Agc hierarchical optimization method and system based on wind power error modeling

Abstract

The application discloses an AGC hierarchical optimization method and system based on wind power error modeling, and belongs to the field of power system dispatching optimization.The method comprises the following steps: constructing a wind power prediction error model and a wind power prediction error dynamic probability boundary model; constructing a two-stage robust optimization model comprising spare capacity configuration and real-time dispatching of unit output; and constructing an AGC hierarchical control architecture comprising a strategic decision layer, a coordination optimization layer and an execution response layer, wherein the strategic decision layer calculates a spare capacity configuration strategy of each unit based on the two-stage robust optimization model as a boundary constraint of the adjustment capacity of each unit in the coordination optimization layer, the coordination optimization layer calculates unit adjustment instructions of each unit with the aim of minimizing area control error (ACE) and adjustment cost, and the instructions are sent to the execution response layer for instruction execution.The application can reduce invalid spare allocation, shorten frequency recovery time and improve adjustment quality, and is helpful to guarantee the economy and stability of the system under high uncertainty.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and control technology, specifically relating to an AGC hierarchical optimization method and system based on wind power error modeling. Background Technology

[0002] Traditional backup planning methods rely on simplified distribution assumptions, making it difficult to characterize the asymmetry and heavy-tailed characteristics of wind power output errors; while existing AGC strategies (such as fixed-parameter PID) are insufficient in terms of black-box nature and interpretability, leading to an imbalance between regulation economy and stability.

[0003] Chinese patent application CN119742770A discloses an automatic generation control method and system for power systems that supports online parameter identification and distributed model prediction. It integrates wind power, photovoltaic (PV), and battery energy storage systems into the power system, coordinating output with traditional thermal power units for frequency regulation. However, it still has the following shortcomings: the prediction stage of the core control model (Distributed Model Predictive Control, DMPC) does not explicitly and meticulously consider the uncertainty of prediction errors for wind and PV output. This patent uses DMPC based on a state-space model for rolling optimization. Although DMPC itself has optimization capabilities, the weight matrices (such as Q, R, H) in its optimization problem need to be pre-set, and the mapping relationship between the complex optimization calculation process inside the controller and the final control command output is relatively opaque to operators, exhibiting certain "black box" characteristics. When system conditions change or the control effect does not meet expectations, the intuitiveness of adjusting these parameters or understanding the controller's decision logic is insufficient, posing challenges to debuggability and interpretability.

[0004] Current research has significant shortcomings in dynamic prediction error modeling and backup reliability assessment, and innovative solutions are urgently needed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a power system optimization method and system based on hierarchical control of multi-source automatic generation control (AGC) using wind power prediction error modeling.

[0006] The present invention adopts the following technical solution.

[0007] In a first aspect, this invention discloses a hierarchical optimization method for AGC based on wind power error modeling, comprising the following steps: Step 1: Construct a wind power prediction error model and a dynamic probability boundary model for wind power prediction error; Step 2: Construct a two-stage robust optimization model that includes reserve capacity configuration and real-time scheduling of unit output, and construct the power balance constraints of the two-stage robust optimization model based on the dynamic probability boundary model of wind power prediction error. Step 3: Construct an AGC hierarchical control architecture including a strategic decision-making layer, a coordination and optimization layer, and an execution response layer. The strategic decision-making layer calculates the reserve capacity configuration strategy of each unit based on the two-stage robust optimization model, which serves as the boundary constraint for the regulation capacity of each unit in the coordination and optimization layer. The coordination and optimization layer calculates the unit regulation instructions of each unit with the goal of minimizing the regional control error ACE and regulation cost, and issues them to the execution response layer for instruction execution.

[0008] More preferably, In step 1, the wind power prediction error model is constructed based on the Weibull-Beta mixed distribution, and the mixed distribution is sampled. The dynamic probability boundary model of wind power prediction error is obtained based on the quantile statistics of the sampling results.

[0009] More preferably, In step 1, the parameters of the wind power prediction error model are updated based on the wind power prediction error data for different operating scenarios, so as to obtain the wind power prediction error model and the corresponding dynamic probability boundary model of wind power prediction error under different operating scenarios; the operating scenarios include different regions. The dynamic probability boundary model for wind power prediction error is as follows:

[0010] in, For time period t ,area r The lower bound of the confidence interval for wind power prediction error; For time period t ,area r The upper bound of the confidence interval for lower wind power prediction error; The Weibull-Beta mixture distribution is fitted based on the wind power prediction error data under the current operating scenario; pl , pu These are the lower and upper quantiles of the confidence interval, respectively. , For each of the mixed distributions The quantile function, which takes the lower and upper quantiles of the confidence interval, is used to output the upper and lower bounds of the confidence interval for wind power prediction error in the corresponding scenario.

[0011] More preferably, In step 2, the two-stage robust optimization model includes the following steps: In the reserve capacity configuration stage, the optimal reserve capacity configuration scheme is calculated with the goal of minimizing the expected reserve cost; in the real-time scheduling stage of unit output, based on the optimal reserve capacity configuration scheme, the optimal real-time scheduling scheme is obtained under the corresponding wind power prediction error scenario by probabilistically relaxing the system power balance constraints at a set confidence level, with the goal of minimizing the adjustment cost and risk premium cost. The two-stage robust optimization model is solved through the following iterative process: the adjustment cost and risk premium cost corresponding to the optimal solution in the real-time scheduling stage of unit output are mathematically expected under all wind power prediction error scenarios and then superimposed as feedback terms onto the objective function of the reserve capacity configuration stage to update the reserve capacity configuration scheme. The iteration is continued until the reserve capacity configuration scheme converges.

[0012] More preferably, During the real-time scheduling phase of unit output, the probabilistic relaxation of the system power balance constraints under a set confidence level is specifically defined as follows:

[0013] in, This is an uncertain variable, representing the wind power prediction error; For the first g The planned output of the conventional generating units; To account for uncertain variables The next g Real-time adjustment of the unit; To account for uncertain variables The actual output of wind power after that; This represents the system load. The confidence level is the chance constraint.

[0014] More preferably, In step 3, the AGC hierarchical control architecture also includes an uncertainty control layer to support the coordinated regulation of frequency control between regions and the dynamic weight configuration of multi-source backup resources. The uncertainty control layer uses the backup capacity configuration strategy of each unit calculated by the strategic decision layer as the constraint boundary. Through robust optimization, it generates the regional control error ACE optimization target value, unit backup call command and the weight coefficient of renewable energy backup and thermal power regulation backup for each region and sends them to the coordination optimization layer to reconstruct the optimization target of the coordination optimization layer.

[0015] More preferably, After receiving the AGC adjustment command from the coordination and optimization layer, the execution response layer tracks the AGC adjustment command using an adaptive PID controller based on linear decision rules. The adaptive PID controller dynamically adjusts its gain coefficient based on the current system's ACE deviation, ACE rate of change, and wind and solar power output rate of change. Specifically:

[0016] in, The base gain is used as the initial reference for control through offline tuning; , , (i=1,2,3) are decision coefficients, which are trained and tuned offline based on historical system data; The current ACE bias, The ACE rate of change The rate of change in the output of new energy sources.

[0017] Secondly, the present invention discloses an AGC hierarchical optimization system based on wind power error modeling based on the method, including a wind power prediction error dynamic probability boundary model construction module, a two-stage robust optimization model construction module, and an AGC hierarchical control architecture construction module. The module for constructing the dynamic probability boundary model of wind power prediction error is used to construct the wind power prediction error model and the dynamic probability boundary model of wind power prediction error. A two-stage robust optimization model construction module is used to construct a two-stage robust optimization model including reserve capacity configuration and real-time scheduling of unit output, and to construct the power balance constraints of the two-stage robust optimization model based on the dynamic probability boundary model of wind power prediction error. The AGC hierarchical control architecture construction module is used to construct an AGC hierarchical control architecture including a strategic decision-making layer, a coordination and optimization layer, and an execution response layer. The strategic decision-making layer calculates the reserve capacity configuration strategy of each unit based on the two-stage robust optimization model, which serves as the boundary constraint for the regulation capacity of each unit in the coordination and optimization layer. The coordination and optimization layer calculates the unit regulation instructions of each unit with the goal of minimizing the regional control error ACE and regulation cost, and issues them to the execution response layer for instruction execution.

[0018] Thirdly, the present invention provides a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of the first aspects of the present invention.

[0019] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects of the present invention.

[0020] The beneficial effects of this invention are that, compared with the prior art, This invention significantly improves the reliability of reserve capacity configuration by introducing a Weibull-Beta hybrid distribution to model wind power prediction errors with high precision. In scenarios where the penetration rate of new energy reaches 50%, it achieves synergy between pre-emptive reserve configuration and in-process adjustment based on a two-stage robust optimization model, resulting in a significant reduction in total cost compared to traditional methods. Simultaneously, the multi-source AGC hierarchical control architecture, improved based on model predictive control (MPC) and linear decision rule (LDR), enhances the ability to suppress regional control error (ACE) fluctuations, significantly shortens the frequency deviation recovery time, and strengthens the dynamic stability of the system under high uncertainty. Furthermore, the structural transparency of LDR-PID and MPC greatly improves the interpretability of the control strategy, facilitating integration with actual dispatching systems, reducing debugging complexity and operation and maintenance costs, and forming a comprehensive benefit of synergistic optimization of economy, stability, and interpretability. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the AGC hierarchical optimization method based on wind power error modeling of the present invention; Figure 2 This is a power grid diagram of the IEEE 118-node system used in the example simulation of Embodiment 2 of the present invention; Figure 3 The actual and upper limit of wind power output on a certain day in Texas, USA, according to Embodiment 2 of the present invention; Figure 4 This is a graph showing the normalized prediction error and the Weibull-Beta mixture fitting in Embodiment 2 of the present invention. Figure 5 The following is a daytime curve of the backup capacity allocation of AGC units under different penetration rates in Embodiment 2 of the present invention, wherein (a) is the backup capacity allocation of AGC units under the scenario of 10% renewable energy penetration, (b) is the backup capacity allocation of AGC units under the scenario of 30% renewable energy penetration, and (c) is the backup capacity allocation of AGC units under the scenario of 50% renewable energy penetration. Figure 6 This is a comparative analysis diagram of two backup allocation methods under different permeability rates in Embodiment 2 of the present invention; Figure 7This is a cost comparison chart of two methods in Embodiment 2 of the present invention, where (a) is a cost comparison of Scheme 1 and 2 at a penetration rate of 10%, (b) is a cost comparison of Scheme 1 and 2 at a penetration rate of 30%, and (c) is a cost comparison of Scheme 1 and 2 at a penetration rate of 50%. Figure 8 This is a disturbance diagram applied to simulate power grid fluctuations in Embodiment 2 of the present invention; Figure 9 This is a cost structure diagram of the three competitive strategies in Embodiment 2 of the present invention; Figure 10 This is a schematic diagram of three ACE control strategies in Embodiment 2 of the present invention; Figure 11 The above are frequency response curves of the three control strategies in Embodiment 2 of the present invention. Figure 12 The diagram shows the response of the joint frequency regulation command between new energy and thermal power in Embodiment 2 of the present invention, where (a) is the new energy AGC command, (b) is the thermal power AGC command, (c) is the new energy AGC response, and (d) is the thermal power AGC response. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0023] like Figure 1 As shown, this invention discloses a hierarchical optimization method for AGC based on wind power error modeling, comprising the following steps: Step 1: Construct a wind power prediction error model and a dynamic probability boundary model for wind power prediction error; Preferably, the wind power prediction error model is constructed based on the Weibull-Beta mixture distribution; those skilled in the art will understand that the error model can also be constructed based on commonly used methods in the field, such as normal distribution or nonparametric estimation.

[0024] Specifically, a wind power prediction error model based on a Weibull-Beta hybrid distribution is constructed by linearly weighting the Weibull and Beta distributions, which can be expressed as:

[0025] in, This represents the standardized wind power prediction error. The weighted coefficients are mixed and satisfy the following conditions: This is used to achieve flexible expression of bimodal features; Let Weibull's probability density function be the error of wind power prediction. Let be the scale parameter of the Weibull distribution. For shape parameters; Let be the probability density function of the Beta distribution of wind power prediction error. The first shape parameter of the Beta distribution controls the degree of left skewness and the thickness of the left tail. This is the second shape parameter of the Beta distribution, which controls the degree of right skewness and the thickness of the right tail.

[0026] The Weibull-Beta mixture distribution is sampled, and the dynamic probability boundary model of wind power prediction error is obtained based on the quantile statistics of the sampling results. Furthermore, by updating the parameters of the wind power prediction error model based on the wind power prediction error data for different operating scenarios, wind power prediction error models and corresponding dynamic probability boundary models for wind power prediction errors under different operating scenarios can be obtained; the operating scenarios include different regions. The dynamic probability boundary model for wind power prediction error is as follows:

[0027] in, For time period t ,area r The lower bound of the confidence interval for wind power prediction error; For time period t ,area r The upper bound of the confidence interval for lower wind power prediction error; The Weibull-Beta mixture distribution is fitted based on the wind power prediction error data under the current operating scenario; pl , pu These are the lower and upper quantiles of the confidence interval, respectively. , For each of the mixed distributions The quantile function, which takes the lower and upper quantiles of the confidence interval, is used to output the upper and lower bounds of the confidence interval for wind power prediction error in the corresponding scenario.

[0028] Step 2: Construct a two-stage robust optimization model that includes reserve capacity configuration and real-time scheduling of unit output, and construct the power balance constraints of the two-stage robust optimization model based on the dynamic probability boundary model of wind power prediction error. The two-stage robust optimization model includes the following steps: In the reserve capacity configuration stage, the optimal reserve capacity configuration scheme is calculated with the goal of minimizing the expected reserve cost; in the real-time scheduling stage, based on the optimal reserve capacity configuration scheme, the optimal real-time scheduling scheme is obtained under the corresponding wind power prediction error scenario by probabilistically relaxing the system power balance constraints at a set confidence level, with the goal of minimizing the adjustment cost and risk premium cost. The two-stage robust optimization model is solved through the following iterative process: the adjustment cost and risk premium cost corresponding to the optimal solution in the real-time scheduling stage are expected under all wind power prediction error scenarios, and then used as feedback terms to be added to the objective function of the reserve capacity configuration stage to update the reserve capacity configuration scheme. The iteration is continued until the reserve capacity configuration scheme converges.

[0029] The risk premium cost is used to characterize the degree of deviation between the real-time scheduling decision and the baseline plan, and is determined in the following manner:

[0030] in, This is an uncertain variable, representing the wind power prediction error; For uncertain variables After implementation, the actual adjustment amount of each unit; As the baseline scheduling plan instruction; It is an L2 norm.

[0031] The probabilistic relaxation of the system power balance constraint under a set confidence level is specifically defined as follows:

[0032] in, For the first g The planned output of the conventional generating units; To account for uncertain variables The next g Real-time adjustment of the unit; To account for uncertain variables The actual output of wind power after that; This represents the system load. The confidence level is the chance constraint.

[0033] Step 3: Construct an AGC hierarchical control architecture including a strategic decision-making layer, a coordination and optimization layer, and an execution response layer. The strategic decision-making layer calculates the reserve capacity configuration strategy of each unit based on the two-stage robust optimization model, which serves as the boundary constraint for the regulation capacity of each unit in the coordination and optimization layer. The coordination and optimization layer calculates the unit regulation instructions of each unit with the goal of minimizing the regional control error ACE and regulation cost, and issues them to the execution response layer for instruction execution.

[0034] After receiving the AGC adjustment command from the coordination and optimization layer, the execution response layer tracks the AGC adjustment command using an adaptive PID controller based on linear decision rules. The adaptive PID controller dynamically adjusts its gain coefficient based on the current system's ACE deviation, ACE rate of change, and wind and solar power output rate of change. Specifically:

[0035] in, The base gain is tuned offline using the traditional Ziegler-Nichols method and used as the initial reference for control. , , (i=1,2,3) are decision coefficients, which are trained and tuned offline based on historical system data; The current ACE bias, The ACE rate of change The rate of change in the output of new energy sources.

[0036] In this invention, the AGC hierarchical control architecture may further include an uncertainty control layer to support the coordinated regulation of frequency control between regions and the dynamic weight configuration of multi-source backup resources. The uncertainty control layer uses the backup capacity configuration strategy of each unit calculated by the strategic decision layer as the constraint boundary. Based on the regional control error ACE of each region measured in real time by the coordination optimization layer, it generates the regional control error ACE optimization target value, unit backup call command and weight coefficient of renewable energy backup and thermal power regulation backup for each region through robust optimization and sends them to the coordination optimization layer to reconstruct the optimization target of the coordination optimization layer. The reconstructed optimization objective in the coordination optimization layer is specifically as follows: the coordination optimization layer uses the regional control error ACE optimization objective value as the tracking target of the regional control error ACE in the original objective function; the unit standby call instruction is used as the expected value of the unit regulation instruction in the coordination optimization layer and is incorporated into the original objective function in the form of soft constraints; the regulation cost in the original objective function is adjusted based on the weight coefficients of renewable energy standby and thermal power regulation standby.

[0037] Example 1: like Figure 1As shown, this invention discloses a hierarchical optimization method for AGC based on wind power error modeling, comprising the following steps: Step 1: Construct a wind power prediction error model and a dynamic probability boundary model for wind power prediction error; Wind power forecasting errors are mainly caused by a combination of factors, including the uncertainty of meteorological input data, insufficient accuracy of numerical weather prediction models, spatiotemporal non-uniformity of wind resources, and local disturbances in wind farms (such as topography, turbulence, and wind shear). Among these, the instability of meteorological boundary conditions is the main factor leading to medium- and long-term forecasting errors, while local micro-meteorological disturbances have a significant impact on short-term forecasting accuracy. In addition, the dynamic response characteristics and control strategies of wind turbines themselves can also cause deviations between predicted and actual output, further exacerbating the randomness and uncontrollability of the errors.

[0038] Statistically, wind power prediction errors exhibit asymmetry, large kurtosis variations, and a heavy-tailed distribution, making it difficult for traditional normality-based models to accurately characterize their true fluctuations. Therefore, this invention introduces a Weibull-Beta hybrid distribution model, combining the fundamental characteristics of wind speed distribution with a flexibly adjustable skewness structure to dynamically model wind power output prediction errors. The Weibull distribution is suitable for characterizing the non-negativity and right-skewness of wind speed distributions and is widely used in wind energy resource assessment and wind power modeling; the Beta distribution, due to its good control over kurtosis and skewness, enhances the model's adaptability to local anomalies.

[0039] In this model, the Weibull distribution is used to characterize wind speed and its fluctuations. Its good fitting performance makes it widely used in wind energy assessment and wind power modeling. Its probability density function is:

[0040] in, Wind speed is used to describe the fundamental characteristics of wind resources. The scale parameter of the Weibull distribution reflects the average level / concentration of wind speed, also known as the scale parameter; The shape parameter reflects the skewness and peak characteristics of the wind speed distribution. When b=1, it is an exponential distribution, and when b=3.67, it is an approximately normal distribution, which is suitable for the statistical characteristics of right skewness of wind speed in wind power.

[0041] Furthermore, introducing a Beta distribution as a supplement helps to finely control the tail thickness and peak position of the error distribution through shape parameters. This is achieved through weighting coefficients. , Adjusting the tail thickness and peak position, its density function is:

[0042] in, Let be a Beta function, representing the normalization constant of the Beta distribution, ensuring that the integral sum of the probability density functions is 1, and satisfying the following conditions: , It is the Gamma function; This represents the standardized wind power prediction error. The first shape parameter of the Beta distribution controls the degree of left skewness and the thickness of the left tail. This is the second shape parameter of the Beta distribution, which controls the degree of right skewness and the thickness of the right tail.

[0043] By linearly superimposing Weibull and Beta distributions, a bimodal mixture probability density function is constructed, where the mixture weights are adjustable parameters, giving the model high flexibility in expressing complex error structures. The linear combination of these distributions forms a composite distribution:

[0044] in, This represents the standardized wind power prediction error. The weighted coefficients are mixed and satisfy the following conditions: This is used to achieve flexible expression of bimodal features; Let Weibull's probability density function be the error of wind power prediction. Let be the scale parameter of the Weibull distribution. For shape parameters; Let be the probability density function of the Beta distribution of wind power prediction error. The first shape parameter of the Beta distribution controls the degree of left skewness and the thickness of the left tail. This is the second shape parameter of the Beta distribution, which controls the degree of right skewness and the thickness of the right tail.

[0045] To ensure the stability and accuracy of the modeling results, this invention first standardizes the historical wind power output prediction error sequence to eliminate the potential impact of data scale differences on the parameter estimation process. The standardization process for the wind power output prediction error sequence is as follows:

[0046] in, for t Standardized prediction error of wind power output at any given moment; for t Actual wind power output at any given moment; fort Real-time wind power output forecast; This refers to the rated power of the wind power installed capacity.

[0047] In terms of parameter estimation, the Expectation-Maximization (EM) algorithm is used to iteratively optimize the parameters of the mixed distribution to achieve the optimal fit in the sense of maximum likelihood.

[0048] in, The optimal parameter vector to be found includes all parameters to be estimated, which is the combination of parameters that maximizes the log-likelihood function; n This represents the total number of wind power prediction error samples. For the first i A standardized wind power prediction error sample; These are mixed weighting coefficients; Let be the probability density function of the Weibull distribution; Let be the probability density function of the Beta distribution of wind power prediction error.

[0049] Furthermore, to verify the model's fit, the Kolmogorov-Smirnov test method is introduced to evaluate how well the established mixture distribution approximates the actual prediction error data, ensuring that the model has good interpretability in a statistical sense.

[0050] Based on the fitting results, the Monte Carlo sampling method is further used to perform multiple simulated samplings on the mixed distribution, thereby generating prediction error confidence intervals under different time periods and regional conditions. Preferably, the boundary calculation formula is as follows:

[0051] in, This represents the lower bound of the confidence interval for wind power prediction error, and the minimum possible value of wind power prediction error at a 90% confidence level. This represents the upper bound of the confidence interval for wind power prediction error, and the maximum possible value of wind power prediction error at a 90% confidence level. is a quantile function, representing the p-th percentile value of a sequence of random variables, where p is the percentile; The wind power prediction error sequence generated by Monte Carlo sampling is a set of wind power prediction error samples obtained by sampling using the Monte Carlo method based on the Weibull-Beta mixture distribution.

[0052] Based on this, a dynamic wind power prediction error probability boundary model is constructed, which can be formally represented as:

[0053] in, t and r These represent time and region identifiers, respectively. For time period t ,area r The lower bound of the confidence interval for wind power prediction error corresponds to the lower quantile of the error at the confidence level (e.g., 5%), indicating that in this scenario, there is a corresponding probability (e.g., 90%, i.e., the selected confidence level) that the prediction error is not less than this value. For time period t ,area r The upper bound of the confidence interval for wind power prediction error corresponds to the high quantile of the error at the confidence level (e.g., 95%), indicating that in this scenario, there is a corresponding probability (e.g., 90%, i.e., the selected confidence level) that the prediction error is not greater than this value. This represents a Weibull-Beta mixture distribution fitted based on the latest data for the current time period and region. It is a quantile function. pl and pu These are the lower and upper quantiles (e.g., 5% and 95%) of the confidence interval. Boundary values ​​can be achieved by periodically updating the distribution parameters with the latest data. , The dynamic evolution of ].

[0054] This dynamic probability boundary model not only reflects the fluctuation pattern of errors within different time windows, but also characterizes the differences in error distribution between regions caused by variations in meteorological conditions and geographical features. To capture the spatiotemporal characteristics of prediction errors, the model uses a sliding time window to periodically update the distribution parameters, enabling the constructed uncertainty boundary to evolve dynamically.

[0055] Ultimately, this dynamic probabilistic boundary is incorporated into the robust scheduling model as part of the uncertainty set of wind power output, participating in the optimization calculation. By inputting the probabilistic boundary as the uncertainty set into the robust scheduling model, the power balance constraints are:

[0056]

[0057] Where G represents the number of conventional generating units. For the first g The active power output of the conventional generating unit; For wind power forecasting, the day-ahead / intra-day active power output forecast of the wind farm; This refers to wind power forecasting errors; As the lower bound of system power balance, after considering wind power prediction errors, the minimum power requirement that the total active power output of the system must meet; The upper bound of system power balance is the maximum power demand that the total active power output of the system must meet after considering wind power prediction errors. The confidence interval for wind power prediction error is the 90% confidence interval obtained from Monte Carlo sampling, which constrains the fluctuation range of wind power prediction error.

[0058] While ensuring system power balance constraints, the model enhances the resilience of the scheduling model by introducing dynamic boundaries, providing a solid probabilistic foundation and decision support for the safety and stability of actual wind power consumption and grid operation.

[0059] Step 2: Construct a two-stage robust optimization model that includes reserve capacity configuration and real-time scheduling of unit output, and construct the power balance constraints of the two-stage robust optimization model based on the dynamic probability boundary model of wind power prediction error. To address the uncertainty in reserve capacity delivery caused by wind power output forecasting errors, this invention constructs a two-stage robust optimization model for scenarios with high renewable energy penetration, aiming to improve the power system's flexible adjustment capabilities and operational safety. The model's design logic involves formulating reserve reservation and real-time adjustment strategies in stages as uncertainty information is gradually revealed, achieving an optimal balance between reserve costs and system risks.

[0060] In step 2, the two-stage robust optimization model includes the following steps: In the reserve capacity configuration stage, the optimal reserve capacity configuration scheme is calculated with the goal of minimizing the expected reserve cost; in the real-time scheduling stage of unit output, based on the optimal reserve capacity configuration scheme, the optimal real-time scheduling scheme is obtained under the corresponding wind power prediction error scenario by probabilistically relaxing the system power balance constraints at a set confidence level, with the goal of minimizing the adjustment cost and risk premium cost. The two-stage robust optimization model is solved through the following iterative process: the adjustment cost and risk premium cost corresponding to the optimal solution in the real-time scheduling stage of unit output are mathematically expected under all wind power prediction error scenarios and then superimposed as feedback terms onto the objective function of the reserve capacity configuration stage to update the reserve capacity configuration scheme. The iteration is continued until the reserve capacity configuration scheme converges.

[0061] The convergence of the backup capacity configuration scheme is specifically defined as follows: when the deviation norm between the backup capacity configuration schemes calculated in two adjacent iterations is less than a preset convergence threshold, and / or the absolute value of the difference between the objective function values ​​of two adjacent iterations is less than a preset convergence threshold, the iteration process is determined to have converged if either condition is met, and the current backup capacity configuration scheme is output as the optimal solution.

[0062] The first phase primarily addresses the "deterministic problem of pre-emptive reserve capacity allocation," namely, how to rationally formulate reserve capacity strategies for various reserve units based on known wind power forecast information and error boundaries before wind power output is realized and the system is operational. The goal of this phase is to minimize expected reserve costs and, considering the uncertainties of wind power, ensure the system has the basic adjustment capability to cope with potential fluctuations through conservative reservations. Its core decision variable is the pre-allocation of reserve capacity, which directly relates to the deployment efficiency of reserve resources at the spatial and turbine levels.

[0063] The second phase addresses the "responsiveness of in-process operation adjustments," specifically how to adjust the system's operating state to ensure safety and stability after actual wind power output is achieved and specific output deviations occur. This phase introduces a chance constraint mechanism, relaxing key operational constraints (such as system power balance and turbine output limits) at a set confidence level to control the probability of risk occurrence, thereby achieving the minimum adjustment cost and risk premium cost under controllable risk. This phase reflects the system's dynamic response capability to wind power output disturbances. The adjustment cost and risk premium corresponding to its optimal solution will be fed back as part of the objective function into the expected cost calculation of the first phase, thus guiding the development of a backup reserve scheme that balances economy and robustness in the first phase.

[0064] The two-stage robust optimization model can be specifically expressed as:

[0065] in, , is the set of decision variables for the first stage. , These represent the reserve capacity (unit: MW) for each generating unit. X for x The feasible domain is determined by the unit's maximum / minimum technical output; For uncertain variables, representing the error of actual wind power output deviating from the predicted value, the range of its value is defined by the uncertainty set formed by the dynamic probability boundary of wind power prediction error generated in step 1; for The set of uncertainties; This is the optimization objective in the second phase; , These are the cost coefficients for reserving upper and lower reserves in the first phase (unit: yuan / MW). , These are the cost coefficients for the actual and downward adjustment commands of the generating units in the second phase; , which are the decision variables for the second stage. , Uncertain variables Once implemented, each unit will actually issue and receive adjustment commands (unit: MW). for The feasible domain is limited by the reserved capacity in the first phase. ; To account for uncertain variables The actual output of wind power after that; For the first g The planned output of the conventional generating units; , The respective g Minimum and maximum output of the unit; Uncertain variables Once implemented, the real-time adjustment amount of the g-th unit; This represents the system load. This is the risk aversion coefficient, used to weigh adjustment costs against systemic risk; The risk premium measures the actual adjustment orders. Deviation from baseline scheduling plan To what extent, Based on day-ahead or intraday wind power and load forecasts, a baseline output plan for each generating unit is obtained by solving a deterministic economic dispatch model. This plan is used to penalize excessive or frequent regulatory actions. ; For probability operators, For the confidence level of the opportunity constraint, e.g. =0.05 indicates that the probability of system power imbalance is allowed to be no more than 5%; , These are the maximum downward adjustment rate and the maximum upward adjustment rate of the g-th unit, respectively.

[0066] The two-stage robust optimization model's inputs include: wind power / load forecast values ​​and the error uncertainty set generated in step 1. Operating parameters of all AGC units , , , Cost coefficient , , , System load and risk parameters Confidence level of opportunity constraints .

[0067] The output of the two-stage robust optimization model is: the optimal backup reservation scheme obtained in the first stage. And for any wind power implementation scenario At that time, the real-time adjustment strategy in the second stage .

[0068] Continuous set of uncertainties This is typically difficult to handle directly. This embodiment employs a polyhedral approximation method for uncertain sets to solve the problem, i.e., from the set... Selected from A representative extreme scenario (such as maximum positive error, maximum negative error, and several intermediate scenarios). And the original set is approximated by the convex hull of these scenes: Through this transformation, the original two-stage robust optimization problem can be equivalently rewritten as a large-scale mixed-integer linear programming (MILP) problem, which can be solved efficiently using commercial solvers. This method is widely used in the field of robust optimization and will not be elaborated upon here.

[0069] Further preferably, to more accurately assess whether the reserve capacity of renewable energy sources such as wind power has reliable delivery capability in actual operation, this invention proposes a Dynamic Credible Capacity (DCC) index. This index aims to quantify the probability of achieving reserve capacity at a predetermined confidence level under conditions of significant uncertainty in wind power output, thereby assisting in the formulation of more robust reserve configuration strategies. This index is based on the cumulative distribution function (CDF) under a mixed probability distribution to inversely deduce the reserve response probability, defined as follows:

[0070] in, For unit i During the period t Dynamic trusted capacity; The infimum; To set the confidence level, As a backup response amplitude, For unit i During the period t The actual wind power output (following a Weibull-Beta mixed distribution).

[0071] The dynamic reliability capacity index is added as a constraint to the first-stage model, and the specific constraint formula is as follows:

[0072] in, For backup unit During the period Reserved capacity; For unit i During the period tThe actual wind power output (generated based on a probabilistic model of actual new energy output generated from a Weibull-Beta hybrid distribution); this constraint ensures that at the confidence level... The reserved backup is reliable and callable.

[0073] By incorporating DCC into robust optimization constraints, the deliverability of the backup plan is significantly improved.

[0074] Step 3: Construct an AGC hierarchical control architecture including a strategic decision-making layer, a coordination and optimization layer, and an execution response layer. The strategic decision-making layer calculates the reserve capacity configuration strategy of each unit based on the two-stage robust optimization model, which serves as the boundary constraint for the regulation capacity of each unit in the coordination and optimization layer. The coordination and optimization layer calculates the unit regulation instructions of each unit with the goal of minimizing the regional control error ACE and regulation cost, and issues them to the execution response layer for instruction execution.

[0075] Compared to traditional AGC control scenarios primarily based on thermal power, AGC control in renewable energy scenarios differs significantly in terms of control objectives, response mechanisms, and uncertainty handling. Firstly, traditional AGC systems primarily utilize high-inertia thermal power units, which offer stable output, slow adjustment response, and highly predictable control models, making them suitable for linear control strategies employing static parameter settings. However, in systems with a high proportion of renewable energy, resources such as wind and solar power exhibit discontinuous output, frequent fluctuations, and weak inertia support capabilities, resulting in a significantly faster rate of frequency change. This necessitates systems with higher levels of rapid response and frequency stability.

[0076] Secondly, the controllability of controlled objects in new energy scenarios is weak and the prediction error is large, making it difficult for traditional AGC methods that rely on static setpoints and fixed controller parameters to adapt to actual operating conditions. In addition, in terms of coordination mechanisms, traditional AGC is mostly for isolated regional control, and in the context of high penetration of new energy, new problems such as source-load spatial decoupling and backup cross-regional allocation need to be addressed.

[0077] The hierarchical control architecture of this invention divides each level into hourly, second-level, and sub-second-level gradient response systems according to the time scale and response requirements of the control task. The specific control logic of the three levels is as follows: 1. Strategic Decision-Making Layer (Hourly Level): At the strategic decision-making layer, the grid dispatching EMS (Energy Management System) integrates wind and solar power output forecasts, load forecasts, and real-time system operating status to construct a dynamic reserve allocation model. This model comprehensively considers the economics, response speed, and physical constraints of regulation resources to form a reserve capacity allocation strategy for future rolling time periods. This layer corresponds to the two-stage robust optimization model in step 2, with an operating cycle of 15 minutes to 1 hour. Its inputs are ultra-short-term wind power / load forecasts, system network topology, and unit status. Its output is the globally optimal reserve capacity pre-allocation plan for several future time periods. The plan defines the upper and lower limits of the standby capacity that each AGC unit can call upon at different times, and issues them to the coordination and optimization layer as boundary constraints on its adjustment capabilities.

[0078] 2. Coordination and Optimization Layer (Second-level): The coordination and optimization layer dynamically optimizes the allocation of AGC commands based on real-time data within the rolling time domain, maximizing the suppression of regional control error (ACE) while improving the grid-connected utilization rate of renewable energy. This layer continuously corrects short-term planning and execution deviations by introducing a rolling forecasting and feedback correction mechanism, enabling rapid tracking of renewable energy fluctuations. A rolling time-domain model predictive control (MPC) framework is adopted, with a cycle typically of 4-5 seconds. Within each rolling cycle: the model inputs are the backup plan x received from the strategic layer, real-time system measurements (frequency, ACE, unit output), and the ultra-short-term (future 1-5 minutes) wind and solar power forecast range updated based on the error model update from step 1.

[0079] With the objective of minimizing the regional control error ACE and adjustment cost within a finite future time domain, online optimization is performed. The specific model of the coordinated optimization layer is as follows:

[0080] in, for t Time-based regional control error, k For the current moment, This refers to the unit control command vector for each unit in the future time domain, i.e., the command vector for each unit in... t Adjustment amount at any time, The 2-norm square of the AGC adjustment command vector represents the total adjustment cost; N To predict the time domain length, λ is the weighting coefficient, used to assess the suppression effect and adjustment cost of weighted ACE. , The first g Taiwanese crew t time, t AGC adjustment command at time -1 For the first g Taiwanese crew t Current output at any given moment; , The first g The lower and upper limits of the output of the generator set; For the first g The ramp rate of the Taiwanese generator set. To control the time step of the cycle, This is the frequency deviation coefficient.

[0081] Solving the above optimization problem yields the optimal control sequence in the future time domain, but only the first control variable is considered. The data is then sent to the execution layer. In the next cycle, the optimization process is restarted based on the new measurement data, allowing for feedback correction.

[0082] 3. Execution Response Layer (Sub-second): The execution response layer specifically operates on various types of control units, requiring them to implement rapid and autonomous responses based on their own capabilities and operating status upon receiving AGC control commands, ensuring the feasibility and timeliness of control actions. This layer receives commands from the coordination layer. And it achieves fast and accurate tracking through an adaptive PID controller based on linear decision rule (LDR) installed on each AGC unit.

[0083] To enhance the adaptive adjustment capability of AGC systems in scenarios with high penetration of new energy sources, this invention improves the traditional PID control method based on linear decision rules. By introducing system state information to dynamically adjust the PID controller parameters, the control strategy can flexibly adapt to system operating conditions, effectively enhancing the stability and robustness of frequency regulation.

[0084] The core idea of ​​linear decision rules is to use key state variables of the current system, such as ACE, ACE change rate, and wind and solar power output change rate, as the basis for decision-making, and dynamically adjust the gain coefficient of the PID controller through a preset linear relationship. The controller parameters are dynamically adjusted according to the real-time state of the system. Assuming... Let the proportional, integral, and derivative gains be respectively. Then, the adaptive gain adjustment based on LDR can be expressed as:

[0085] in, Based on the gain, (i=1,2,3) are decision coefficients. Indicates the current ACE deviation. The ACE rate of change The rate of change in the output of new energy sources.

[0086] In this invention, the base gain of the PID controller , , The traditional Ziegler-Nichols method was used for offline tuning as the initial reference for control; decision coefficients... , , (i=1,2,3) are determined through offline training: First, historical system operation data are collected, and a linear regression model is constructed with ACE deviation, ACE change rate, and new energy output change rate as inputs and PID gain as output. The optimal coefficients are obtained by solving the least squares method. In the embodiment, based on the IEEE-118 node system simulation, the preferred range of basic gain values ​​is... , , The typical values ​​are 1.2, 0.3, and 0.1, respectively; the preferred ranges and typical values ​​of the other decision coefficients are shown in Table 1.

[0087] Table 1. Optimal ranges and typical values ​​of decision coefficients for adaptive gain adjustment based on LDR.

[0088] These parameters can be obtained through offline training using historical data, enabling the system to achieve excellent grid frequency control performance evaluation indicators CPS1 and ACE suppression performance under high renewable energy penetration rates.

[0089] This parameter setting can improve the CPS1 performance evaluation index of the power grid frequency control and reduce ACE fluctuations, and has good engineering feasibility.

[0090] Compared to the traditional fixed-parameter PID control method, the LDR method gives the controller a certain "sensing" and "judgment" capability, enabling it to optimize and adjust its behavior in real time according to the magnitude and trend of system disturbances, thereby achieving more refined and efficient frequency control.

[0091] In designing the control strategy, the differences between new energy control resources and traditional thermal power resources in terms of response speed, inertia characteristics, and control capabilities were considered. New energy equipment has the advantage of rapid response in the early stages of frequency disturbances, making it suitable for high-sensitivity control tasks; while thermal power units, due to their greater mechanical inertia and stronger continuous support capabilities, provide stable power support in the later stages of frequency adjustment. LDR-PID effectively coordinates the action paths of the two types of resources by dynamically adjusting parameter weights under different operating conditions, making the frequency control process more hierarchical and continuous.

[0092] Furthermore, the algorithm maintains stable system operation under small deviations, avoiding system oscillations caused by frequent adjustments; when large disturbances occur, it enhances the system's dynamic response and robustness by increasing the controller gain. Overall, this improved PID algorithm features a simple structure, low computational cost, and strong engineering feasibility, and exhibits good compatibility with the MPC hierarchical control structure.

[0093] In this invention, based on the hierarchical control system, an uncertainty control layer based on a robust optimization mechanism for uncertainty disturbances can be further established to support the coordinated adjustment of frequency control between regions and the dynamic weight allocation of multi-source backup resources. Against the backdrop of significantly enhanced volatility in new energy sources, the ACE changes between regions due to differences in resource structure and load distribution tend to become uncoordinated, necessitating mechanism design to achieve cross-regional frequency response coordination.

[0094] The control mechanism proposed in this invention incorporates factors such as wind and solar power output prediction errors and uncertainties in tie-line switching power into a unified uncertainty modeling framework, and constructs regional ACE flexible coupling constraints to dynamically adjust the allowable range of frequency deviations between regions. When frequency disturbances are small, the region mainly relies on local resources to complete the adjustment, avoiding unnecessary resource linkage; when the deviation exceeds a certain threshold, the regional coupling strength is automatically enhanced, guiding frequency regulation from local to system-level response, and achieving the goal of system-level frequency coordinated stability.

[0095] Considering the impact of changes in system inertia level, wind and solar power volatility, and the response capabilities of various resources on reserve demand, an adaptive adjustment mechanism for the reserve responsibility ratio of new energy and thermal power is constructed. Under operating conditions with insufficient system inertia, the reserve ratio undertaken by thermal power resources is increased to enhance the system's inertial response capability to frequency disturbances. When the system stability margin is sufficient, the proportion of new energy regulation is reasonably increased to alleviate the pressure on conventional units and improve the utilization rate of new energy.

[0096] The uncertainty control layer is expressed as the following mathematical model under the unified robust framework:

[0097] in, The system's comprehensive decision vector is a multi-dimensional column vector, including the ACE optimization target value of each region, the standby call instructions of each unit, and the weight coefficients of renewable energy standby and thermal power regulation standby. It is the core solution object of the robust optimization model. For regional sets (such as power grid zones, such as North China, East China, etc.). i For regional indexing; For unit Spare configuration quantity, , Units k Minimum and maximum standby configuration; For a set of units; For the region The ACE weighting coefficients can be determined based on the equivalent inertia of the region. The settings include adjusting the capacity ratio and historical ACE statistical characteristics. In this embodiment, it is preferable to... That is, the smaller the inertia of a region, the greater the weight of its ACE in the global target, so as to suppress its frequency fluctuations first. This is the coupling relaxation penalty coefficient; Allocate costs for reserves; , They are respectively regions i With the region j ACE value; For the region i and j The maximum permissible ACE deviation threshold (unit: MW); This represents the relaxation amount of ACE flexible coupling between regions; Let be the set of edges connecting regions. i,j () is a region pair index; , Scenes Weighting coefficients for renewable energy reserves and thermal power regulation reserves; Represents a set of renewable energy units; Represents a set of thermal power units; This represents a set of uncertain scenarios arising from fluctuations in wind and light intensity and changes in load. Indicates the region i The load; Indicating uncertainty in scenarios Below, area i The actual total active power output of all generator units within the facility is the decision variable. x and scene The function; In uncertain scenarios Below, area i The net exchange power (with positive input) of the tie line to the external power grid is a decision variable. x and scene The function; This is represented as an uncertain scenario. The total reserve capacity required by the system to maintain stable operation.

[0098] Regional dynamic weights and It adaptively adjusts based on the current system status, reflecting the dynamic switching of regulation responsibilities between new energy sources and thermal power.

[0099] The model's inputs include real-time measured ACE, tie-line power, system inertia estimates, and uncertainty set. Its output includes the optimized target value of ACE for each region and the standby call instructions for each unit. The output includes the weighting coefficients for renewable energy reserves and thermal power regulation reserves. The results are directly used to guide the setting of optimization objectives and the generation of constraints for the MPC in the coordinated optimization layer, thereby achieving cross-regional, multi-resource coordinated frequency control.

[0100] Specifically, the ACE optimization target values ​​for each region output by the uncertainty control layer are directly passed to the MPC model of the coordinated optimization layer as input parameters, serving as the ACE reference trajectory that the model needs to track, or directly replacing the ACE in its objective function. t ) variables, ensuring that the second-level optimization objective of MPC aligns with the system-level robust coordination objective. For each unit's standby call instruction, it represents the standby reference value that each unit should be called up under the current system state. This value will be used as the expected value of the unit adjustment instructions in the coordination optimization layer MPC model, and incorporated into the MPC objective function as a soft constraint (e.g., by adding...). (Items) or directly incorporated into the constraints, constraining the MPC optimization results to track the global reserve call strategy determined by robust optimization. The weight coefficients of renewable energy reserves and thermal power regulation reserves reflect the allocation of regulation responsibilities between new energy and thermal power in the current scenario. Both will be transformed into differentiated penalty coefficients for regulation commands of new energy units and thermal power units in the MPC objective function of the coordination optimization layer.

[0101] To explicitly reflect the dynamic adjustment of the "uncertainty control layer" to the optimization objective of the "coordination optimization layer," the original objective function of the coordination optimization layer is reconstructed. The specific quantitative form of the reconstructed MPC objective function of the coordination optimization layer is as follows:

[0102] in, To determine the overall objective function value of the reconstructed Coordinated Optimization Layer Model Predictive Control (MPC), the system generates the current optimal control command by solving for the control sequence that minimizes this function value. The prediction horizon of MPC represents the number of steps to roll the prediction forward from the current time. t This is the current control moment; To predict the index of future time steps in the time domain, the value is... ; This is the standard subscript representation of model predictive control, indicating that based on the current time... t The system state, for the future... Forecast values ​​or control plans for time-varying variables; To coordinate the prediction of the system in the future by the optimization layer Actual value of the area control error at any given time; This is a document issued from the upper layer (uncertainty control layer) to this layer regarding the future. The target reference value for optimizing the regional control error that the system is expected to track at all times; ACE deviation penalty weighting coefficient, which is a constant. This value is used to measure the system's tolerance for frequency and tie-line power deviations. The larger the weight, the more the system tends to prioritize eliminating ACE deviations in order to maintain grid frequency stability. A collection of new energy (such as wind power and photovoltaic) units or power plants that participate in AGC regulation; i For the collection of new energy units Individual unit index variables; A collection of traditional thermal power units participating in AGC regulation; j For thermal power units Individual unit index variables; The coordinated optimization layer is calculated as a decision variable, and is intended to be used in the future. t + k The information is sent to the first person at each time. Taiwan New Energy Unit and the first Unit adjustment instructions for thermal power units (expected output adjustment value); and This is a decision made and issued by the upper layer (uncertainty control layer) based on minute-level robust decision calculations, targeting the first... The first new energy unit and the jth thermal power unit in the future Backup call instructions at any time; Let represent the square of the Euclidean norm (norm 2), which serves as a quadratic penalty term here, intended to minimize the actual control command in the form of a soft constraint. ) and upper-level reference instructions ( The degree of deviation between ACE and target, or the actual ACE and target. The degree of deviation; In order to target the i The adjustment penalty coefficient (i.e. adjustment cost weight) of the new energy unit in the coordination and optimization layer. The smaller this value is, the more the objective function tends to call the new energy unit for adjustment. In order to target the j The adjustment penalty coefficient of the thermal power unit in the coordination optimization layer.

[0103] Furthermore, the dynamic weights output by the uncertainty control layer ( The preferred quantitative mapping relationship between the above-mentioned penalty coefficient and the above-mentioned penalty coefficient is as follows:

[0104] in, This serves as the base penalty coefficient. Using the above quantitative formula, when the system requires new energy sources to undertake more regulatory tasks, the uncertainty control layer will increase... This directly corresponds to reducing the adjustment penalty coefficient for new energy units in the MPC objective function. This encourages new energy generating units to undertake more regulation tasks. Through the aforementioned information transmission path, the minute-to-second robust decision-making of the uncertainty control layer can provide precise target guidance and constraint boundaries for the second-level rolling optimization of the coordination optimization layer.

[0105] This model, as the core optimization module within the coordination and optimization layer, is a crucial link in the three-layer AGC control architecture, enabling multi-timescale collaboration of "hourly planning, second-level control, and sub-second-level execution." It uses hourly planning from the strategic decision-making layer as a constraint and second-level real-time data from the coordination and optimization layer as dynamic input. Through robust optimization, it provides precise decision-making basis for the MPC framework of the coordination and optimization layer, enabling it to generate second-level AGC instructions adapted to the uncertainties of new energy sources. This, in turn, guides the sub-second-level rapid execution of the execution response layer. Simultaneously, through bidirectional feedback, it provides second-level support to the coordination and optimization layer and hourly dynamic correction to the strategic decision-making layer. This upgrades the three-layer architecture from "top-down instruction issuance" to "multi-timescale dynamic closed-loop collaboration," ultimately improving the robustness and control efficiency of the entire AGC system under high new energy penetration rates.

[0106] This hierarchical MPC control framework improves the structural adaptability and control flexibility of the entire AGC system by decomposing the time scale and physical boundary of the system frequency control task, and provides a foundation for the subsequent introduction of robust optimization and adaptive control.

[0107] Example 2: Reference Figure 1 This is the second embodiment of the present invention. This embodiment provides a power system optimization method and system based on Weibull-Beta hybrid distribution for wind power prediction error modeling and multi-source AGC hierarchical control, including: Obtain the topology and operation data of the power grid within the control range of a certain dispatch master station; obtain the output limits, real-time output and controllable status of conventional units and new energy power plants under the dispatch master station; obtain the active power planned value of the next planned point issued by the dispatch master station to the new energy power plants and conventional units within the control range; obtain the historical data of wind power output in the dispatch area.

[0108] To verify the effectiveness of the proposed dynamic reserve robust optimization model and multi-source AGC hierarchical control strategy, this example uses a simulation experiment based on the IEEE 118-bus system. The system consists of 10 AGC units, and their node distribution and output range are shown in Table 2. The simulation platform uses Matlab / Simulink, and the wind power output data is derived from actual operating data from Texas, USA.

[0109] Table 2. Distribution of AGC Units

[0110] Figure 2 The IEEE 118-node power grid architecture was demonstrated. Figure 3 This provides data on wind power output and its upper limit on a certain day in Texas.

[0111] To improve the accuracy of wind power output modeling, this invention introduces the Weibull-Beta hybrid distribution to establish a wind power probability model. In this model, the Weibull distribution is used to characterize wind speed and its fluctuations. Its good fitting performance makes it widely used in the fields of wind energy assessment and wind power modeling. Furthermore, a Beta distribution is introduced as a supplement, allowing for fine-tuning of the tail thickness and peak position of the error distribution through shape parameters. This is achieved through weighting coefficients. , Adjusting the tail thickness and peak position, its density function is:

[0112] in, Let be a Beta function, representing the normalization constant of the Beta distribution, ensuring that the integral sum of the probability density functions is 1, and satisfying the following conditions: , It is the Gamma function; This represents the standardized wind power prediction error. The first shape parameter of the Beta distribution controls the degree of left skewness and the thickness of the left tail. This is the second shape parameter of the Beta distribution, which controls the degree of right skewness and the thickness of the right tail.

[0113] By linearly superimposing Weibull and Beta distributions, a bimodal mixture probability density function is constructed, where the mixture weights are adjustable parameters, giving the model high flexibility in expressing complex error structures. The linear combination of these distributions forms a composite distribution:

[0114] in, This represents the standardized wind power prediction error. The weighted coefficients are mixed and satisfy the following conditions: This is used to achieve flexible expression of bimodal features; Let Weibull's probability density function be the error of wind power prediction. Let be the scale parameter of the Weibull distribution. For shape parameters; Let be the probability density function of the Beta distribution of wind power prediction error. The first shape parameter of the Beta distribution controls the degree of left skewness and the thickness of the left tail. This is the second shape parameter of the Beta distribution, which controls the degree of right skewness and the thickness of the right tail.

[0115] The fitting effect of the mixed distribution based on this example is as follows: Figure 4 As shown, the mixture probability density function, including the Weibull-Beta component and its Weibull and Beta components, reflects the heavy-tailed effect. The mixture distribution parameters are shown in Table 3. Table 3. Weibull-Beta Mixture Distribution Parameters

[0116] To address the uncertainty in reserve capacity delivery caused by wind power output forecasting errors, this patent constructs a two-stage robust optimization model for scenarios with high renewable energy penetration, aiming to improve the power system's flexible adjustment capabilities and operational safety. The model's design logic involves formulating reserve reservation and real-time adjustment strategies in stages as uncertainty information is gradually revealed, achieving an optimal balance between reserve costs and system risks.

[0117] The first phase primarily addresses the "deterministic problem of pre-emptive reserve capacity allocation," namely, how to rationally formulate reserve capacity strategies for various reserve units based on known wind power forecast information and error boundaries before wind power output is realized and the system is operational. The goal of this phase is to minimize expected reserve costs and, considering the uncertainties of wind power, ensure the system has the basic adjustment capability to cope with potential fluctuations through conservative reservations. Its core decision variable is the pre-allocation of reserve capacity, which directly relates to the deployment efficiency of reserve resources at the spatial and turbine levels.

[0118] The second phase addresses the "responsiveness of in-process operation adjustments," specifically how to adjust the system's operating state to ensure its safety and stability after actual wind power output has been achieved and specific output deviations have occurred. This phase introduces a chance constraint mechanism, relaxing key operational constraints (such as system power balance and turbine output limits) at a set confidence level to control the probability of risk occurrence, thereby achieving the minimum adjustment cost and risk premium cost under controllable risk. This phase reflects the system's dynamic response capability to wind power output disturbances and serves as a test of the effectiveness of the configuration strategy in the first phase.

[0119] Based on the aforementioned probabilistic model, this example combines a dynamic robust optimization algorithm to optimize the all-weather standby capacity scheduling of AGC units. Figure 5The data shows the daily standby allocation of 10 AGC units at the IEEE 118 node under different renewable energy penetration rates. Figure 5 In the diagram, each line represents the amount of standby space allocated to each unit throughout the day.

[0120] The results show that as penetration increases, more AGC units are mobilized to participate in vertical adjustment tasks to meet system backup requirements. Furthermore, Figure 6 The performance of the two backup allocation methods was compared. Figure 6 The study demonstrates the allocation of upstream and downstream reserves for new energy resources and thermal power resources at different penetration rates. The results show that when the wind power penetration rate reaches 50%, the reserve capacity undertaken by wind power exceeds that of thermal power for the first time. As the penetration rate of new energy gradually increases, the upstream and downstream reserves of wind power gradually increase, and at a penetration rate of 50%, it surpasses that of thermal power. This situation reflects the dominant role of new energy in system regulation under high penetration scenarios.

[0121] In terms of economy, Figure 7 The total daily cost of the method of this invention was compared with that of the traditional fixed-ratio standby configuration method. Figure 7 In the above, Scheme 1 is the method of the present invention, and Scheme 2 is the traditional fixed ratio capacity configuration method. As the penetration rate increases, the method of the present invention effectively reduces operating costs while maintaining system stability, demonstrating higher economic efficiency.

[0122] Furthermore, to adapt to the frequency regulation pressure brought about by the high proportion of new energy access, this invention constructs a hierarchical AGC control architecture based on MPC. This architecture divides the system control tasks into three levels: strategic decision-making, coordination optimization, and execution response. Structurally, it realizes information flow and feedback loop between the upper and lower levels, ensuring that each control module can operate independently while also possessing the ability to coordinate and collaborate.

[0123] To enhance the adaptive adjustment capability of AGC systems in scenarios with high penetration of new energy sources, this invention improves the traditional PID control method based on linear decision rules. By introducing system state information to dynamically adjust the PID controller parameters, the control strategy can flexibly adapt to system operating conditions, effectively enhancing the stability and robustness of frequency regulation.

[0124] The core idea of ​​linear decision rules is to use key state variables of the current system, such as ACE, ACE change rate, and wind and solar power output change rate, as the basis for decision-making, and dynamically adjust the gain coefficient of the PID controller through a preset linear relationship. Assume... Let the proportional, integral, and derivative gains be respectively. Then, the adaptive gain adjustment based on LDR can be expressed as:

[0125] in, Based on the gain, (i=1,2,3) are decision coefficients. Indicates the current ACE deviation. The ACE rate of change The rate of change in the output of new energy sources.

[0126] Compared to the traditional fixed-parameter PID control method, the LDR method gives the controller a certain "sensing" and "judgment" capability, enabling it to optimize and adjust its behavior in real time according to the magnitude and trend of system disturbances, thereby achieving more refined and efficient frequency control.

[0127] In designing the control strategy, the differences between new energy control resources and traditional thermal power resources in terms of response speed, inertia characteristics, and control capabilities were considered. New energy equipment has the advantage of rapid response in the early stages of frequency disturbances, making it suitable for high-sensitivity control tasks; while thermal power units, due to their greater mechanical inertia and stronger continuous support capabilities, provide stable power support in the later stages of frequency adjustment. LDR-PID effectively coordinates the action paths of the two types of resources by dynamically adjusting parameter weights under different operating conditions, making the frequency control process more hierarchical and continuous.

[0128] Furthermore, the algorithm maintains stable system operation under small deviations, avoiding system oscillations caused by frequent adjustments; when large disturbances occur, it enhances the system's dynamic response and robustness by increasing the controller gain. Overall, this improved PID algorithm features a simple structure, low computational cost, and strong engineering feasibility, and exhibits good compatibility with the MPC hierarchical control structure.

[0129] To verify the regulation performance of the proposed AGC strategy under disturbance response, this invention introduces the following... Figure 8 The randomly generated disturbances shown are used to test three AGC control strategies: Comparing the performance of three AGC strategies: Strategy 1: Traditional PID strategy; Strategy 2: LDR-PID Improved Control; Strategy 3: Hierarchical MPC+LDR-PID collaborative control (the method of this invention).

[0130] The performance of the three AGC strategies is compared as follows: from Figure 9 As can be seen from the cost curves, the three curves show the cost changes within 300 seconds. Strategy 3 has the lowest cost while ensuring system performance, but its cost is almost the same as that of Strategy 1. Strategy 2 has the highest cost in order to maintain system stability.

[0131] Three strategies for ACE variation are as follows Figure 11Further from the ACE response ( Figure 10 ) and frequency control effect ( Figure 11 From the perspective of strategy 3, it outperforms strategy 2 and strategy 1 in both indicators, demonstrating a stronger system adjustment capability.

[0132] Table 4 summarizes the key performance indicators of each control strategy. The results show that strategy 3 outperforms other strategies in terms of CPS1, root mean square frequency difference, root mean square ACE, and daily adjustment cost, and has the best adjustment effect.

[0133] Table 4. Comparison of key performance indicators for the three control strategies

[0134] also, Figure 12 Comparison curves of AGC control commands and responses under three strategies are presented. (a) represents the AGC command for new energy sources; (b) represents the AGC command for thermal power plants; (c) represents the AGC response for new energy sources; and (d) represents the AGC response for thermal power plants. The results show that strategy 3 exhibits the smallest frequency deviation and the best AGC command tracking performance, particularly in the regulation of new energy sources.

[0135] Example 3: An embodiment of the present invention provides a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of Embodiment 1.

[0136] Example 4: The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the embodiments.

[0137] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0138] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0139] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0140] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A hierarchical optimization method for AGC based on wind power error modeling, characterized in that, Includes the following steps: Step 1: Construct a wind power prediction error model and a dynamic probability boundary model for wind power prediction error; Step 2: Construct a two-stage robust optimization model that includes reserve capacity configuration and real-time scheduling of unit output, and construct the power balance constraints of the two-stage robust optimization model based on the dynamic probability boundary model of wind power prediction error. Step 3: Construct an AGC hierarchical control architecture including a strategic decision-making layer, a coordination and optimization layer, and an execution response layer. The strategic decision-making layer calculates the reserve capacity configuration strategy of each unit based on the two-stage robust optimization model, which serves as the boundary constraint for the regulation capacity of each unit in the coordination and optimization layer. The coordination and optimization layer calculates the unit regulation instructions of each unit with the goal of minimizing the regional control error ACE and regulation cost, and issues them to the execution response layer for instruction execution.

2. The AGC hierarchical optimization method according to claim 1, characterized in that: In step 1, the wind power prediction error model is constructed based on the Weibull-Beta mixed distribution, and the mixed distribution is sampled. The dynamic probability boundary model of wind power prediction error is obtained based on the quantile statistics of the sampling results.

3. The AGC hierarchical optimization method according to claim 2, characterized in that: In step 1, the parameters of the wind power prediction error model are updated based on the wind power prediction error data for different operating scenarios, so as to obtain the wind power prediction error model and the corresponding dynamic probability boundary model of wind power prediction error under different operating scenarios; the operating scenarios include different regions. The dynamic probability boundary model for wind power prediction error is as follows: in, For time period t ,area r The lower bound of the confidence interval for wind power prediction error; For time period t ,area r The upper bound of the confidence interval for lower wind power prediction error; The Weibull-Beta mixture distribution is fitted based on the wind power prediction error data under the current operating scenario; pl , pu These are the lower and upper quantiles of the confidence interval, respectively. , For each of the mixed distributions The quantile function, which takes the lower and upper quantiles of the confidence interval, is used to output the lower and upper bounds of the confidence interval for wind power prediction error in the corresponding scenario.

4. The AGC hierarchical optimization method according to claim 3, characterized in that: In step 2, the two-stage robust optimization model includes the following steps: In the reserve capacity configuration stage, the optimal reserve capacity configuration scheme is calculated with the goal of minimizing the expected reserve cost; in the real-time scheduling stage of unit output, based on the optimal reserve capacity configuration scheme, the optimal real-time scheduling scheme is obtained under the corresponding wind power prediction error scenario by probabilistically relaxing the system power balance constraints at a set confidence level, with the goal of minimizing the adjustment cost and risk premium cost. The two-stage robust optimization model is solved through the following iterative process: the adjustment cost and risk premium cost corresponding to the optimal solution in the real-time scheduling stage of unit output are mathematically expected under all wind power prediction error scenarios and then superimposed as feedback terms onto the objective function of the reserve capacity configuration stage to update the reserve capacity configuration scheme. The iteration is continued until the reserve capacity configuration scheme converges.

5. The AGC hierarchical optimization method according to claim 4, characterized in that: During the real-time scheduling phase of unit output, the probabilistic relaxation of the system power balance constraints under a set confidence level is specifically defined as follows: in, This is an uncertain variable, representing the wind power prediction error; For the first g The planned output of the conventional generating units; To account for uncertain variables The next g Real-time adjustment of the unit; To account for uncertain variables The actual output of wind power after that; This represents the system load. The confidence level is the chance constraint.

6. The AGC hierarchical optimization method according to claim 1, characterized in that: In step 3, the AGC hierarchical control architecture also includes an uncertainty control layer to support the coordinated regulation of frequency control between regions and the dynamic weight configuration of multi-source backup resources. The uncertainty control layer uses the backup capacity configuration strategy of each unit calculated by the strategic decision layer as the constraint boundary. Through robust optimization, it generates the regional control error ACE optimization target value, unit backup call command and the weight coefficient of renewable energy backup and thermal power regulation backup for each region and sends them to the coordination optimization layer to reconstruct the optimization target of the coordination optimization layer.

7. The AGC hierarchical optimization method according to claim 5, characterized in that: After receiving the AGC adjustment command from the coordination and optimization layer, the execution response layer tracks the AGC adjustment command using an adaptive PID controller based on linear decision rules. The adaptive PID controller dynamically adjusts its gain coefficient based on the current system's ACE deviation, ACE rate of change, and wind and solar power output rate of change. Specifically: in, Basic gain; , , (i=1,2,3) are the decision coefficients; The current ACE bias, The ACE rate of change The rate of change in the output of new energy sources.

8. A hierarchical AGC optimization system based on wind power error modeling, according to any one of claims 1-7, comprising a dynamic probability boundary model construction module for wind power prediction error, a two-stage robust optimization model construction module, and an AGC hierarchical control architecture construction module, characterized in that: The module for constructing the dynamic probability boundary model of wind power prediction error is used to construct the wind power prediction error model and the dynamic probability boundary model of wind power prediction error. A two-stage robust optimization model construction module is used to construct a two-stage robust optimization model including reserve capacity configuration and real-time scheduling of unit output, and to construct the power balance constraints of the two-stage robust optimization model based on the dynamic probability boundary model of wind power prediction error. The AGC hierarchical control architecture construction module is used to construct an AGC hierarchical control architecture including a strategic decision-making layer, a coordination and optimization layer, and an execution response layer. The strategic decision-making layer calculates the reserve capacity configuration strategy of each unit based on the two-stage robust optimization model, which serves as the boundary constraint for the regulation capacity of each unit in the coordination and optimization layer. The coordination and optimization layer calculates the unit regulation instructions of each unit with the goal of minimizing the regional control error ACE and regulation cost, and issues them to the execution response layer for instruction execution.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

Images