for this step.

[0145] Step 25: To unify the risks within the rolling domain to a single risk level, define the health risk indicator for the current sampling period as follows:

[0146] ,

[0147] In the formula, This indicates the health risk indicators for the current sampling period; This indicates the rolling forecast length, which is the number of steps considered forward.

[0148] Step 26: To ensure that measurement consistency and risk consistency both participate in online correction, a reference risk level is defined, and an online correction residual vector is constructed, expressed as:

[0149] ,

[0150] In the formula, Indicates time The running measurement vector, which is the same variable as the running measurement vector defined in step 32, is expressed as follows:

[0151] ,

[0152] In the formula, This represents the measured combined output power, in MW. This represents the measured value of grid-connected power deviation, in MW. This indicates the measured adjustment rate, in MW / s. This indicates the temperature measurement value of the vulnerable core component, in °C. Indicates the digital twin model at time... The predicted output of the running measurement vector is the same function as the measurement mapping in step 32; Represents the measurement residual weighting matrix; This represents the risk residual weighting coefficient; This indicates the health risk indicators for the current sampling period; This indicates the reference risk level.

[0153] Step 27: Update the digital twin model parameter vector using a recursive approach. This yields the unconstrained update parameter vector for the next time step. , represented as:

[0154] ,

[0155] ,

[0156] In the formula, This represents the intermediate parameter vector at the next time step without any boundary constraints applied. This represents the online correction gain matrix, which determines the strength of the parameter's response to the residual. Represents the fundamental gain matrix; Indicates the level of risk The variable gain adjustment function is expressed as follows:

[0157] ,

[0158] in, Indicates the lower bound of the gain. Let represent the upper bound of the gain, and satisfy . ; Indicates the risk limit threshold; when the risk level When it rises, Monotonically increasing, thus improving the sensitivity to parameter updates; when the risk level At lower levels, Take a smaller value to make the update smoother.

[0159] Then, a preset boundary constraint is applied to the unconstrained update parameter vector to obtain the digital twin model parameter vector for the next time step. , represented as:

[0160] ,

[0161] in, Represents the lower bound vector of the parameters; Represents the upper bound vector of the parameters; This represents the element-wise limiting operator, used to constrain the parameter vector within a preset allowable range. The element-wise limiting operator satisfies:

[0162] ,

[0163] In the formula, The dimension of the parameter vector of the digital twin model is a dimensionless integer. Represents unconstrained update of parameter vector The One component; Indicates the first The allowable lower bound of each model parameter; Indicates the first The upper bound of each model parameter; This represents the parameter vector of the digital twin model at the next time step after applying boundary constraints. The One component;

[0164] Step 28, after boundary constraints Feedback is sent to the joint prediction model for rolling prediction in the next sampling period, thus forming a closed-loop mechanism of prediction, risk output, online correction, and re-prediction.

[0165] Step 3: Construct a digital twin model, use historical operating data to update the current operating status and health status of vulnerable core components of the hydropower, wind power, solar power and pumped storage power station system, and predict the operating status of the system in the future preset time domain.

[0166] Furthermore, step 3 includes:

[0167] Step 31: Construct a digital twin model comprising three elements: device operating status, health status, and measurement mapping. The device operating status includes combined output power status, regulation rate status, operating mode status, and temperature status. The discrete-time state evolution relationship of the digital twin model is expressed as follows:

[0168] ,

[0169] In the formula, The extended state vector of the digital twin is used to characterize the operating status and health status of vulnerable core components of hydropower, wind power, solar power and pumped storage power station systems. Represents the state transition matrix; Represents the control input matrix; The bias term is represented by the state evolution function of the digital twin model, which is denoted as . And satisfy:

[0170] ,

[0171] Used to characterize deterministic effects such as environmental disturbances, inherent losses and reference drift, the combined output state evolution law of hydropower, wind power and photovoltaic under the action of control commands, as well as the health evolution law of vulnerable core components under the stress of power ramping, start-stop switching and core temperature change, are uniformly written into the same extended state space, thereby ensuring that the operation state inference and the health state inference are consistent within the same model framework.

[0172] Step 32: To establish consistency constraints between the twin state and the measured data, define the running measurement vector. Represented as:

[0173] ,

[0174] And satisfy:

[0175] ,

[0176] In the formula, The same variable as the running measurement vector used in step 26 to construct the online correction residual vector; This is the measurement error term; The measurement mapping is the same function as the measurement mapping in step 26, and its expression is:

[0177] ,

[0178] In the formula, Represents the measurement output matrix; This indicates the measurement bias term.

[0179] Step 33: Based on the posterior state estimate of the previous time step, the control command of the previous time step, and the historical operation data, construct the prior state estimate of the current time step. Then, using the unified operation measurement vector defined in steps 26 and 32 By fusing and correcting the prior state estimates, we obtain the posterior state estimate vector at the current time step. , represented as:

[0180] ,

[0181] In the formula, Indicates time Prior state estimation vector before measurement fusion; Indicates time The posterior state estimation vector after measurement fusion; Indicates time The posterior state estimation vector; Indicates time The control command vector; Indicates time The parameter vector of the digital twin model; This represents the state fusion gain matrix, used to adjust the strength of the effect of measurement residuals on state correction;

[0182] Step 34, using the posterior state estimation vector at the current time. As an initial condition for rolling prediction, in the future control command sequence Given the conditions, forward extrapolation is performed according to the state evolution relationship of the digital twin model to obtain the future extended state prediction sequence and the corresponding future operation measurement prediction sequence within the rolling prediction window, which are expressed as follows:

[0183] ,

[0184] ,

[0185] In the formula, the state evolution function of the digital twin model and the discrete-time state evolution relationship of the digital twin model in step 31 are expressed by the same expression, which is:

[0186] ,

[0187] in, Indicates the future number The digital twin of the step extends the state prediction vector; Represents the extended state transition matrix; This represents the extended control input matrix; Indicates the extended bias term; , and All are composed of model parameter vectors Characterized and updated online.

[0188] Step 35: Extract the healthy state sub-vector from the extended state vector, denoted as: ,in, A health status selection matrix is ​​used to extract health status predictions for vulnerable core components from the extended states.

[0189] To achieve digital twin state estimation and prediction, and to ensure that the online correction parameters output in step 2 can be updated consistently at the twin layer for the device state and the health characterization of vulnerable core components, this step constructs a digital twin model that includes three types of elements: operating state, health state, and measurement mapping.

[0190] After obtaining online calibration parameters, risk prediction sequences, and health risk indicators, risk-driven twin calibration is performed. The digital twin model is updated to ensure parameter consistency based on the online calibration parameters. Using the device's operating state and health status at the current sampling time as initial conditions, a joint recursive deduction is performed on the device's operating state and the health status of vulnerable core components at multiple future sampling times, forming a future state prediction result consistent with the rolling prediction window. During the deduction process, the influence mechanism of control command adjustment on the joint output response and the degradation evolution mechanism of vulnerable core components are expressed in the same state space, ensuring consistency between future state prediction and future health prediction at the mechanistic level. Based on the deduction results, the future state prediction sequence of the device and the future health prediction sequence of vulnerable core components are output, serving as input for subsequent prediction feasibility verification, operational constraint generation, and risk adaptive constraint tightening. Simultaneously, the future state prediction results and risk prediction sequences are correlated to ensure that subsequent rolling optimization solutions and maintenance strategy decisions are based on the same twin prediction results and meet the closed-loop consistency requirement.

[0191] Step 4: Based on the current operating status and the health status of core components, construct a risk-driven energy consumption-lifetime coupled objective function, and obtain the optimal control command through rolling optimization.

[0192] Furthermore, step 4 includes:

[0193] Step 41, define the rolling optimization decision variables, represented as: ,in, The sequence of control commands to be optimized. For at any time Calculate and plan for future moments The vector of control instructions to be executed; The rolling prediction length represents the number of steps considered forward.

[0194] Step 42, construct the tracking performance cost term, represented as:

[0195] ,

[0196] In the formula, For reference target output, It is a positive semi-definite weight matrix, used to reflect the importance of different output components to tracking performance;

[0197] The energy loss cost term is constructed and represented as follows:

[0198] ,

[0199] In the formula, For the predicted active power at future moments, To predict the power consumption of auxiliary equipment, For equivalent efficiency, the value ranges from 0 to 1 and is determined by digital twin parameters. It participates in the calculation, enabling energy consumption assessments to be updated synchronously with online corrections;

[0200] The construction lifetime-switching joint cost term is represented as:

[0201] ,

[0202] In the formula, It is a dimensionless weighting coefficient used to balance the contribution of climbing impact, start-stop switching and temperature difference impact to lifespan loss. This is a start / stop switching indicator, with a value of 0 or 1, determined by whether the discrete operating mode of adjacent steps changes, and originating from the mode selection and strategy switching in the control command; The equivalent temperature prediction value of the vulnerable core components is obtained by digital twin prediction and is used to characterize the driving effect of temperature shock on degradation. It is a norm 2; It is a dimensionless quantity used to characterize the impact intensity of the command climbing slope;

[0203] Step 43, construct the risk-driven twin correction energy consumption-lifetime coupling objective function, expressed as:

[0204] ,

[0205] In the formula, , , It is a dimensionless weighting coefficient used to weigh tracking performance, energy loss, and lifetime switching costs.

[0206] Combination Figure 2 As shown, step 4 further includes:

[0207] Step 44, construct the set of vectors of inequality constraint functions. This encompasses constraints such as power limits, ramp rate boundaries, temperature boundaries, start-stop logic, and equipment safety boundaries. The joint modeling and risk assessment device outputs a rolling domain risk sequence and defines the risk level for the current period, expressed as:

[0208] ,

[0209] In the formula, For dimensionless risk levels, For rolling domain risk sequences;

[0210] Construct a risk-tightened feasible region, denoted as:

[0211] ,

[0212] In the formula, This corresponds to the upper bound vector of the constraint. The mapping coefficient from risk to constraint margin;

[0213] To mitigate frequent fluctuations in the feasible region caused by short-term volatility in risk estimates, a risk smoothing value is introduced. Risk of price limits , respectively represented as:

[0214] ,

[0215] ,

[0216] In the formula, For smoothing coefficients, For bounded operators, This is the upper limit threshold for risk.

[0217] This construction ensures that the higher the risk level, the more conservative the constraints, and improves the time consistency of the feasible region and the stability of online solutions.

[0218] Step 45, employ the rolling shift operator. The initial candidate control sequence is generated as follows:

[0219] ,

[0220] In the formula, The optimal control command sequence obtained in the previous sampling period is indicated by the superscript. Indicates the optimal solution; This serves as the initial candidate control sequence for the current cycle, used to improve the convergence speed of the online solution and maintain control continuity.

[0221] Step 46, in By performing a local quadratic transformation on the objective function in the vicinity, a quadratic approximation is obtained, which is expressed as:

[0222] ,

[0223] In the formula, It is the control sequence increment, representing the adjustment amount relative to the operating point. It is the gradient vector of the objective function with respect to the control sequence, reflecting the marginal impact of each control action on energy consumption and lifetime cost; It is a symmetric, positive semi-definite second-order curvature approximation matrix, used to characterize local variation trends and ensure the convexity of subproblems; superscript Represents the transpose of a vector;

[0224] However, when the device is operating in a highly nonlinear, strongly coupled, or near-constraint boundary condition, the single-point quadratic approximation may exceed its effective range, making... The insufficient characterization of the true curvature causes the update direction obtained in subsequent solutions to deviate from the true descent direction. Therefore, this step is improved by introducing a trust region radius. By limiting the applicability of the quadratic approximation and regularizing the curvature matrix to enhance numerical stability, the objective function is approximated as follows:

[0225] ,

[0226] In the formula, This is the regularization coefficient, which takes a non-negative value and is used to improve the stability of curvature approximation; It is the identity matrix; The radius of the trust region is used to constrain the magnitude of a single update to not exceed the local approximate reliable interval.

[0227] This improvement enhances convergence reliability and solution quality under strongly nonlinear conditions while maintaining online solvability.

[0228] Step 47: Solve the sequential quadratic programming subproblem within the risk-tightening feasible region to obtain the optimal control increment, expressed as:

[0229] ,

[0230] In the formula, It is the optimal solution to the subproblem;

[0231] Constraints This means that the higher the risk, the more conservative the solution space, avoiding lifetime impact and feasibility risks. When a subproblem becomes infeasible, a deterministic backoff strategy is implemented: the trust region radius is adjusted. Update it to its current value and trust region shrinkage coefficient The product of the regularization coefficients and the regularization coefficients Update it to its current value and regularization amplification factor. The product of these is then reconstructed and the subproblems are solved until a feasible solution is obtained. , ;

[0232] Step 48: Transform the subproblem solution into a new candidate control sequence, and update the magnitude using risk level scheduling, expressed as:

[0233] ,

[0234] In the formula, It is the updated control sequence; It is the step size coefficient, which controls the step size used in each iteration. The proportion, risk The larger, The smaller the size, the more conservative the updates, thus prioritizing lifespan friendliness and feasibility margin in high-risk phases, and reducing energy consumption costs in low-risk phases. Maximum base step size;

[0235] Step 49: Output the optimal instruction and proceed to the next time step. The closed loop will use the first control value of the updated sequence as the instruction to be issued, represented as:

[0236] ,

[0237] In the formula, It is a moment The optimal control command; This indicates the first step of taking the control sequence.

[0238] This instruction enters the instruction shaping and distribution phase, completes the amplitude limiting, rate limiting, and formatting distribution, and is then output to the execution layer, subsequently proceeding to the next sampling time. The device's current state, future state, risk output, correction, and optimization closed loop.

[0239] Under the combined effect of the energy consumption-lifetime coupled objective function model and the risk-adaptively tightened set of operating constraints, a twin-corrected sequential quadratic programming rolling optimization algorithm is used for online solution. The optimization problem is transformed into a sequential quadratic programming subproblem constructed and solved in each sampling period through a rolling window solution strategy. The rolling shift result of the optimal control command sequence in the previous sampling period is used as the initial value for hot start to improve solution convergence and control continuity. In each solution, the optimal control command that satisfies the set of operating constraints is output, and after command shaping through the control output interface, it is issued to satisfy the dynamic response and boundary constraints of the execution layer. At the same time, the optimal control command is substituted back into the digital twin prediction link to generate future state and risk prediction results consistent with the executable control strategy, so as to provide a consistent prediction basis for the next step of maintaining strategy decision-making and updating historical operating data.

[0240] Step 5: Under the action of the optimal control command, update the current operating status of the hydropower, wind power, solar power and pumped storage power station system, and predict the operating status of the system in the future preset time domain.

[0241] After obtaining the executable optimal control scheme, the maintenance strategy decision and historical operation data update are performed based on the risk prediction sequence obtained under the constraints of the optimal control scheme. The maintenance trigger intensity is calculated based on the risk prediction sequence within the rolling prediction window to characterize the cumulative exceedance of future risks in the time dimension. The maintenance decision process is initiated when the maintenance trigger intensity reaches the preset trigger condition.

[0242] Step 6: Perform a feasibility check on the predicted future operating state and generate operating constraints. If the check is not met, trigger constraint update and return to the rolling optimization step to continue optimization; if the check is met, trigger the issuance of the optimal control command shaping.

[0243] The feasibility verification process for predicting the future state of the device derived from digital twin simulation involves comparing the operational trajectory within the future rolling prediction window against grid connection rules, device safety boundaries, and operational strategy boundaries to arrive at a feasibility verification conclusion. Based on this conclusion, an operational constraint set is generated, which includes at least constraints on joint output adjustment limits, adjustment rate constraints, start-stop and mode switching constraints, safe operation constraints for vulnerable core components, and lifespan protection-related constraints. If the feasibility verification is not met, the operational constraint set is updated and output. Subsequently, the constraint margin of the operational constraint set is adjusted based on health risk indicators. Risk mapping adjustment ensures that constraint margins converge when health risk levels rise and release when health risk levels fall. Risk smoothing is employed to suppress frequent constraint changes caused by short-term risk fluctuations, and amplitude limiting is used to restrict the tightening of constraints to avoid feasible region collapse. The output is a set of operational constraints that has been adaptively tightened by risk. Under the constraints of the adaptively tightened set of operational constraints, the risk-driven twin correction energy consumption-lifetime coupling objective function model is updated. The joint output command tracking performance, energy loss, lifetime loss, and switching cost are co-expressed in the same objective function. The objective function model and the adaptively tightened set of operational constraints are used together as inputs for subsequent rolling optimization solutions.

[0244] Step 7: Generate maintenance strategies based on the generated risk prediction sequences, and write back the maintenance information to update the historical operation data to achieve integrated closed-loop iteration of maintenance and control strategies.

[0245] Further, step 7 includes:

[0246] Step 71, to predict the risk sequence As input, the maintenance trigger strength is calculated using the following formula:

[0247] ,

[0248] In the formula, To maintain the evaluation window length and satisfy , These are dimensionless weighting coefficients. Indicates the sampling sequence number The future number is obtained based on online correction parameters and digital twin predictions. Step risk prediction value, risk threshold To the maximum allowable risk level, For positive part operators, satisfying ;

[0249] Step 72, Define the set of candidate maintenance actions Each action within the set Given a defined maintenance strategy, the optimal maintenance action satisfies:

[0250] ,

[0251] In the formula, Indicates the action The maintenance cost This indicates the cost of the shutdown. This represents the residual risk sequence after the action is performed. This is the risk trade-off coefficient;

[0252] Step 73, Maintenance Action At any moment Once the maintenance is completed, the system will write the maintenance action type, execution timestamp, key state quantities before and after execution, and corresponding risk sequences into the historical operation data, and simultaneously update the current state and health status of the device, serving as the unified starting point data for the next sampling cycle of joint modeling and risk assessment, digital twin state prediction, constraint tightening, and rolling optimization.

[0253] During the maintenance decision-making process, a set of candidate maintenance actions is constructed, and each candidate action undergoes a calculable cost assessment and residual risk assessment. The cost assessment includes at least the cost of maintenance resource consumption and the cost of downtime loss. The residual risk assessment characterizes the remaining risk level of the system in the subsequent rolling window after the maintenance action is executed. A trade-off mechanism unifies the cost assessment and residual risk assessment into a comprehensive evaluation criterion for maintenance actions. Based on the comprehensive evaluation criterion, the optimal maintenance action is determined and maintenance execution is triggered. After the maintenance is completed, the current operating status of the device and the health status of vulnerable core components are updated. At the same time, the information of this maintenance action, the maintenance execution time, the status before and after maintenance, and the risk prediction information related to maintenance are written into the historical operating data. The updated historical operating data is fed back to the control command input, joint modeling and risk assessment, and digital twin correction prediction link as the starting data for the next sampling cycle rolling prediction, risk constraint tightening, and rolling optimization solution, thereby forming a closed-loop operation mechanism integrating maintenance and control, and ensuring that the life protection target of vulnerable core components is continuously reflected in the continuous control cycle.

[0254] To further illustrate the superiority of the risk-driven twin correction control method for combined hydropower, wind power, and solar power output described in this invention, it is compared with the traditional combined hydropower, wind power, and solar power output control system. Figure 3 The statistical comparison results of power regulation switching intensity between a traditional hydro-wind-solar combined power output control system and the method of this invention under the same operating conditions are presented. This index, based on the same AGC command input, quantifies the equivalent change in power regulation amplitude per unit time to reflect the switching frequency of the system during continuous regulation. Figure 3It can be seen that traditional systems exhibit high switching intensity under the strategy of pursuing fast power point tracking. However, this invention, by jointly modeling the control command adjustment behavior with the degradation process of vulnerable core components and introducing a risk-driven twin correction mechanism, effectively suppresses unnecessary frequent switching while meeting adjustment requirements, providing favorable operating conditions for subsequent life management and risk control.

[0255] Figure 4 This paper demonstrates the time evolution characteristics of cumulative lifespan loss of vulnerable core components under a conventional system and the method of this invention during continuous AGC (Automatic Guided Control) adjustments. It can be observed that the conventional control strategy, due to its failure to explicitly consider the impact of adjustment behavior on equipment degradation, exhibits a faster rate of lifespan loss growth. In contrast, this invention incorporates lifespan loss into a unified modeling and optimization framework during the instruction generation stage, effectively suppressing the rate of lifespan loss growth through risk-driven twin correction and constraint adjustment. These results demonstrate that this invention can achieve proactive management of the service status of vulnerable core components while ensuring system adjustment performance, showcasing the advantages of a lifespan-friendly control strategy.

[0256] Figure 5 This paper presents the changes in health risk indicators of vulnerable core components estimated based on a digital twin model during operation. These risk indicators comprehensively reflect the component's degradation state and future failure trend, representing a concentrated embodiment of the results of simultaneous modeling and twin correction. As shown in the figure, the risk indicators of traditional systems continuously increase over time. However, this invention, by introducing an online correction mechanism and risk-driven constraint adjustment, makes the risk evolution process smoother, avoiding the rapid accumulation of high-risk states. This demonstrates that this invention can transform equipment health risk from post-event assessment into a pre-event constraint in control decisions, effectively improving the safety and controllability of system operation.

[0257] Figure 6 The equivalent cumulative operating energy consumption index of the conventional system and the method of this invention was compared over a complete operating cycle. This index, based on the basic energy consumption, comprehensively considers the additional energy loss caused by frequent power switching, reflecting the overall energy consumption level under different control strategies. The results show that the conventional system, due to its more aggressive adjustment behavior, has a significant proportion of additional losses, while the present invention effectively reduces switching-related energy consumption through a smooth adjustment strategy under risk constraints. This result demonstrates that incorporating lifetime and risk factors into control decisions helps to achieve a more economical operating mode at the system level.

[0258] Figure 7This paper presents a comparison of the equivalent comprehensive operating costs of traditional systems and the method of this invention under a unified evaluation caliber. This comprehensive cost index simultaneously considers operating energy consumption, the lifespan depletion of vulnerable core components, and adjustment-related costs to reflect the system's operating characteristics from a full lifecycle perspective. It can be seen that this invention significantly outperforms traditional control strategies in terms of comprehensive cost. This is mainly due to its explicit introduction of lifespan depletion and health risks into the objective function, and the achievement of a synergistic trade-off through risk-driven twin-correction rolling optimization. The results show that this invention exhibits superior long-term economic efficiency and operational rationality in frequent adjustment scenarios.

Claims

1. A risk-driven twin correction control method for combined hydropower, wind power, and solar power output, characterized in that, Includes the following steps: Step 1: Receive the automatic power generation control command from the power grid and form the decision input for this cycle based on the command; Step 2: Establish a joint prediction model for instruction adjustment and deterioration of vulnerable core components, and perform online correction of the joint prediction model based on historical operating data to generate risk prediction sequences and health risk indicators. Step 3: Construct a digital twin model, use historical operating data to update the current operating status and health status of vulnerable core components of the hydropower, wind power, solar power and pumped storage power station system, and predict the operating status of the system in the future preset time domain; Step 4: Based on the current operating status and the health status of core components, construct a risk-driven energy consumption-lifetime coupled objective function, and obtain the optimal control command through rolling optimization; Step 5: Under the action of the optimal control command, update the current operating status of the hydropower, wind power, solar power and pumped storage power station system, and predict the operating status of the system in the future preset time domain. Step 6: Perform a feasibility check on the simulated future operating state and generate operating constraints. If the check is not met, trigger constraint update and return to the rolling optimization step to continue optimization; if the check is met, trigger the optimal control command shaping and issuance. Step 7: Generate maintenance strategies based on the generated risk prediction sequences, and write back the maintenance information to update the historical operation data to achieve integrated closed-loop iteration of maintenance and control strategies.

2. The risk-driven twin correction control method for combined hydropower, wind power, and solar power output according to claim 1, characterized in that, Step 2 includes: Step 21, in each sampling period Within, obtain the operating state vector of the wind-solar combined power system. Control command vector Health status vector of vulnerable core components And the digital twin model parameter vector available at the current cycle prediction start time. ; Step 22, construct the simultaneous prediction model, expressed as: ， ， ， In the formula, Indicates time-based Information on the future Predicted values ​​of the operating state vector of the combined wind and solar power system; Indicates time-based Information on the future Candidate control command vector for each step; Indicates the future number The rate of health degradation of easily damaged core components; Indicates time-based Information on the future Predicted values ​​of the health status vector of vulnerable core components; Indicates time Available model parameter vectors; Indicates the sampling period; This represents the system state transition mapping function driven by control commands; This represents a health degradation mapping function that is jointly determined by the operating state, control commands, current health state, and model parameters. Step 23: Obtain the system operating state prediction sequence covering the rolling prediction window by simultaneously establishing the prediction model. Health status prediction sequence ; Step 24, Define the risk mapping function , used to represent the future 1 The health risk prediction value for each step is expressed as: ， In the formula, Indicates the future number The dimensionless health risk prediction value of the step; Represents a health state vector The Middle Predicted values ​​for each component; Indicates the first The minimum health boundary allowed for each health state component; when When, the corresponding component risk is 0; when At that time, the corresponding component risk increases monotonically as health status declines; Step 25: To unify the risks within the rolling domain to a single risk level, define the health risk indicator for the current sampling period as follows: ， In the formula, This indicates the health risk indicators for the current sampling period; This indicates the rolling forecast length, i.e., the number of steps considered forward; Step 26: To ensure that measurement consistency and risk consistency both participate in online correction, a reference risk level is defined, and an online correction residual vector is constructed, expressed as: ， In the formula, This represents the measurement residual weighting matrix. Indicates the digital twin model at time... The predicted output for the running measurement vector, , Represents the measurement output matrix; Indicates the measurement bias term; This represents the risk residual weighting coefficient; This indicates the health risk indicators for the current sampling period; Indicates the reference risk level; Indicates time The running measurement vector is represented as: ， In the formula, This indicates the measured value of the combined output power; This indicates the measured value of grid-connected power deviation; This indicates the measured value of the adjustment rate; This indicates the temperature measurement value of the vulnerable core component; Step 27: Update the digital twin model parameter vector using a recursive approach. This yields the unconstrained update parameter vector for the next time step. , is represented as: ， ， In the formula, This represents the intermediate parameter vector at the next time step without any boundary constraints applied. Represents the online correction gain matrix; Represents the fundamental gain matrix; Indicates the level of risk A variable gain adjustment function is used to adaptively adjust the parameter update magnitude according to the current risk level; Then, a preset boundary constraint is applied to the unconstrained update parameter vector to obtain the digital twin model parameter vector for the next time step. , is represented as: ， in, Represents the lower bound vector of the parameters; Represents the upper bound vector of the parameters; This represents the element-wise limiting operator, used to constrain the parameter vector within a preset allowable range. The element-wise limiting operator satisfies: ， In the formula, The dimension of the parameter vector of the digital twin model; Represents unconstrained update of parameter vector The One component; Indicates the first The allowed lower bound of each model parameter; Indicates the first The upper bound of each model parameter; This represents the parameter vector of the digital twin model at the next time step after applying boundary constraints. The One component; Step 28, after boundary constraints Feedback is fed into the joint prediction model for rolling prediction in the next sampling period.

3. The risk-driven twin correction control method for combined hydropower, wind power, and solar power output according to claim 2, characterized in that, Step 3 includes: Step 31: Construct a digital twin model comprising three elements: device operating status, health status, and measurement mapping. The device operating status includes combined output power status, regulation rate status, operating mode status, and temperature status. The discrete-time state evolution relationship of the digital twin model is expressed as follows: ， In the formula, The extended state vector of the digital twin is used to characterize the operating status and health status of vulnerable core components of hydropower, wind power, solar power and pumped storage power station systems. Represents the state transition matrix; Represents the control input matrix; Indicates the bias term; Step 32: To establish consistency constraints between the twin state and the measured data, define the running measurement vector. satisfy: ， In the formula, This is the measurement error term; Step 33: Based on the posterior state estimate of the previous time step, the control command of the previous time step, and the historical operation data, construct the prior state estimate of the current time step. Reuse time Running measurement vector By fusing and correcting the prior state estimates, we obtain the posterior state estimate vector at the current time step. , is represented as: ， In the formula, Indicates time Prior state estimation vector before measurement fusion; Indicates time The posterior state estimation vector after measurement fusion; Indicates time The posterior state estimation vector; Indicates time The control command vector; Indicates time The parameter vector of the digital twin model; Represents the state fusion gain matrix; Step 34, using the posterior state estimation vector at the current time. As an initial condition for rolling prediction, in the future control command sequence Given the conditions, forward extrapolation is performed according to the state evolution relationship of the digital twin model to obtain the future augmented state prediction sequence and the corresponding future operational measurement prediction sequence within the rolling prediction window, which are expressed as follows: ， ； In the formula, The state evolution function of the digital twin model is used to generate the prior state estimate at the current time step based on the posterior state estimate of the previous time step, control commands, and model parameters. Step 35: Extract the healthy state sub-vector from the extended state vector, denoted as: ,in, Select a matrix for health status.

4. The risk-driven twin correction control method for combined hydropower, wind power, and solar power output according to claim 3, characterized in that, Step 4 includes: Step 41, define the rolling optimization decision variables, represented as: ,in, The sequence of control commands to be optimized. For at any time Calculate and plan for future moments The vector of control instructions to be executed; Step 42, construct the tracking performance cost term, represented as: ， In the formula, For reference target output, It is a positive semidefinite weight matrix; The energy loss cost term is constructed and represented as follows: ， In the formula, For the predicted active power at future moments, To predict the power consumption of auxiliary equipment, For equivalent efficiency, Indicates the sampling period; The construction lifetime-switching joint cost term is represented as: ， In the formula, These are dimensionless weighting coefficients. This is a start / stop switching indicator, with a value of 0 or 1. This is the predicted equivalent temperature value for vulnerable core components. It is a norm 2; Step 43, construct the risk-driven twin correction energy consumption-lifetime coupling objective function, expressed as: ， In the formula, , , These are dimensionless weighting coefficients.

5. The risk-driven twin correction control method for combined hydropower, wind power, and solar power output according to claim 4, characterized in that, Step 4 also includes: Step 44, construct the set of vectors of inequality constraint functions. The joint modeling and risk assessment device outputs a rolling domain risk sequence and defines the risk level for the current period. Construct a risk-tightened feasible region, denoted as: ， In the formula, This corresponds to the upper bound vector of the constraint. The mapping coefficient from risk to constraint margin; Introducing risk smoothing value Risk of price limits , respectively represented as: ， ， In the formula, For smoothing coefficients, For bounded operators, This is the upper limit threshold for risk. Step 45: Employ the rolling shift operator. The initial candidate control sequence is generated as follows: ， In the formula, The optimal control command sequence obtained in the previous sampling period is indicated by the superscript. Indicates the optimal solution; Step 46, in By performing a local quadratic transformation on the objective function in the vicinity, a quadratic approximation is obtained, which is expressed as: ， In the formula, It is the control sequence increment, representing the adjustment amount relative to the operating point. It is the gradient vector of the objective function with respect to the control sequence. It is a symmetric positive semi-definite second-order curvature approximation matrix; superscript Represents the transpose of a vector; Introducing the trust region radius By limiting the applicability of the quadratic approximation and regularizing the curvature matrix, the objective function is approximated as follows: ， In the formula, is the regularization coefficient, taking a non-negative value. It is the identity matrix; Step 47: Solve the sequential quadratic programming subproblem within the risk-tightening feasible region to obtain the optimal control increment, expressed as: ， In the formula, It is the optimal solution to the subproblem; When a subproblem becomes infeasible, a deterministic backoff strategy is implemented: the trust region radius is adjusted. Update it to its current value and trust region shrinkage coefficient The product of the regularization coefficients and the regularization coefficients Update it to its current value and regularization amplification factor. The product of these is then reconstructed and the subproblems are solved until a feasible solution is obtained. Step 48: Transform the subproblem solution into a new candidate control sequence, and update the magnitude using risk level scheduling, expressed as: ， In the formula, It is the updated control sequence; It is the step size coefficient, which controls the step size used in each iteration. The proportion; Maximum base step size; Step 49: Output the optimal instruction and proceed to the next time step. The closed loop will use the first control value of the updated sequence as the instruction to be issued, represented as: ， In the formula, It is a moment The optimal control command; This indicates the first step of taking the control sequence.

6. The risk-driven twin correction control method for combined hydropower, wind power, and solar power output according to claim 5, characterized in that, Step 7 includes: Step 71, to predict the risk sequence As input, the maintenance trigger strength is calculated using the following formula: ， In the formula, To maintain the evaluation window length and satisfy , These are dimensionless weighting coefficients. Indicates the sampling sequence number The future number is obtained based on online correction parameters and digital twin predictions. Step risk prediction value, risk threshold To the maximum allowable risk, For positive part operators, satisfying ; Step 72, Define the set of candidate maintenance actions Each action within the set Given a defined maintenance strategy, the optimal maintenance action satisfies: ， In the formula, Indicates the action The maintenance cost This indicates the cost of the shutdown. This represents the residual risk sequence after the action is performed. This is the risk trade-off coefficient; Step 73, Maintenance Action At any moment Once the maintenance is completed, the system will write the maintenance action type, execution timestamp, key state quantities before and after execution, and corresponding risk sequences into the historical operation data, and simultaneously update the current state and health status of the device, serving as the unified starting point data for the next sampling cycle of joint modeling and risk assessment, digital twin state prediction, constraint tightening, and rolling optimization.

Images