Decision-making method and device for recovery of backbone network frame of wind power plant supporting power system

Abstract

The invention provides a wind power plant supporting power system backbone network frame recovery decision-making method and device. The method comprises the following steps: acquiring a real-time recovery state of a power grid, ultra-short-term prediction information of a wind power plant and a climbing rate of a thermal power generating unit; a strategy network is generated by self-learning historical data of a wind power plant, and the recovery sequence of the power transmission wind power plant and the thermal power generating unit is determined by combining Monte Carlo tree search. And adjusting the active output according to the prediction information and the climbing rate, and finally executing recovery through the station control equipment. According to the method, the recovery sequence and the active power output are dynamically adjusted by using an online decision and strategy network, the wind power uncertainty is dealt with, and the system recovery process is accelerated.

Description

Technical Field

[0001] This application belongs to the field of power restoration, and in particular relates to a decision-making method and device for restoring the backbone grid of a wind farm-supported power system. Background Technology

[0002] Although wind power generation has the advantages of fast start-up speed, low start-up power, and fast ramp-up speed, which is conducive to accelerating the system recovery process, the high uncertainty of its active power output is the most severe challenge faced in the recovery process.

[0003] Existing technologies for wind farm support grid reconfiguration primarily involve establishing offline optimization models and solving them to formulate grid reconfiguration plans based on typical wind power scenarios. While this approach can effectively address most wind power output scenarios, it cannot effectively cover all scenarios, and its optimality is difficult to guarantee.

[0004] Offline contingency planning methods cannot adapt to changes in wind power scenarios in real time and are insufficient to meet the needs of online decision-making. Summary of the Invention

[0005] The purpose of this application is to overcome the deficiencies in the prior art and provide a method and apparatus for decision-making on the restoration of the backbone grid of a wind farm-supported power system.

[0006] This application provides a decision-making method for the restoration of the backbone power grid of a wind farm-supported power system, including:

[0007] Acquire real-time grid recovery status, wind farm ultra-short-term output prediction information, and thermal power unit ramp-up rate; generate a strategy network through self-learning based on historical wind farm output scenarios and conventional unit startup parameters;

[0008] Based on the strategy network and the real-time recovery status of the power grid, Monte Carlo tree search is used to determine the recovery sequence of the transmission wind farm and thermal power unit.

[0009] Based on the ultra-short-term power output prediction information and the ramp-up rate of thermal power units, the active power output of wind farms and thermal power units is determined.

[0010] According to the restoration sequence and the active power output, the restoration of the backbone grid supporting the wind farm is carried out through the station control equipment.

[0011] Optionally, based on the historical power output scenarios of the wind farm and the start-up parameters of conventional units, a policy network is generated through self-learning, including:

[0012] The historical power output scenarios of the wind farm are divided into a preset number of typical scenario clusters using the binary k-means clustering algorithm;

[0013] Key features are obtained by performing feature dimensionality reduction on each of the typical scene clusters using a sparse autoencoder.

[0014] Based on the aforementioned key features, the recovery process of each typical scenario cluster is simulated using Monte Carlo tree search to generate the weight parameters of the policy network.

[0015] Optionally, based on the strategy network and the real-time recovery status of the power grid, the Monte Carlo tree search is used to determine the recovery order of the transmission wind farm and thermal power unit, including:

[0016] Calculate the power betweenness of the transmission line based on the importance of the node load and the proportion of generator capacity;

[0017] The weights of candidate lines in the Monte Carlo tree search are determined by the power betweenness factor, which is determined based on the node load importance and the generator capacity ratio.

[0018] Optionally, based on the ultra-short-term power output prediction information and the ramp-up rate of the thermal power units, the active power output of the wind farm and the thermal power units is determined, including:

[0019] Maximize the total output of the wind farm within the next 5 time steps;

[0020] Minimize the power output fluctuation of wind farms between adjacent time steps;

[0021] Linear constraints are applied to the ramp-up rate of thermal power units.

[0022] Optionally, according to the restoration sequence and the active power output, the restoration of the backbone grid supported by the wind farm is performed through the station control equipment, including:

[0023] The droop coefficient is calculated in real time based on the system frequency deviation.

[0024] The active power output reference value of the fan is adjusted based on the droop coefficient, and a fixed reserve capacity is reserved. The reserve capacity is determined by the product of the maximum frequency deviation per unit value and the droop coefficient.

[0025] Optionally, the Monte Carlo tree search includes:

[0026] The search tree is decomposed into multiple candidate routes, and the evaluation metrics of each candidate route are calculated in parallel by multiple threads.

[0027] The evaluation indicators are combined, and the candidate line with the highest evaluation indicator is selected as the recovery target.

[0028] Optionally, determining the active power output of the wind farm and the thermal power unit based on the ultra-short-term power output prediction information and the ramp-up rate of the thermal power unit further includes:

[0029] Calculate the wind farm's adjustability coefficient based on the dynamic difference between the wind turbine's current speed and the limit.

[0030] The wind farm output adjustment amount is allocated based on the adjustable capacity coefficient.

[0031] Optionally, before restoring the backbone grid supported by the wind farm via the station control equipment, according to the restoration sequence and the active power output, the process further includes:

[0032] Verify whether the transient process of the node voltage in the recovery path exceeds the limit;

[0033] If the limit is exceeded, the Monte Carlo tree search will be retried to generate the corrected recovery order.

[0034] This application also provides a wind farm-supported power system backbone grid restoration decision-making device, comprising:

[0035] The acquisition module acquires real-time grid recovery status, wind farm ultra-short-term output prediction information, and thermal power unit ramp-up rate.

[0036] The strategy module generates a strategy network through self-learning based on the historical power output scenarios of wind farms and the start-up parameters of conventional units.

[0037] The sequence module determines the recovery sequence of the transmission wind farm and thermal power unit based on the strategy network and the real-time recovery status of the power grid using Monte Carlo tree search.

[0038] The power output module determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term power output prediction information and the ramp rate of the thermal power unit.

[0039] The execution module, according to the restoration sequence and the active power output, performs the restoration of the backbone grid supported by the wind farm through the station control equipment.

[0040] Optionally, the strategy module generates a strategy network through self-learning based on the historical output scenarios of the wind farm and the startup parameters of conventional units, including:

[0041] The historical power output scenarios of the wind farm are divided into a preset number of typical scenario clusters using the binary k-means clustering algorithm;

[0042] Key features are obtained by performing feature dimensionality reduction on each of the typical scene clusters using a sparse autoencoder.

[0043] Based on the aforementioned key features, the recovery process of each typical scenario cluster is simulated using Monte Carlo tree search to generate the weight parameters of the policy network.

[0044] Optionally, the sequence module determines the recovery sequence of the transmission wind farm and thermal power unit using Monte Carlo tree search based on the strategy network and the real-time recovery status of the power grid, including:

[0045] Calculate the power betweenness of the transmission line based on the importance of the node load and the proportion of generator capacity;

[0046] The weights of candidate lines in the Monte Carlo tree search are determined by the power betweenness factor, which is determined based on the node load importance and the generator capacity ratio.

[0047] Optionally, the output module determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term output prediction information and the thermal power unit ramp rate, including:

[0048] Maximize the total output of the wind farm within the next 5 time steps;

[0049] Minimize the power output fluctuation of wind farms between adjacent time steps;

[0050] Linear constraints are applied to the ramp-up rate of thermal power units.

[0051] Optionally, the execution module, according to the recovery sequence and the active power output, performs the restoration of the backbone grid supported by the wind farm through the station control equipment, including:

[0052] The droop coefficient is calculated in real time based on the system frequency deviation.

[0053] The active power output reference value of the fan is adjusted based on the droop coefficient, and a fixed reserve capacity is reserved. The reserve capacity is determined by the product of the maximum frequency deviation per unit value and the droop coefficient.

[0054] Optionally, the Monte Carlo tree search includes:

[0055] The search tree is decomposed into multiple candidate routes, and the evaluation metrics of each candidate route are calculated in parallel by multiple threads.

[0056] The evaluation indicators are combined, and the candidate line with the highest evaluation indicator is selected as the recovery target.

[0057] Optionally, the power output module determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term power output prediction information and the thermal power unit ramp rate, and further includes:

[0058] Calculate the wind farm's adjustability coefficient based on the dynamic difference between the wind turbine's current speed and the limit.

[0059] The wind farm output adjustment amount is allocated based on the adjustable capacity coefficient.

[0060] Optionally, before the execution module performs the restoration of the backbone grid supported by the wind farm through the station control equipment according to the restoration sequence and the active power output, it further includes:

[0061] Verify whether the transient process of the node voltage in the recovery path exceeds the limit;

[0062] If the limit is exceeded, the Monte Carlo tree search will be retried to generate the corrected recovery order.

[0063] The beneficial effects of this application are:

[0064] This application provides a method for decision-making regarding the restoration of a power grid backbone supported by wind farms, comprising: acquiring real-time grid restoration status, short-term power output prediction information of wind farms, and ramp-up rates of thermal power units; generating a strategy network through self-learning based on historical power output scenarios of wind farms and startup parameters of conventional units; determining the restoration sequence of transmission wind farms and thermal power units using Monte Carlo tree search according to the strategy network and the real-time grid restoration status; determining the active power output of wind farms and thermal power units according to the short-term power output prediction information and the ramp-up rates of thermal power units; and executing the restoration of the wind farm-supported backbone grid through the power station control equipment according to the restoration sequence and the active power output. This application, through online decision-making and strategy network generation, fully utilizes real-time prediction information to dynamically adjust the restoration sequence and active power output of wind farms and conventional units to cope with the uncertainties of wind power and accelerate the system restoration process. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the decision-making process for the restoration of the backbone power grid of the wind farm supporting the power system in this application;

[0066] Figure 2 This is a schematic diagram of the wind farm supporting power system backbone grid restoration decision-making device in this application. Detailed Implementation

[0067] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it will be understood that various forms of implementation of the present disclosure are possible and should not be limited to the embodiments set forth herein. Rather, the embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0068] S101. Obtain real-time grid recovery status, wind farm ultra-short-term output prediction information, and thermal power unit ramp rate.

[0069] The real-time power grid recovery status includes:

[0070] The node electrical parameters, such as voltage values ​​and system frequency, have been restored.

[0071] The status of a power line, for example: a restored power line or a line that has not been restored.

[0072] Equipment operating status includes: thermal power unit status and wind farm grid connection status.

[0073] The system acquires the real-time recovery status of the power grid, as well as the dynamic parameters of thermal power units, real-time ramp rate, wind farm output prediction data, and ultra-short-term prediction values.

[0074] Furthermore, based on actual grid information, an optimization model is established that includes evaluation indicators and constraints for the restoration of the wind farm-supported backbone grid, serving as the basis for formulating the backbone grid restoration plan.

[0075] The optimized model evaluation metrics are:

[0076]

[0077] Where E is the total generating capacity of the unit, and N is the number of time periods. W P represents the number of wind farms awaiting restoration. Wi,t Let P be the active power output of the i-th wind farm at time t, where Δt represents the time interval, T is the recovery time interval, and T = N * Δt. j (t) is the output function of the j-th conventional unit, N G This represents the number of units awaiting restoration of normal operation.

[0078] The constraints are:

[0079]

[0080] Among them, U k U is the voltage value at node k. k min and U k max Here, represents the upper and lower limits of the voltage value at node k, and f is the system frequency; f min and f max These are the upper and lower limits of the system frequency, respectively; P l For the power flow of line l; P max The upper limit of the power flow of line l; t min P is the minimum startup time of unit t; Gj,t P represents the active power output of the j-th conventional unit during time period t. Gsj,t P represents the starting power of the j-th conventional unit during time period t; Gsq Let be the starting power required for the q-th unit, which starts during the t-th time period.

[0081] S102. Based on the historical power output scenarios of wind farms and the start-up parameters of conventional units, a strategy network is generated through self-learning.

[0082] The decision-making and valuation networks for wind farm-supported backbone network recovery are trained iteratively and continuously updated. The main steps are as follows:

[0083] 1. Use the binary k-means algorithm to generate Nwp typical wind farm output scenarios, setting i=1, j=1;

[0084] 2. For the i-th scenario, the old and new strategies and the valuation network are combined with Monte Carlo tree search to simulate backbone network recovery decisions, resulting in two types of network recovery schemes and E. new and E old Two evaluation metrics, let i = i + 1;

[0085] 3. Compare E new and E old Size, if E new >E old If , then j = j + 1;

[0086] 4. Repeat the above process until i = N wp until;

[0087] 5. If j > 0.5N wp If the condition is met, save the new network and the samples it generates; otherwise, save the old network and the samples it generates.

[0088] 6. Based on the sparse autoencoder training method, the retained samples are used to train the retained network to generate a new policy network and a valuation network.

[0089] 7. Repeat steps 1-6 until the computation time reaches the required time, then output the policy network.

[0090] The steps of the binary k-means clustering algorithm used for wind power scenario clustering are as follows:

[0091] 1. Treat all wind power scenarios as a cluster;

[0092] 2. The cluster is divided into two using the basic k-means clustering algorithm;

[0093] 3. Select the cluster with the most wind power scenarios and continue to use the k-means algorithm for clustering until the required number of clusters is obtained;

[0094] 4. Select the centroid of each cluster as a typical wind power scenario for subsequent calculations.

[0095] The main function of the valuation network is to quickly estimate the decision indicators of the overall recovery plan under the recovery status of a certain unit, while the main function of the strategy network is to quickly estimate the probability of a certain alternative route being selected under the recovery status of a certain unit.

[0096] Monte Carlo tree search can gradually build a local decision tree by searching the tree. This process mainly consists of four steps: selection, expansion, simulation, and backtracking.

[0097] The method uses a valuation network instead of the "simulation" part. The new and old strategies, the valuation network, and Monte Carlo tree search are combined to simulate backbone network recovery decisions. The steps are as follows:

[0098] Selection: Starting from the root node, after calculating the Integrated Upper Confidence Interval (IUCT) value for each node, the node with the largest IUCT value is selected for the next step of expansion. The formula for this indicator is as follows:

[0099]

[0100] Among them, P s The probability of node s being selected can be obtained from the estimated network; N s N represents the number of times node s is visited; c ε is the number of times the parent node of node s is visited; ε is a scaling factor; Q s The average value of the results of the child nodes of the evaluation node s in the valuation network.

[0101] Expansion: During expansion, search for the next possible power transmission line to be restored, and use the system state after the power transmission line is restored as the next level node of the Monte Carlo tree.

[0102] Backtracking: The valuation network is used to evaluate the state of each node, and the parameters of each node in the tree are updated in reverse.

[0103] The new and old strategies and valuation networks are combined with Monte Carlo tree search to simulate the backbone network recovery decision-making process.

[0104] S103. Based on the strategy network and the real-time recovery status of the power grid, Monte Carlo tree search is used to determine the recovery sequence of the transmission wind farm and thermal power unit.

[0105] In this step, the Monte Carlo tree search differs from the Monte Carlo tree search in step S02. It retains the simulation part, makes full use of real-time wind power prediction information, and simulates the backbone grid restoration process to improve the effectiveness of decision-making.

[0106] The Monte Carlo tree search steps, including the simulation part, are as follows:

[0107] Selection: Starting from the root node, selection is guided by calculating the Improved Upper Confidence Interval (MUCT) metric value for each node. The selection process follows a clear rule: prioritize nodes with the highest MUCT metric values ​​for expansion or simulation.

[0108] The formula for the MUCT index is as follows:

[0109]

[0110] in, This represents the average of the top 70% of simulation results for node c, where n is the number of times node c's parent node was visited. c Let C be the number of times node c is visited. p It is a real number greater than 0.

[0111] Extension: Branch pruning technology is employed to optimize the extension process. By reducing the number of child nodes, the algorithm's search depth and breadth are increased, and search efficiency is significantly improved. Branch pruning technology optimizes the search path by starting the search from the latest node and searching backwards, identifying and avoiding the recurrence of node states with the same parent node in subsequent extensions until a new unit is connected.

[0112] Simulation: The system state is continuously evaluated using a valuation network to quickly estimate the optimal decision metric values. This process increases the likelihood of selecting alternative routes with higher decision metric values, thus guiding the simulation process.

[0113] Backtracking: Update the parameters of each node in the tree in reverse order.

[0114] S104. Determine the active power output of the wind farm and the thermal power unit based on the ultra-short-term power output prediction information and the ramp-up rate of the thermal power unit.

[0115] Based on real-time forecast information of restored wind farms in the power grid and the ramp-up capability of thermal power plants, a rolling optimization model for the active power output of wind farms is established to calculate the active power output of each wind farm and thermal power unit in the next time step, and then distribute the calculation to each station for execution through the dispatch communication system.

[0116] The optimization objective of the optimization model mainly consists of two parts. The first part is to maximize the active power output of wind power, which is consistent with the objective of unit recovery. In addition, to further suppress the fluctuation of wind farm output, the second part is set to minimize the fluctuation of wind farm output.

[0117] The optimization objective is shown in the following formula:

[0118]

[0119] Among them, P Wi,t+k ΔP represents the active power output of the i-th wind farm in the k-th time period after the current time. Wi,t+k γ represents the difference in active power output between the k-th and (k-1)-th time periods for the i-th wind farm; γ is the control weight of the wind farm.

[0120] The constraints of the optimization model in the above steps mainly consider the active power balance constraints of the system, as shown in the following equation:

[0121]

[0122] in, The active power that the i-th wind farm can provide in the k-th time period; U t This serves as the system's overall active power reserve.

[0123] Based on the kinetic energy change between the initial speed and the speed limit of the wind turbine, an evaluation index for the wind turbine's adjustability is established. Based on the wind turbine's adjustability, an optimization model for the active power output of each wind turbine is established. The active power reference value of each wind turbine is calculated and solved, and each wind turbine is controlled according to the reference value.

[0124] The expression for the adjustable capacity parameter of the i-th wind turbine at time k is:

[0125]

[0126] Where, ω r，min Generally, 0.7 pu and ω are taken. r，max Generally, 1.2 pu is used.

[0127] The objective function for optimizing the model is:

[0128]

[0129] in, It is the change in reference active power of the i-th wind turbine. It is the reference active power variation distributed in the wind farm according to the adjustable capacity of each wind turbine, that is:

[0130]

[0131] Among them, C i (k) represents the adjustable capacity of the i-th wind turbine at time k, and C WF (k) represents the total adjustable capacity of the wind farm, i.e.:

[0132]

[0133] S105. According to the restoration sequence and the active power output, the backbone grid supported by the wind farm is restored through the station control equipment.

[0134] Based on the frequency regulation characteristics of conventional synchronous generator sets, droop control is introduced into the power limiting control of wind turbines to establish a frequency regulation controller, which adjusts the active power output of wind turbines in real time according to the active power reference value and system frequency changes.

[0135] Power limiting control consists of two parts: speed control and pitch control. Speed ​​control is used in low-to-medium wind conditions, while pitch control is used in high-wind conditions.

[0136] Based on power limiting control, droop control is introduced. By simulating the power-frequency static characteristic curve of a synchronous generator's primary frequency regulation, an active power change ΔP proportional to the frequency deviation is introduced into the active power reference value. The active power output of the wind turbine is changed according to the frequency deviation, as shown in the following formula:

[0137] ΔP=-RΔf

[0138] Where R is the droop coefficient and Δf is the frequency deviation.

[0139] Based on the power limiting control strategy, the wind turbine reserves a fixed reserve capacity Pmar to ensure that the wind turbine has primary frequency regulation capability at any time. The fixed reserve capacity Pmar is obtained based on its set droop coefficient R, and its calculation formula is shown below.

[0140]

[0141] Where, σ g % represents the droop coefficient of synchronous generators in the power system, and σ represents the droop coefficient. w % represents the simulation droop factor for the wind farm, K f K refers to the ratio of the frequency regulation capability of a wind farm to that of a synchronous generator. f The larger the value, the lower the frequency regulation requirement for the wind farm. This paper takes K as the value. f =2; This refers to the per-unit value of the fixed reserve capacity of wind turbine units. This is the per-unit value of the maximum allowable frequency deviation of the system; This is the per-unit value for the minimum stable output of the fan; The active power reference per unit value is reserved for the wind turbine after it has been made available for backup.

[0142] During the backbone network restoration decision-making process, the above S101-S105 need to be executed in a rolling manner, which determines the order of line restoration in the final backbone network reconfiguration scheme obtained by this restoration decision-making framework.

[0143] like Figure 2 As shown, this application provides a wind farm-supported power system backbone grid restoration decision-making device, comprising:

[0144] Module 201 acquires the real-time recovery status of the power grid, the ultra-short-term output prediction information of wind farms, and the ramp-up rate of thermal power units.

[0145] Strategy module 202 generates a strategy network through self-learning based on the historical power output scenarios of wind farms and the start-up parameters of conventional units;

[0146] Sequence module 203 determines the recovery sequence of transmission wind farms and thermal power units by using Monte Carlo tree search based on the strategy network and the real-time recovery status of the power grid.

[0147] The output module 204 determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term output prediction information and the ramp rate of the thermal power unit.

[0148] The execution module 205 performs the restoration of the backbone grid supported by the wind farm through the station control equipment according to the restoration sequence and the active power output.

[0149] Furthermore, the strategy module generates a strategy network through self-learning based on the historical output scenarios of the wind farm and the startup parameters of conventional units, including:

[0150] The historical power output scenarios of the wind farm are divided into a preset number of typical scenario clusters using the binary k-means clustering algorithm;

[0151] Key features are obtained by performing feature dimensionality reduction on each of the typical scene clusters using a sparse autoencoder.

[0152] Based on the aforementioned key features, the recovery process of each typical scenario cluster is simulated using Monte Carlo tree search to generate the weight parameters of the policy network.

[0153] Furthermore, the sequence module determines the recovery sequence of the transmission wind farm and thermal power unit using Monte Carlo tree search based on the strategy network and the real-time recovery status of the power grid, including:

[0154] Calculate the power betweenness of the transmission line based on the importance of the node load and the proportion of generator capacity;

[0155] The weights of candidate lines in the Monte Carlo tree search are determined by the power betweenness factor, which is determined based on the node load importance and the generator capacity ratio.

[0156] Furthermore, the output module determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term output prediction information and the thermal power unit ramp rate, including:

[0157] Maximize the total output of the wind farm within the next 5 time steps;

[0158] Minimize the power output fluctuation of wind farms between adjacent time steps;

[0159] Linear constraints are applied to the ramp-up rate of thermal power units.

[0160] Furthermore, the execution module, based on the recovery sequence and the active power output, performs the restoration of the backbone grid supported by the wind farm through the station control equipment, including:

[0161] The droop coefficient is calculated in real time based on the system frequency deviation.

[0162] The active power output reference value of the fan is adjusted based on the droop coefficient, and a fixed reserve capacity is reserved. The reserve capacity is determined by the product of the maximum frequency deviation per unit value and the droop coefficient.

[0163] Furthermore, the Monte Carlo tree search includes:

[0164] The search tree is decomposed into multiple candidate routes, and the evaluation index of each candidate route is calculated in parallel by multiple threads. The evaluation indexes are merged, and the candidate route with the highest evaluation index is selected as the recovery target.

[0165] Furthermore, the power output module, based on the ultra-short-term power output prediction information and the thermal power unit ramp rate, determines the active power output of the wind farm and the thermal power unit, and also includes:

[0166] Calculate the wind farm's adjustability coefficient based on the dynamic difference between the wind turbine's current speed and the limit.

[0167] The wind farm output adjustment amount is allocated based on the adjustable capacity coefficient.

[0168] Furthermore, before the execution module performs the restoration of the backbone grid supported by the wind farm through the station control equipment according to the restoration sequence and the active power output, it also includes:

[0169] Verify whether the transient process of the node voltage in the recovery path exceeds the limit;

[0170] If the limit is exceeded, the Monte Carlo tree search will be retried to generate the corrected recovery order.

[0171] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A decision-making method for restoring the backbone power grid of a wind farm-supported power system, characterized in that, include: Acquire real-time grid recovery status, wind farm ultra-short-term output forecast information, and thermal power unit ramp-up rate; Based on the historical power output scenarios of wind farms and the start-up parameters of conventional units, a strategy network is generated through self-learning. Based on the strategy network and the real-time recovery status of the power grid, Monte Carlo tree search is used to determine the recovery sequence of the transmission wind farm and thermal power unit. Based on the ultra-short-term power output prediction information and the ramp-up rate of thermal power units, the active power output of wind farms and thermal power units is determined. According to the restoration sequence and the active power output, the restoration of the backbone grid supporting the wind farm is carried out through the station control equipment.

2. The decision-making method for restoring the backbone grid of a wind farm-supported power system according to claim 1, characterized in that, Based on historical wind farm output scenarios and conventional turbine startup parameters, a policy network is generated through self-learning, including: The historical power output scenarios of the wind farm are divided into a preset number of typical scenario clusters using the binary k-means clustering algorithm; Key features are obtained by performing feature dimensionality reduction on each of the typical scene clusters using a sparse autoencoder. Based on the aforementioned key features, the recovery process of each typical scenario cluster is simulated using Monte Carlo tree search to generate the weight parameters of the policy network.

3. The decision-making method for restoring the backbone grid of a wind farm-supported power system according to claim 1, characterized in that, Based on the strategy network and the real-time recovery status of the power grid, the Monte Carlo tree search is used to determine the recovery sequence of transmission wind farms and thermal power units, including: Calculate the power betweenness of the transmission line based on the importance of the node load and the proportion of generator capacity; The weights of candidate lines in the Monte Carlo tree search are determined by the power betweenness factor, which is determined based on the node load importance and the generator capacity ratio.

4. The decision-making method for restoring the backbone grid of a wind farm-supported power system according to claim 1, characterized in that, Based on the ultra-short-term power output prediction information and the ramp-up rate of thermal power units, the active power output of wind farms and thermal power units is determined, including: Maximize the total output of the wind farm within the next 5 time steps; Minimize the power output fluctuation of wind farms between adjacent time steps; Linear constraints are applied to the ramp-up rate of thermal power units.

5. The decision-making method for restoring the backbone grid of a wind farm-supported power system according to claim 1, characterized in that, According to the restoration sequence and the active power output, the restoration of the backbone grid supported by the wind farm is performed through the station control equipment, including: The droop coefficient is calculated in real time based on the system frequency deviation. The active power output reference value of the fan is adjusted based on the droop coefficient, and a fixed reserve capacity is reserved. The reserve capacity is determined by the product of the maximum frequency deviation per unit value and the droop coefficient.

6. The decision-making method for restoring the backbone grid of a wind farm-supported power system according to claim 1, characterized in that, The Monte Carlo tree search includes: The search tree is decomposed into multiple candidate routes, and the evaluation metrics of each candidate route are calculated in parallel by multiple threads. The evaluation indicators are combined, and the candidate line with the highest evaluation indicator is selected as the recovery target.

7. The decision-making method for restoring the backbone grid of a wind farm-supported power system according to claim 4, characterized in that, Determining the active power output of wind farms and thermal power units based on the aforementioned ultra-short-term power output prediction information and the ramp-up rate of thermal power units also includes: Calculate the wind farm's adjustability coefficient based on the dynamic difference between the wind turbine's current speed and the limit. The wind farm output adjustment amount is allocated based on the adjustable capacity coefficient.

8. The decision-making method for restoring the backbone grid of a wind farm-supported power system according to claim 1, characterized in that, According to the restoration sequence and the active power output, before the restoration of the backbone grid supported by the wind farm is performed through the station control equipment, the following steps are also included: Verify whether the transient process of the node voltage in the recovery path exceeds the limit; If the limit is exceeded, the Monte Carlo tree search will be retried to generate the corrected recovery order.

9. A decision-making device for restoring the backbone grid of a wind farm-supported power system, characterized in that, include: The acquisition module acquires real-time grid recovery status, wind farm ultra-short-term output prediction information, and thermal power unit ramp-up rate. The strategy module generates a strategy network through self-learning based on the historical power output scenarios of wind farms and the start-up parameters of conventional units. The sequence module determines the recovery sequence of the transmission wind farm and thermal power unit based on the strategy network and the real-time recovery status of the power grid using Monte Carlo tree search. The power output module determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term power output prediction information and the ramp rate of the thermal power unit. The execution module, according to the restoration sequence and the active power output, performs the restoration of the backbone grid supported by the wind farm through the station control equipment.

10. The wind farm-supported power system backbone grid restoration decision-making device according to claim 9, characterized in that, The strategy module generates a strategy network through self-learning based on historical wind farm output scenarios and conventional unit startup parameters, including: The historical power output scenarios of the wind farm are divided into a preset number of typical scenario clusters using the binary k-means clustering algorithm; Key features are obtained by performing feature dimensionality reduction on each of the typical scene clusters using a sparse autoencoder. Based on the aforementioned key features, the recovery process of each typical scenario cluster is simulated using Monte Carlo tree search to generate the weight parameters of the policy network.

11. The wind farm-supported power system backbone grid restoration decision-making device according to claim 9, characterized in that, The sequence module determines the recovery sequence of transmission wind farms and thermal power units using Monte Carlo tree search based on the strategy network and the real-time recovery status of the power grid, including: Calculate the power betweenness of the transmission line based on the importance of the node load and the proportion of generator capacity; The weights of candidate lines in the Monte Carlo tree search are determined by the power betweenness factor, which is determined based on the node load importance and the generator capacity ratio.

12. The wind farm-supported power system backbone grid restoration decision-making device according to claim 9, characterized in that, The power output module determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term power output prediction information and the thermal power unit ramp rate, including: Maximize the total output of the wind farm within the next 5 time steps; Minimize the power output fluctuation of wind farms between adjacent time steps; Linear constraints are applied to the ramp-up rate of thermal power units.

13. The wind farm-supported power system backbone grid restoration decision-making device according to claim 9, characterized in that, The execution module, based on the recovery sequence and the active power output, performs the restoration of the backbone grid supported by the wind farm through the station control equipment, including: The droop coefficient is calculated in real time based on the system frequency deviation. The active power output reference value of the fan is adjusted based on the droop coefficient, and a fixed reserve capacity is reserved. The reserve capacity is determined by the product of the maximum frequency deviation per unit value and the droop coefficient.

14. The wind farm-supported power system backbone grid restoration decision-making device according to claim 9, characterized in that, The Monte Carlo tree search includes: The search tree is decomposed into multiple candidate routes, and the evaluation metrics of each candidate route are calculated in parallel by multiple threads. The evaluation indicators are combined, and the candidate line with the highest evaluation indicator is selected as the recovery target.

15. The wind farm-supported power system backbone grid restoration decision-making device according to claim 12, characterized in that, The power output module determines the active power output of the wind farm and the thermal power unit based on the ultra-short-term power output prediction information and the thermal power unit ramp rate, and also includes: Calculate the wind farm's adjustability coefficient based on the dynamic difference between the wind turbine's current speed and the limit. The wind farm output adjustment amount is allocated based on the adjustable capacity coefficient.

16. The wind farm-supported power system backbone grid restoration decision-making device according to claim 9, characterized in that, Before the execution module performs the restoration of the backbone grid supported by the wind farm through the station control equipment according to the restoration sequence and the active power output, it also includes: Verify whether the transient process of the node voltage in the recovery path exceeds the limit; If the limit is exceeded, the Monte Carlo tree search will be retried to generate the corrected recovery order.

Images