Model-free reactive voltage control method based on zero-order feedback optimization

Abstract

The invention relates to the technical field of power system operation and control, in particular to a model-free reactive voltage control method based on zero-order feedback optimization, and the method comprises the steps: constructing a power grid reactive voltage optimization model, and obtaining reactive voltage optimization target values at different moments based on a target function; calculating gradient estimation values of the target function at different moments according to the reactive voltage optimization target values at different moments; extracting a gradient updating step length at the next moment of the target moment according to the gradient estimation values of the target function at different moments; and updating the reactive voltage control instruction at the next moment based on an accelerated gradient projection iterative algorithm. The objective of the invention is to solve the technical problems of large calculation amount and insufficient effectiveness of reactive voltage control.

Description

Technical Field

[0001] The present invention relates to the technical field of power system operation and control, and in particular to a model-free reactive voltage control method based on zero-order feedback optimization. Background Art

[0002] Driven by energy and environmental issues, the proportion of renewable energy in the power grid is increasing, and large-scale, high-penetration renewable energy power generation and grid connection have become the forefront and hot spot in the international energy and power fields. Due to the large amount of renewable energy, dispersion and strong volatility, it has brought a series of negative impacts on the voltage quality and dispatching operation of the power grid. At the same time, renewable energy is often connected to the grid through power electronic inverters, which have flexible and high-speed regulation capabilities. In order to efficiently optimize and control distributed renewable energy generation and improve the voltage quality of the power grid with high renewable energy penetration, voltage control has become an important part of the power grid regulation and operation.

[0003] In traditional power grids, methods based on power grid models are often used to achieve reactive power control and optimize the active power loss and voltage distribution of power grids. However, traditional model-based optimization control methods rely on accurate system model parameters, while the ideal model of the power grid is difficult to obtain, and this model-based optimization method is difficult to apply in practice. In order to deal with the problem of incomplete models of power grids, data-driven control methods have been proposed in recent years. This method attempts to infer the behavior pattern of the power grid by utilizing a large amount of historical operating data. However, most of the existing data-driven control methods require the construction of an equivalent data-driven model of the system from the data, which involves high-dimensional regression calculation problems and increases the complexity of the calculation; on the other hand, there is often a deviation between the data-driven estimation model and the actual physical system model. Even if a data-driven module is constructed, the difference between it and the actual physical characteristics of the power grid may reduce the effectiveness of the control strategy. Summary of the invention

[0004] In order to solve the technical problems of large reactive voltage control calculation amount and insufficient effectiveness, the purpose of the present invention is to provide a model-free reactive voltage control method based on zero-order feedback optimization. The technical solution adopted is as follows:

[0005] A model-free reactive voltage control method based on zero-order feedback optimization, the method comprising:

[0006] Construct a reactive power voltage optimization model for the power grid, and obtain the reactive power voltage optimization target values ​​at different times based on the objective function;

[0007] Calculate the estimated gradient value of the objective function at different times according to the reactive voltage optimization target value at different times;

[0008] Extract the gradient update step size at the next target moment according to the estimated gradient value of the target function at different moments;

[0009] The reactive voltage control command at the next moment is updated based on the accelerated gradient projection iterative algorithm.

[0010] Furthermore, obtaining the reactive voltage optimization target value at different times includes:

[0011] The voltage amplitude column vectors of observable voltage nodes and the active power value column vectors of common coupling nodes at different times of the power grid are obtained, and the reactive voltage optimization target values ​​at different times are calculated according to the voltage amplitude reference values ​​of observable voltage nodes of the power grid, the preset voltage deviation penalty cost coefficients, and the preset active network loss optimization cost coefficients.

[0012] Further, calculating the estimated gradient value of the objective function at different times according to the reactive voltage optimization target value at different times includes:

[0013] According to the difference between the reactive voltage optimization target values ​​at different times, the gradient estimation value of the objective function at different times is calculated based on the preset smoothing coefficient and the standard normal distribution random vector.

[0014] Further, extracting the gradient update step size at the next moment of the target moment according to the gradient estimation value of the target function at different moments includes:

[0015] Select a target time and calculate the active constraint set of the power grid at the time before the target time;

[0016] Extracting an inactive constraint set based on an active constraint set of the power grid at the previous moment;

[0017] Calculate the gradient update step size at the next target moment based on the set of inactive constraints.

[0018] Furthermore, a target time is selected and the active constraint set of the power grid at the previous time before the target time is calculated, including:

[0019] Obtain the control input vector of the power grid at the previous moment of the target moment and the estimated value of the objective function gradient;

[0020] The active constraint set of the power grid at the previous moment is extracted based on the upper and lower limit thresholds of the control input vector and the estimated value of the gradient of the objective function at the previous moment.

[0021] Furthermore, extracting the inactive constraint set based on the active constraint set of the power grid at the previous moment includes:

[0022] The elements in the active constraint set are removed from the control input vector of the power grid at the previous moment of the target moment, and the control input vector includes at least the control instructions of the reactive power output of the inverter-based distributed power source and the control instructions of the reactive power output of the static reactive compensation device.

[0023] Further, calculating the gradient update step size at the next moment of the target moment according to the inactive constraint set includes:

[0024] Extract the difference vector between the control input vector at the target moment and the previous moment and the objective function gradient estimate;

[0025] Extracting a first element vector belonging to the inactive constraint set from the difference vector of the control input vector, and calculating the square value of the Euclidean norm of the first element vector;

[0026] Extracting a second element vector belonging to the inactive constraint set from the difference vector of the estimated value of the objective function gradient;

[0027] Calculate the dot product of the first element vector and the second element vector;

[0028] The gradient update step size for the next moment is calculated based on the square value of the Euclidean norm of the first element vector and the dot product.

[0029] Further, updating the reactive voltage control instruction at the next moment based on the accelerated gradient projection iterative algorithm includes:

[0030] Calculate candidate control variables of the power grid at the next moment of the target moment according to gradient descent;

[0031] The feasible region of the control input vector at the next moment is calculated based on the candidate control variables at the next moment by using random disturbances.

[0032] Furthermore, the candidate control variables of the power grid at the next moment after the target moment are calculated according to the gradient descent, including:

[0033] Calculate the gradient descent direction at the target moment based on the candidate control variables at the target moment, the estimated value of the objective function gradient, and the gradient update step size at the next moment of the target moment;

[0034] The projection operator is used to map the gradient descent direction at the target moment to the feasible domain of the power grid control input vector at the next moment, and the candidate control variables of the power grid at the next moment are obtained.

[0035] Furthermore, by using random disturbances, the feasible domain of calculating the control input vector at the next moment based on the candidate control variables at the next moment includes:

[0036] Randomly initialize the random disturbance vector of the power grid at the next moment;

[0037] Adjust the intensity of the random disturbance vector of the power grid at the next moment according to a preset positive scalar;

[0038] The control input vector at the next moment is calculated according to the candidate control variables of the power grid at the next moment and the weighted random disturbance vector.

[0039] The present invention has the following beneficial effects:

[0040] The model-free reactive voltage control method based on zero-order feedback optimization provided by the present invention realizes model-free closed-loop reactive voltage control by calculating the estimated gradient values ​​of the objective function at different moments according to the reactive voltage optimization target values ​​at different moments, extracting the gradient update step size at the next moment of the target moment according to the estimated gradient values ​​of the objective function at different moments, and iteratively updating the reactive voltage control instructions based on the accelerated gradient iteration algorithm. The method does not require accurate system model parameters, nor does it require the construction of an equivalent data-driven model of the power grid. Through the zero-order feedback optimization method, it only needs to calculate the estimated gradient values ​​of the objective function at different moments according to the reactive voltage optimization target values ​​at different moments, thereby avoiding the explicit gradient calculation and reducing the amount of calculation. On the other hand, the method does not rely on the accurate model parameters of the power grid, is easy to deploy and implement in the power grid, and can realize the optimized operation of the power grid in a scenario with incomplete models. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A method flow chart of a model-free reactive voltage control method based on zero-order feedback optimization provided by an embodiment of the present invention;

[0043] Figure 2 A flow chart of a method for obtaining a gradient update step size at the next moment of a target moment provided by an embodiment of the present invention;

[0044] Figure 3 A flow chart of a method for updating reactive voltage control instructions at the next moment provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a model-free reactive voltage control method based on zero-order feedback optimization proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0046] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0047] A specific scheme of a model-free reactive voltage control method based on zero-order feedback optimization provided by the present invention is described in detail below in conjunction with the accompanying drawings.

[0048] See also Figure 1 , which shows a method flow chart of a model-free reactive voltage control method based on zero-order feedback optimization provided by an embodiment of the present invention, the method comprising:

[0049] Step S100: constructing a power grid reactive voltage optimization model, and obtaining reactive voltage optimization target values ​​at different times based on an objective function;

[0050] Step S100 specifically includes:

[0051] Step S110: Obtain the voltage amplitude column vectors of observable voltage nodes and the active power value column vectors of common coupling nodes at different times of the power grid, and calculate the reactive voltage optimization target values ​​at different times according to the voltage amplitude reference values ​​of observable voltage nodes of the power grid, the preset voltage deviation penalty cost coefficient, and the preset active network loss optimization cost coefficient; In this embodiment, a model is established with the inverter-based distributed power source and static VAR compensation device in the power grid as the controlled devices:

[0052] The inverter-based distributed power model can be expressed as:

[0053]

[0054] Where N DG represents the set of nodes in the power grid that are equipped with inverter-based distributed power sources, i represents the grid node index; Q Di,t represents the control instruction of the reactive power output of the distributed power source at node i at time t; |Q Di,t | indicates Q Di,t The absolute value of represents the active power output of the distributed generation at node i at time t; S Di represents the rated capacity of the distributed generation at node i;

[0055] The static VAR compensation device model can be expressed as:

[0056]

[0057] Where N SVC represents the set of nodes in the power grid where static VAR compensation devices are installed, i represents the grid node index; Q Ci,trepresents the control instruction of the reactive power output of the static VAR compensation device at node i at time t; Q Ci,min and Q Ci,max They represent the lower and upper limits of reactive power output of the static VAR compensation device at node i respectively;

[0058] Obtain the voltage amplitude column vectors of observable voltage nodes of the power grid at different times and the active power value column vectors of common coupling nodes. The objective function of the power grid reactive voltage optimization model can be expressed as:

[0059] F t =c 1 ||V t -V ref || 2 +c 2 p PCC,t

[0060] In the formula, F t represents the optimization objective function value of the reactive power voltage control of the power grid at time t; V t is a obs dimensional column vector, each element of which represents the voltage amplitude of the observable voltage node in the power grid at time t, n obs is the total number of observable voltage nodes in the power grid; V ref is a obs dimensional column vector, each element of which represents the voltage amplitude reference value of the observable voltage node in the power grid at time t; PCC,t represents the column vector of the active power value of the common coupling node of the controlled power grid at time t; c 1 is a scalar greater than zero, indicating the penalty cost coefficient for voltage deviation of the controlled power grid; c 2 is a scalar greater than zero, indicating the active power loss optimization cost coefficient of the controlled power grid; the typical values ​​can be taken as c 1 =500, c 2 =1;

[0061] Step S200: calculating the estimated gradient value of the objective function at different times according to the reactive voltage optimization target value at different times; specifically, performing model-free reactive voltage control based on zero-order feedback optimization on the power grid, wherein the time index t≥0 is a non-negative integer;

[0062] Step S200 specifically includes:

[0063] Step S210: calculating the estimated gradient value of the objective function at different times according to the difference between the reactive voltage optimization target values ​​at different times, and based on a preset smoothing coefficient and a standard normal distribution random vector;

[0064] For example, the gradient estimate of the reactive power voltage optimization objective function at time t can be expressed as:

[0065]

[0066] In the formula, Represents the estimated value of the objective function gradient at time t, which is an n u dimensional vector, n u It is the sum of the total number of distributed power sources and static VAR compensation devices in the controlled power grid; F t+1 represents the optimization objective function value of the reactive power voltage control of the power grid at time t+1; δ represents the preset smoothing coefficient, which is a positive scalar and preferably takes a value of 0.01; v t represents the independent and identically distributed standard normal random vectors generated at time t;

[0067] In this embodiment, the difference between the reactive voltage optimization target values ​​at different times is used, and the preset smoothing coefficient and the voltage amplitude column vector of the observable voltage node of the power grid are combined to calculate the gradient estimate of the objective function. This method avoids the explicit gradient calculation in the traditional method and simplifies the calculation process. In practical applications, especially when the power grid is large and the topology is complex, the explicit calculation of the gradient may be very complex and time-consuming. The zero-order feedback optimization method only needs to obtain the gradient estimate through a simple difference calculation, which greatly reduces the amount of calculation and complexity; on the other hand, it does not rely on the precise power grid model parameters, but directly constructs the objective function based on the measured value, and estimates the gradient through the zero-order feedback optimization method. Since the actual power grid model is often not completely accurate, especially when the power grid structure changes or there are unknown parameters, this method can still operate effectively in the case of an incomplete model, which improves practicality.

[0068] See also Figure 2 , step S300: extracting the gradient update step size at the next moment of the target moment according to the gradient estimation value of the objective function at different moments; it should be noted that if the target moment is 0, the gradient update step size at the next moment is a preset initialization value;

[0069] Step S300 specifically includes:

[0070] Step S310: Select a target time and calculate the active constraint set of the power grid at the previous time of the target time;

[0071] Step S310 specifically includes:

[0072] Step S311: Obtain the control input vector of the power grid at the previous moment of the target time and the estimated value of the objective function gradient; obtain the control input vector u at the previous moment t-1 t-1, which contains the control instructions of all controllable devices (such as inverter-based distributed power sources and static VAR compensation devices) at time t-1; the gradient estimate of the objective function at time t-1 can be It can be calculated according to the reactive voltage optimization target value at different times in step S200;

[0073] Step S312: extracting the active constraint set of the power grid at the last moment based on the upper and lower limit thresholds of the control input vector and the estimated value of the gradient of the objective function at the last moment;

[0074] The active constraint set of the power grid at the previous moment of the target moment can be expressed as:

[0075]

[0076] In the formula, J t-1 represents the set of active constraints of the power grid at time t-1; j represents the element number index of the control input vector, which is in the range of [0,n u ] is a positive integer; u t-1 represents the control input vector at time t-1; represents the control input vector u t-1 ut-1 represents the upper threshold of the control input vector at time t-1; Indicates the lower threshold of the control input vector at time t-1; Represents the estimated value of the objective function gradient at time t-1; ∧ represents a logical OR operation; ∨ represents a logical AND operation; This embodiment extracts the active constraint set at the previous moment to ensure that the control variable update at the next moment will not violate any physical or engineering restrictions. The algorithm can focus on those constraints that have a real impact on the current control state, thereby ensuring the effectiveness of the control variables and the robustness of the algorithm, and improving the control accuracy.

[0077] Step S320: extracting an inactive constraint set based on the active constraint set of the power grid at the last moment;

[0078] Step S320 specifically includes:

[0079] Step S321: remove the elements in the active constraint set from the control input vector of the power grid at the previous moment of the target moment, the control input vector at least includes the control instructions of the reactive power output of the distributed power source based on the inverter and the control instructions of the reactive power output of the static reactive power compensation device, that is, u in step S312 t-1 The elements in contain the control instructions for the reactive power output of all inverter-based distributed generation in the power grid at time t-1. And the control instructions for reactive power output of all static VAR compensation devices

[0080] The set of inactive constraints can be expressed as:

[0081] I t-1 ={1,...,n u}\J t-1

[0082] In the formula, I t-1 represents the set of grid inactive constraints at time t-1; n u is the sum of the total number of distributed generation sources and static VAR compensation devices in the controlled power grid; J t-1 represents the active constraint set of the power grid at time t-1; by extracting the inactive constraint set I t-1 , the model can focus more on the control variables that have not reached the upper and lower limits, thereby ensuring that these variables can be effectively adjusted in the control variable update at the next moment; by extracting the inactive constraint set, those control variables that have reached the upper and lower limits can be ignored, further simplifying the calculation process, and the final control strategy can more accurately reflect the actual operating status of the power grid.

[0083] Step S330: Calculate the gradient update step size at the next moment of the target moment according to the inactive constraint set;

[0084] Step S330 specifically includes:

[0085] Step S331: extract the difference vector between the control input vector at the target moment and the previous moment and the objective function gradient estimate; specifically, the difference vector of the control input vector can be expressed as: s t =u t -u t-1 , for an n u dimensional vector; the difference vector of the objective function gradient estimate can be expressed as: For an n u dimensional vector;

[0086] Step S332: extracting the first element vector belonging to the inactive constraint set from the difference vector of the control input vector, and calculating the square value of the Euclidean norm of the first element vector; specifically, the first element vector can be expressed as: The difference vector s represents the control input vector t The element number index set in belongs to the set I t-1 A vector consisting of all elements of ;

[0087] Step S333: extracting the second element vector belonging to the inactive constraint set from the difference vector of the objective function gradient estimate; specifically, the second element vector can be expressed as: The difference vector z represents the estimated value of the objective function gradient t The element number index set in belongs to the set It-1 A vector consisting of all elements of ;

[0088] Step S334: Calculate the dot product of the first element vector and the second element vector;

[0089] Step S335: Calculate the gradient update step size at the next moment according to the square value of the Euclidean norm of the first element vector and the dot product;

[0090] The gradient update step size for calculating the next moment of the target moment can be expressed as:

[0091]

[0092] In the formula, α t+1 Represents the gradient update step size in the accelerated gradient projection iterative algorithm at time t+1; The difference vector s represents the control input vector t The element number index set in belongs to the set I t-1 A vector consisting of all elements of Representation vector The square of the second norm of ; The difference vector z represents the estimated value of the objective function gradient t The element number index set in belongs to the set I t-1 A vector consisting of all elements of ; T represents the transpose operator; the step size α is updated by calculating the gradient at time t+1 t+1 , the step size can be dynamically adjusted according to the gradient change at the current moment. Since the gradient update step size is calculated based on the elements of the inactive constraint set, this helps the algorithm to better explore the solution space and avoid falling into the local optimal solution. It can more accurately adjust the control variables that have not reached their upper and lower limits, thereby improving the control accuracy;

[0093] See also Figure 3 , step S400: updating the reactive voltage control instruction at the next moment based on the accelerated gradient projection iterative algorithm;

[0094] Step S400 specifically includes:

[0095] Step S410: calculating candidate control variables of the power grid at the next moment after the target moment according to gradient descent;

[0096] Step S410 specifically includes:

[0097] Step S411: Calculate the gradient descent direction at the target time according to the candidate control variables at the target time, the estimated value of the objective function gradient, and the gradient update step size at the next time of the target time, which can be expressed as: W t represents the candidate control variable at time t, which is an nu dimensional vector; α t+1 Represents the gradient update step size in the accelerated gradient projection iterative algorithm at time t+1; Represents the estimated value of the objective function gradient at time t;

[0098] Step S412: Use a projection operator to map the gradient descent direction at the target moment to the feasible domain of the power grid control input vector at the next moment, and obtain the candidate control variables of the power grid at the next moment; wherein the projection operator can be expressed as:

[0099]

[0100] Where, u represents U t+1 The symbol of the optimization decision variable in the vector; U t+1 represents the feasible domain of the control input vector at time t+1; ||·|| 2 represents the second norm; the candidate control variables of the power grid at the next moment can be expressed as:

[0101]

[0102] Where W t+1 represents the candidate control variable at time t+1, which is an n u dimensional vector; by using the projection operator to map the gradient descent direction to the feasible domain of the next moment, it is ensured that the updated control instructions are physically feasible, and invalid or infeasible control instructions can be avoided, thereby improving the stability of the entire control process; the projection operator simplifies the calculation process and can ignore those control variables that are already in the boundary state, simplifying the calculation process;

[0103] Step S420: using random disturbances, based on the candidate control variables at the next moment, calculate the feasible region of the control input vector at the next moment;

[0104] Step S420 specifically includes:

[0105] Step S421: Randomly initialize the random disturbance vector of the power grid at the next moment, which can be expressed as v t+1 , represents the independent and identically distributed standard normal random vector generated at time t+1, which is an n u dimensional vector; the generation of a standard normal random vector can be completed by a random number generator, and it is necessary to ensure that each element obeys the standard normal distribution;

[0106] Step S422: adjusting the intensity of the random disturbance vector of the power grid at the next moment according to a preset positive scalar;

[0107] Step S423: Calculate the control input vector at the next moment according to the candidate control variables of the power grid at the next moment and the weighted random disturbance vector;

[0108] The control input vector at the next moment can be expressed as:

[0109] u t+1 =W t+1 +δv t+1

[0110] In the formula, u t+1 represents the control input vector at time t+1, u t+1 The elements in contain the control instructions for the reactive power output of all inverter-based distributed power sources in the controlled power grid at time t+1. And the control instructions for reactive power output of all static VAR compensation devices In actual power grid operation, the actual value of the objective function may fluctuate due to various interferences and uncertainties. The introduction of random disturbances can help better cope with the uncertainties in the power grid. By adding random disturbances in each iteration, the algorithm can quickly respond to changes in the power grid state, improving the real-time performance of the control strategy.

[0111] Finally, control and execution instructions are sent to the control devices in the power grid (distributed power sources based on inverters and static VAR compensation devices) t+1 Then, set t=t+1 and return to step S200.

[0112] In summary, this model-free reactive voltage control method based on zero-order feedback optimization not only simplifies the implementation of the control algorithm by introducing random disturbances and zero-order feedback mechanisms, but also enhances its robustness and exploratory nature. This method avoids dependence on precise model parameters by updating the objective function value in real time and estimating the gradient based on it, so that the algorithm can still operate effectively in the face of changes in the grid structure or incomplete model accuracy. In addition, by dynamically adjusting the gradient update step size and introducing random disturbances to calculate the control input vector, this method can avoid falling into the local optimal solution, promote global optimization, and quickly respond to changes in the grid state, improving the real-time and accuracy of the control strategy. Therefore, this method has shown significant advantages in the actual grid reactive voltage control and can effectively improve the operation efficiency and stability of the grid.

[0113] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0114] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A model-free reactive voltage control method based on zero-order feedback optimization, characterized in that: The method comprises: Construct a reactive power voltage optimization model for the power grid, and obtain the reactive power voltage optimization target values ​​at different times based on the objective function; Calculate the estimated gradient value of the objective function at different times according to the reactive voltage optimization target value at different times; Extract the gradient update step size at the next target moment according to the estimated gradient value of the target function at different moments; The reactive voltage control command at the next moment is updated based on the accelerated gradient projection iterative algorithm.

2. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 1, characterized in that: Obtaining reactive power voltage optimization target values ​​at different times includes: The voltage amplitude column vectors of observable voltage nodes and the active power value column vectors of common coupling nodes at different times of the power grid are obtained, and the reactive voltage optimization target values ​​at different times are calculated according to the voltage amplitude reference values ​​of observable voltage nodes of the power grid, the preset voltage deviation penalty cost coefficients, and the preset active network loss optimization cost coefficients.

3. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 1, characterized in that: The estimated gradient values ​​of the objective function at different times are calculated according to the reactive voltage optimization target values ​​at different times, including: According to the difference between the reactive voltage optimization target values ​​at different times, the gradient estimation value of the objective function at different times is calculated based on the preset smoothing coefficient and the standard normal distribution random vector.

4. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 1, characterized in that: The gradient update step size of the next target moment is extracted according to the estimated gradient value of the target function at different moments, including: Select a target time and calculate the active constraint set of the power grid at the time before the target time; Extracting an inactive constraint set based on an active constraint set of the power grid at the previous moment; Calculate the gradient update step size at the next target moment based on the set of inactive constraints.

5. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 4, characterized in that: Select the target time and calculate the active constraint set of the power grid at the previous time of the target time, including: Obtain the control input vector of the power grid at the previous moment of the target moment and the estimated value of the objective function gradient; The active constraint set of the power grid at the previous moment is extracted based on the upper and lower limit thresholds of the control input vector and the estimated value of the gradient of the objective function at the previous moment.

6. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 5, characterized in that: The inactive constraint set extracted based on the active constraint set of the power grid at the previous moment includes: The elements in the active constraint set are removed from the control input vector of the power grid at the previous moment of the target moment, and the control input vector includes at least the control instructions of the reactive power output of the inverter-based distributed power source and the control instructions of the reactive power output of the static reactive compensation device.

7. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 6, characterized in that: The gradient update step size for calculating the next moment of the target moment based on the inactive constraint set includes: Extract the difference vector between the control input vector at the target moment and the previous moment and the objective function gradient estimate; Extracting a first element vector belonging to the inactive constraint set from the difference vector of the control input vector, and calculating the square value of the Euclidean norm of the first element vector; Extracting a second element vector belonging to the inactive constraint set from the difference vector of the estimated value of the objective function gradient; Calculate the dot product of the first element vector and the second element vector; The gradient update step size for the next moment is calculated based on the square value of the Euclidean norm of the first element vector and the dot product.

8. The model-free reactive voltage control method based on zero-order feedback optimization according to any one of claims 1 to 7, characterized in that: The reactive voltage control instructions for the next moment are updated based on the accelerated gradient projection iterative algorithm, including: Calculate candidate control variables of the power grid at the next moment of the target moment according to gradient descent; The feasible region of the control input vector at the next moment is calculated based on the candidate control variables at the next moment by using random disturbances.

9. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 8, characterized in that: The candidate control variables of the power grid at the next moment calculated according to the gradient descent target moment include: Calculate the gradient descent direction at the target moment based on the candidate control variables at the target moment, the estimated value of the objective function gradient, and the gradient update step size at the next moment of the target moment; The projection operator is used to map the gradient descent direction at the target moment to the feasible domain of the power grid control input vector at the next moment, and the candidate control variables of the power grid at the next moment are obtained.

10. The model-free reactive voltage control method based on zero-order feedback optimization according to claim 9, characterized in that: Using random disturbances, the feasible domain for calculating the control input vector at the next moment based on the candidate control variables at the next moment includes: Randomly initialize the random disturbance vector of the power grid at the next moment; Adjust the intensity of the random disturbance vector of the power grid at the next moment according to a preset positive scalar; The control input vector at the next moment is calculated according to the candidate control variables of the power grid at the next moment and the weighted random disturbance vector.

Images