A county distributed photovoltaic cluster hierarchical AGC optimization scheduling method based on irradiance margin factor

Abstract

The application discloses a county distributed photovoltaic cluster hierarchical AGC optimization scheduling method based on irradiation margin factors, and comprises the following steps: S1. County power distribution network topology modeling and photovoltaic cluster division; S2. Irradiation margin factor definition and dynamic calculation; S3. Mapping of irradiation margin factors to three-level AGC constraints; S4. Three-level time scale coordinated AGC architecture execution; S5. Cloud shadow spatial correlation modeling and robust power allocation; S6. Abnormal degradation strategy and federal auxiliary calibration. The application solves the problems that traditional AGC parameters are not suitable for photovoltaics, multi-level scheduling constraints are disconnected, and distribution strategies fail under spatial correlation fluctuations.

Description

Technical Field

[0001] This invention relates to the field of power system automatic control technology, specifically to a hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor. Background Technology

[0002] With the continuous growth of distributed photovoltaic (PV) capacity in county-level power distribution networks, PV penetration in some counties has exceeded 80% of distribution transformer capacity. While large-scale distributed PV grid connection increases the proportion of clean energy, it also poses a severe challenge to the active power balance regulation of the power grid. PV output is affected by fluctuations in irradiance, exhibiting rapid changes on the order of seconds or even sub-seconds, and the PV output of adjacent transformer areas within the same county shows significant spatial correlation due to cloud shading effects.

[0003] Existing Automatic Generation Control (AGC) technologies primarily target conventional thermal and hydropower units with rotating inertia and bidirectional continuous regulation capabilities. Their regulation cycles are mainly on the order of minutes, and adjustable capacity is characterized by parameters such as ramp rate, frequency dead zone, and response delay. However, distributed photovoltaic inverters lack rotating inertia. Their only way to participate in active power regulation is to reserve a reduction margin below the maximum power point and achieve downward regulation through limiting control, making traditional parameter systems unsuitable. At the dispatch architecture level, existing solutions mostly focus on two-level coordination between the provincial government and power plants, failing to achieve effective connection from minute-level macro-instructions to sub-second-level local balance at the county spatial scale. Furthermore, the boundaries of dispatch constraints at each level are prone to disconnection. Regarding power allocation strategies, traditional AGC treats each regulation unit as an independent object, neglecting spatial correlation caused by cloud shading effects. In scenarios with spatially correlated fluctuations, allocation coefficients become ineffective, easily leading to AGC command overshoot and frequency oscillations.

[0004] Therefore, there is an urgent need for a county-level distributed photovoltaic cluster optimization scheduling method that can characterize the real-time adjustable range of photovoltaics, achieve effective connection of multiple time scales, and adapt to spatially related fluctuation scenarios. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor, so as to solve the problems that traditional AGC parameters are not applicable to photovoltaics, multi-level scheduling constraints are disconnected, and allocation strategies fail under spatially related fluctuations.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows.

[0007] A hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor includes the following steps: S1. County-level power distribution network topology modeling and photovoltaic cluster division: Establish the four-level object mapping relationship of county-level power distribution network, namely county level, feeder level, transformer area level and inverter level, and construct the photovoltaic aggregated output model of each level; S2. Definition and dynamic calculation of irradiance margin factor: Define the irradiance margin factor of a single inverter and its adjustable capacity upper limit, design a dynamic update mechanism including short-term irradiance prediction correction and rapid reset of irradiance mutation, and perform multi-level aggregation calculation to obtain the adjustable capacity upper limit of the substation area, feeder and county level. S3. Mapping of Irradiation Margin Factor to Level 3 AGC Constraints: The irradiation margin factor is mapped to the capacity constraint of the minute-level instruction decomposition layer, the upper and lower bound constraints of the power adjustment of the second-level model prediction control layer, and the tracking saturation boundary of the sub-second-level station balance layer, respectively. S4. Three-level time-scale coordinated AGC architecture execution: At the minute-level layer, the main grid AGC signal is received and weighted and decomposed to the county level according to the irradiance margin factor; at the second-level layer, the county-level target is decomposed to the feeder level through model predictive control rolling optimization; at the sub-second-level layer, the feeder target is allocated to the inverter through consistency control. S5. Cloud shading spatial correlation modeling and robust power allocation: Based on the geographical distance of the transformer area and the cloud movement speed, a spatial correlation coefficient matrix is ​​established, an ellipsoidal uncertainty set is constructed, and the robust power allocation coefficient that satisfies the frequency deviation constraint is solved. S6. Anomaly Degradation Strategy and Federated Auxiliary Calibration: A graded degradation strategy is designed for anomaly scenarios, and the irradiance margin factor prediction model is periodically calibrated using a federated learning framework.

[0008] Preferably, in step S2, the irradiance margin factor of a single inverter is... Defined as:

[0009] in, For the first feeder line number The first district Taiwan inverter The irradiance margin factor at time t has a value range of [0,1]. For the first feeder line number The first in the district Taiwanese photovoltaic inverters The maximum power output at a given moment; For the first feeder line number The first in the district Taiwanese photovoltaic inverters The actual contribution at any given moment; The rated active capacity of the inverter; Define the upper limit of the adjustable capacity of a single inverter. for: .

[0010] Preferably, the dynamic update mechanism in step S2 specifically includes: Real-time value update: Real-time value of irradiance margin factor Updated periodically according to the definition; Predicted values ​​updated: Predicted values ​​for the irradiance margin factor This is obtained by introducing a short-term irradiation prediction correction term, based on the predicted maximum power point output. The calculation is as follows:

[0011] in, To predict the time step; Irradiation mutation detection and fast reset: Defining the rate of irradiation change for:

[0012] in, for The rate of change of irradiance at any given time; For the first feeder line number The first district Location of Taiwan inverter The radiation intensity at a given time, expressed in W / m². ; The sampling interval; when When an irradiation mutation is detected, a fast reset is triggered: the current time is used directly. The real-time measured values ​​replace the predicted values ​​to update the irradiance margin factor, and a margin update signal is sent to the upper layer. The threshold for determining irradiation mutations; The multi-level aggregation calculation method in step 2 specifically includes: District-level aggregation: Upper limit of adjustable capacity at the district level The sum of the upper bounds of the adjustable capacity of each inverter, and the substation-level irradiance margin factor. Calculated using a capacity-weighted average; Feeder-level aggregation: Upper limit of adjustable capacity at the feeder level It is the sum of the upper limits of the adjustable capacity of each subordinate transformer station; County-level aggregation: Upper limit of adjustable capacity at the county level It is the sum of the upper limits of the adjustable capacity of each feeder under its jurisdiction; The predicted values ​​of the upper bounds of adjustable capacity at each level are aggregated using the same method, and the irradiance margin factor is predicted based on each inverter. Obtained through calculation.

[0013] Preferably, in step S3, the three-level mapping specifically includes: Minute-level mapping: County-level power adjustment meets requirements ,in, To allocate to the county The amount of power adjustment; For the county exist The upper bound of the adjustable capacity is predicted based on the predicted irradiance margin factor. Define county Irradiation margin factor weighted decomposition weights for:

[0014] In the formula, To allocate to the county The weighting coefficients are determined according to the proportion of the upper limit of the adjustable capacity forecast of each county to the total. This is the sum of the predicted upper bounds of the adjustable capacity for all counties involved in the adjustment process; Second-level mapping: Feeder power adjustment meets requirements ,in, For the first The feeder in the model predictive control Power adjustment amount per prediction step; The rolling step size for model predictive control; For the first One feeder in The upper bound of the adjustable capacity at any given time is calculated by aggregating the predicted irradiance margin factors of each substation under the feeder line. Sub-second mapping: Inverter power adjustment meets... ,in, For the first Taiwan inverter The amount of power adjustment at any given time; For this inverter in The real-time upper limit of the adjustable capacity at any given time.

[0015] Preferably, in step S4: The minute-level layer employs a bounded decomposition method weighted by the irradiation margin factor: ,in, To allocate to the county The power adjustment target; The weights are assigned to the irradiation margin factor defined in step 3. The regional control deviation signal issued by the main network AGC; The second-level layer aims to minimize the frequency deviation and uses the irradiance margin factor as a rolling constraint to establish a model predictive control optimization problem. The sub-second level layer employs a consistency control law with communication delay compensation to distribute power adjustment reference values ​​to each inverter.

[0016] Preferably, the bounded decomposition method of irradiation margin factor weighting in step 4 also satisfies the following constraints:

[0017] Among them, the equality constraint guarantees the total power adjustment. Regional control deviation signal issued by the main network Balance; inequality constraints ensure that the allocation to each county does not exceed the upper bound of its adjustable capacity prediction value. ; when season And report the shortfall to the main network dispatcher; In the model predictive control optimization problem established at the second-level layer in step 4: Define the time domain length of the model predictive control prediction as: Step, control the time domain length as Step, rolling step length is The model predictive control objective function is:

[0018] in, For the county In the The power adjustment reference value for each prediction step is expanded by minute-level instructions using linear interpolation or step hold. For all within the county M The feeder in the future The sum of power adjustments at each moment; For the first The increment of the power adjustment of the feeder; For tracking deviation weighting matrix; For the first The feeder in the future The power adjustment increment for each control step; To control the incremental weighting matrix; This represents the weighted quadratic norm.

[0019] The constraints include:

[0020] Among them, the first set of constraints is the irradiance margin factor mapping constraint established in step 3, which ensures that the power adjustment of each feeder does not exceed the upper bound prediction value of the adjustable capacity. For the first The maximum power change rate constraint of a single feeder within a single model prediction control step; The sub-second-level real-time balancing of station consistency in step 4 adopts a consistency control law with communication delay compensation:

[0021] in, For the first Time derivative of the consistency state variable of the inverter; For the first The neighbor set of the inverter in the communication topology; The adjacency weights of the communication topology; For the neighbors Delay The state variables after that, For inverters and Communication delay between them; This is the uniformity gain coefficient; For reference tracking gain coefficient; For the reference adjustment ratio at the district level, by Calculated; The convergence condition of the consistency control law is: when hour, This holds true for all inverters, provided that the communication topology is connected and Satisfy the time delay stability condition:

[0022] in, The maximum node degree in the communication topology; This represents the maximum communication delay.

[0023] Preferably, in step S5, the ellipsoidal uncertainty set is defined as:

[0024] in, for An ellipsoidal set of uncertainties at time points; The power output fluctuation vector for each transformer substation; for The output fluctuation covariance matrix at time t, This is a diagonal matrix composed of the standard deviations of the power output fluctuations of each transformer area. for The spatial correlation coefficient matrix at time points; For uncertain budget parameters; The total number of stations participating in AGC regulation; The robust power amortization coefficient is a minimax robust optimization problem: The constraints are:

[0025] in, Taiwan District The apportionment coefficient; Taiwan District The power adjustment reference value; Taiwan District The output fluctuation; This is a regional control deviation signal; Taiwan District The upper limit of the adjustable capacity.

[0026] Preferably, the elements in the spatial correlation coefficient matrix are established using the following formula:

[0027] in, Taiwan District Taiwan exist The spatial correlation coefficient of photovoltaic output at time t, with a value range of [0,1]; Taiwan District Taiwan Geographical distance, in km; for The speed of cloud movement at any given time, expressed in km / h; The direction of cloud movement and the area Pointing to the station area The angle between directions; For spatial attenuation scale parameters; This is the time delay attenuation scale parameter; The robust power sharing coefficient is solved by utilizing the duality of the ellipsoidal uncertainty set, transforming the minimax robust optimization problem into a deterministic second-order cone programming problem:

[0028] The constraints include:

[0029] in, As an auxiliary variable, it represents the maximum frequency deviation proxy amount under the most unfavorable scenario; This is the vector of robust power amortization coefficients; Covariance matrix Cholesky decomposition factor; A vector composed of power adjustment reference values ​​for each transformer area; Taiwan District The apportionment coefficient; For the first One standard basis vector; It is the Euclidean norm.

[0030] Preferably, in step S6, the abnormal scenarios include communication interruption, sudden irradiation changes, and inverter failure to operate. The graded degradation strategy for each abnormal scenario specifically includes: County-level communication interruption degradation: When the communication link between the county-level dispatch center and the feeder controller is interrupted for more than a threshold time, the feeder is degraded to the local model predictive control autonomous mode; Irradiation mutation downgrade: When the rate of irradiation change exceeds the threshold for irradiation mutation judgment, the irradiation margin factor is quickly reset. Inverter failure to operate and degradation: When the area edge controller detects that the actual output deviation of the inverter exceeds the allowable range, it determines that the inverter has failed to operate and removes it from the consistency topology; The federally assisted calibration mechanism in step S6 includes: Each substation maintains a local irradiance margin factor prediction model. Calculate the local gradient within each calibration period. ,in, For the first feeder line number Gradient parameters of the local model in each transformer area; For local loss functions; For local model parameters; Upload the gradient to the county-level scheduling center for federated aggregation:

[0031] in, These are global model parameters; The learning rate; For the first feeder line number The number of inverters in each distribution zone is used as the aggregation weight; The total number of inverters in all participating districts within the county; The aggregated global model parameters are distributed to each station area to update the local model.

[0032] Preferably, step S6 is followed by step S7: execution of the overall closed-loop scheduling process: forming a complete closed-loop scheduling process from the main grid area control deviation signal to the power output adjustment of the transformer area inverter, the specific process is as follows: The mainnet AGC main site operates on a cyclical basis. Send regional control deviation signal To each county; County-level dispatch terminal receiving And calculate the power adjustment targets for each county. This is then converted into power adjustment reference values ​​for each feeder and sent to each feeder controller; Each feeder controller operates in a periodic manner. The model predictive control rolling optimization is executed, the power adjustment amount of each feeder is output, and the reference value of power adjustment for each transformer area is calculated by combining the robust amortization coefficient and sent to the transformer area edge controller; Each area edge controller periodically The consistency control law is executed to distribute the reference value to each inverter and output the power command; Each inverter performs limiting control and feeds back the actual output to the distribution area edge controller. The distribution area edge controller updates the real-time value of the irradiance margin factor and reports it level by level, forming a complete closed-loop scheduling.

[0033] Due to the adoption of the above technical solutions, the technical progress achieved by this invention is as follows.

[0034] This invention proposes an irradiance margin factor as a normalized adjustable domain description index that integrates irradiance conditions, maximum power point margin, and inverter capacity constraints, and defines the corresponding physical quantity of the upper bound of adjustable capacity. This index system is specifically designed for photovoltaic active power control scenarios under reserved reduction margin mode, and can accurately characterize the active power adjustable range of photovoltaic clusters at all levels in real time, overcoming the limitation that traditional frequency regulation delay classification parameters are not applicable to photovoltaic inverters without inertia.

[0035] This invention establishes a three-level mapping relationship from irradiance margin factor to minute-level capacity constraints, second-level power adjustment constraints, and sub-second-level tracking saturation boundary. It also constructs a hierarchical AGC architecture for county-level distributed photovoltaic systems, with the irradiance margin factor serving as the constraint link connecting all three levels. This architecture enables scheduling optimization at each level to share a unified source of constraints, ensuring that power adjustment commands issued by upper levels do not exceed the actual adjustable range of the lower-level photovoltaic clusters, thus resolving the problem of disconnected constraint boundaries between different levels in existing multi-timescale frameworks.

[0036] This invention establishes a spatial correlation coefficient matrix based on station spacing and cloud movement speed, and constructs an ellipsoidal set of most unfavorable fluctuation uncertainties. The robust power allocation coefficient is solved using frequency deviation constraints as hard constraints. This method can still guarantee frequency quality requirements under the most unfavorable spatial correlation fluctuation scenario, effectively avoiding AGC command overshoot and frequency oscillations under traditional allocation strategies.

[0037] This invention designs a tiered degradation strategy for abnormal scenarios such as communication interruptions, sudden changes in irradiance, and inverter failure to operate, ensuring the control continuity of the hierarchical AGC architecture under abnormal operating conditions. The federated learning-assisted calibration mechanism continuously improves the prediction accuracy of the irradiance margin factor while protecting the data privacy of each distribution area, further enhancing the long-term operational reliability of the scheduling system. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a diagram of the architecture of the present invention; Figure 3 This is the irradiation margin factor calculation and AGC command allocation process of the present invention; Figure 4 This is a schematic diagram of cloud shading spatial correlation modeling and robust power allocation in this invention. Detailed Implementation

[0039] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0040] A hierarchical AGC optimization scheduling method for county-level distributed photovoltaic (PV) clusters based on irradiance margin factors is proposed. Based on the county-level distribution network topology and PV cluster division, an irradiance margin factor is defined and its dynamic calculation method is established. Then, a mapping relationship between the irradiance margin factor and the three-level AGC constraint boundary is established. On this constraint transmission chain, a three-level coordinated AGC architecture at the minute, second, and sub-second levels is constructed. Simultaneously, a robust allocation strategy based on cloud shading spatial correlation and an anomaly degradation mechanism are introduced to ensure the robustness and continuity of scheduling, ultimately forming a complete closed loop from the main grid area control deviation signal to the inverter output adjustment in the distribution area. Combined with... Figures 1 to 3 As shown, it includes the following steps: S1. County-level power distribution network topology modeling and photovoltaic cluster division.

[0041] Establish a four-level object mapping relationship for the county-level distribution network: county level, feeder level, transformer area level, and inverter level, and construct photovoltaic aggregated output models for each level; specifically, the following steps are included: S11. Establish a four-level object mapping relationship for the county-level power distribution network.

[0042] Assume the county-level power distribution network includes feeder line, number feeder line The first district, the feeder line number Each area includes Taiwanese photovoltaic inverter. Define a four-level object set: County-level object set. Feeder layer object collection Set of objects in the platform area ( Inverter layer object collection ( ).

[0043] in, This represents the total number of feeders in the county's power distribution network. For the first The number of substations under the jurisdiction of the feeder line; For the first feeder line number Number of photovoltaic inverters in each distribution area; A collection of objects at the county level; A collection of feeder layer objects; For the first A collection of feeder area layer objects; For the first feeder line number A collection of inverter layer objects for each transformer substation.

[0044] S12. Establish the aggregated output model of photovoltaic clusters in each transformer area.

[0045] No. feeder line number The first in the district Taiwanese photovoltaic inverters The actual active power output at any given moment is Its maximum power output is The rated active capacity of the inverter is In the reserved reduction margin mode, the actual operating point of the photovoltaic inverter is lower than the maximum power point, satisfying:

[0046] in, For the first feeder line number The first district Taiwan inverter The actual contribution at any given moment; For this inverter in Maximum power output under constant irradiation conditions; This is the rated active capacity of the inverter.

[0047] The power output of the district-level aggregated power is:

[0048] in, For the first feeder line number Each district The convergence of moments contributes to the effort.

[0049] The feeder-level aggregation output is:

[0050] in, For the first One feeder in The convergence of moments contributes to the effort.

[0051] S2. Definition and dynamic calculation of irradiation margin factor.

[0052] Define the irradiance margin factor and its adjustable capacity upper limit for a single inverter, design a dynamic update mechanism that includes short-term irradiance prediction correction and rapid reset for irradiance abrupt changes, and perform multi-level aggregation calculations to obtain the adjustable capacity upper limits for the substation area, feeder, and county level; specifically, the following steps are included: S21. Define the irradiance margin factor for a single inverter.

[0053] No. feeder line number The first district Taiwan inverter Irradiance margin factor at time Defined as:

[0054] in, For the first feeder line number The first district Taiwan inverter The irradiance margin factor at time [0,1] takes values ​​in the range [0,1]; molecule The power difference between the inverter's current operating point and its maximum power point represents the current downward adjustable active power margin; the denominator is... The maximum active power output that the inverter can actually achieve under the current irradiation conditions is taken as the smaller value between the maximum power point output and the rated capacity.

[0055] Define the upper limit of the adjustable capacity of a single inverter. for:

[0056] In the formula, For the first feeder line number The first district Taiwan inverter The upper limit of the adjustable capacity at any given time, in kW, represents the absolute value of the maximum active power that the inverter can currently adjust downwards.

[0057] S22. Design a dynamic update mechanism for the irradiation margin factor.

[0058] ① Real-time value update: Real-time value of irradiance margin factor Update the data periodically according to the definition in step 21.

[0059] ②Predicted value update: Predicted value of irradiance margin factor This is obtained by introducing a short-term irradiation prediction correction term, based on the predicted maximum power point output. The calculation is as follows:

[0060] in, for Always Predicted value of the moment-to-moment irradiance margin factor; This is based on short-term irradiance prediction. Predicted output value at the maximum power point at any given time; To predict the time step.

[0061] ③ Irradiation mutation detection and fast reset: Design the logic for irradiation mutation detection and fast reset, and define the irradiation change rate. for:

[0062] in, for The rate of change of irradiance at any given time; For the first feeder line number The first district Location of Taiwan inverter The radiation intensity at a given time, expressed in W / m². ; The sampling interval. When When this is detected as an irradiation mutation, a fast reset is triggered: directly using... The irradiance margin factor is updated in real time, replacing the predicted value, and a margin update signal is sent to the upper layer. The threshold for determining irradiation mutations.

[0063] S23. Establish a multi-level aggregation calculation method for the upper bound of the adjustable active capacity of photovoltaic clusters.

[0064] ① District-level aggregation: Upper limit of adjustable capacity at the district level The sum of the upper bounds of the adjustable capacity of each inverter, and the substation-level irradiance margin factor. The capacity-weighted average is used for calculation. Details are as follows: Upper limit of adjustable capacity at the substation level for:

[0065] in, For the first feeder line number Each district The upper bound of the adjustable capacity at any given time.

[0066] District-level irradiance margin factor Defined as a capacity-weighted average:

[0067] in, For the first feeder line number Each district The aggregate irradiance margin factor at any given time.

[0068] ② Feeder-level aggregation: Upper limit of adjustable capacity at the feeder level This is the sum of the upper limits of the adjustable capacity of all subordinate transformer substations. Details are as follows: upper limit of adjustable capacity at the feeder level for:

[0069] in, For the first One feeder in The upper bound of the adjustable capacity at any given time.

[0070] ③ County-level aggregation: Upper limit of adjustable capacity at the county level This is the sum of the upper limits of the adjustable capacities of all subordinate feeders. Details are as follows: County-level adjustable capacity upper limit for:

[0071] in, For the county area The upper limit of the total adjustable capacity at any given time.

[0072] Correspondingly, the predicted values ​​of the upper limits of adjustable capacity at each level Using the same aggregation method, predict the irradiance margin factor for each inverter in step 22. It is obtained from basic calculations and is used for minute-level and second-level scheduling.

[0073] S3. Mapping of irradiation margin factor to third-level AGC constraint.

[0074] The irradiance margin factor is mapped to the capacity constraint of the minute-level command decomposition layer, the upper and lower bound constraints of the power adjustment of the second-level model prediction control layer, and the tracking saturation boundary of the sub-second-level station balancing layer, respectively. The three-level mapping specifically includes: minute-level mapping, second-level mapping, and sub-second-level mapping. This step specifically includes the following steps: S31. Establish the mapping from the irradiation margin factor to the minute-level instruction decomposition layer.

[0075] Minute-level layer receives regional control deviation signals issued by the main network AGC. This needs to be broken down into power adjustment targets for each county. Define the county. The capacity constraint in minute-level instruction decomposition is:

[0076] in, To allocate to the county The amount of power adjustment; For the county exist The upper bound of the adjustable capacity is calculated based on the predicted irradiance margin factor. This constraint ensures that the power adjustment instructions allocated to each county do not exceed the predicted adjustable range of its photovoltaic cluster.

[0077] Define county Irradiation margin factor weighted decomposition weights for:

[0078] in, To allocate to the county The weighting coefficients are determined according to the proportion of the upper limit of the adjustable capacity forecast of each county to the total. It is the sum of the predicted upper bounds of the adjustable capacity of all counties involved in the adjustment.

[0079] S32. Establish the mapping of irradiation margin factor to second-level MPC layer.

[0080] Within each county, the second-level MPC uses the upper bound prediction value of the feeder-level adjustable capacity as the upper and lower bound constraints for power adjustment. The feeder in the MPC Power adjustment amount per prediction step satisfy:

[0081] in, For the first The feeder in the MPC Power adjustment amount per prediction step; This is the rolling step size for MPC; For the first One feeder in The upper bound of the adjustable capacity at any given time is calculated by aggregating the predicted irradiance margin factors of each substation under the feeder.

[0082] S33. Establish the mapping of the irradiation margin factor to the sub-second level station equilibrium layer.

[0083] Within each station area, sub-second consistency control uses the real-time value of the station-level irradiance margin factor as the local tracking saturation boundary. feeder line number The first in the district Power regulation of the inverter Satisfying saturation constraints:

[0084] in, For the first Taiwan inverter The amount of power adjustment at any given time (downward reduction is positive); For this inverter in The real-time upper limit of the adjustable capacity at any given time. This constraint ensures that the adjustment amount of the consistent control law output does not exceed the actual adjustable range of each inverter.

[0085] S4. Three-level time-scale coordination AGC architecture execution.

[0086] At the minute-level, the main grid AGC signal is received and weighted according to the irradiance margin factor to the county level; at the second-level, the county-level target is decomposed to the feeder level through model predictive control rolling optimization; at the sub-second-level, the feeder target is allocated to the inverter through consistency control. Specifically, the following steps are included: S41. Minute-level mainnet AGC instruction decomposition.

[0087] Assume the mainnet AGC is in The area control deviation signal issued at any time is It needs to be broken down into power adjustment targets for each county. The minute-level stratification employs a bounded decomposition method weighted by the irradiance margin factor.

[0088] in, To allocate to the county The power adjustment target; Weight the irradiance margin factor defined in step 31. This is the regional control deviation signal issued by the main network AGC.

[0089] Simultaneously satisfying bounded constraints:

[0090] Among them, the equality constraint guarantees the total power adjustment. Regional control deviation signal issued by the main network Balance; inequality constraints ensure that the allocation to each county does not exceed the upper bound of its adjustable capacity prediction value. .

[0091] when season The shortfall will be reported to the main network dispatcher.

[0092] The output interface variable at the minute-level is the power adjustment target for each county. The update cycle is (minutes).

[0093] S42. Second-level county feeder MPC rolling rescheduling.

[0094] Within each county, power adjustment targets are received at the minute-level. This needs to be decomposed into power adjustment amounts for each feeder within a second-level timescale. With the goal of minimizing frequency deviation and using the irradiance margin factor as a rolling constraint, an MPC optimization problem is established. Details are as follows: Define the prediction time domain length of MPC (Model Predictive Control) as: Step, control the time domain length as Step, rolling step length is The objective function of MPC is:

[0095] in, For the county In the The power adjustment reference value for each prediction step is expanded by minute-level instructions using linear interpolation or step hold. For all within the county M The feeder in the future The sum of power adjustments at each moment; For the first The increment of the power adjustment of the feeder; For tracking deviation weighting matrix; For the first The feeder in the future The power adjustment increment for each control step; To control the incremental weighting matrix; This represents the weighted quadratic norm.

[0096] The constraints include:

[0097]

[0098] Among them, the first set of constraints is the irradiance margin factor mapping constraint established in step 32, which ensures that the power adjustment of each feeder does not exceed the upper bound prediction value of the adjustable capacity. For the first Maximum power change rate constraint for a single feeder within a single MPC step.

[0099] MPC per Perform a single rolling solution, executing only the first step of the control input, and output the power adjustment for each feeder. As an interface variable at the second level, the feeder power adjustment amount is allocated as a reference value for power adjustment in each transformer area according to the proportion of the upper limit of the adjustable capacity of each area.

[0100] in, To be assigned to the feeder line number Reference values ​​for power adjustment in each distribution area; For the first All under the feeder line Nm The sum of the upper limits of the adjustable capacity of each transformer area; The robust power allocation coefficient for this substation area (obtained from step 53) is used to correct for the impact of cloud cover spatial correlation on allocation. When robust correction is not considered, .

[0101] Reference values ​​for power adjustment in each transformer area The data is sent to the corresponding edge controller of the distribution area.

[0102] S43. Sub-second level real-time balance of distribution area consistency.

[0103] A consistency control law with communication delay compensation is adopted to distribute the power adjustment reference value to each inverter, as follows: Within each transformer area, the feeder power adjustment amount is received at the second-level. The portion allocated to this station area This adjustment needs to be distributed to each inverter within a sub-second timescale through a consensus protocol.

[0104] The inverter communication topology within the distribution area is defined as shown in Figure 1. ,in, For the first feeder line number The collection of all inverters within a single distribution area; This is the communication edge set between inverters.

[0105] Definition of the first Consistency state variables of Taiwan inverter Adjust its normalized power ratio:

[0106] in, For the first The consistency state variable of the inverter has a value range of [0,1], which represents the proportion of the inverter's used adjustable capacity to its upper limit of adjustable capacity.

[0107] The consistency control law with communication delay compensation is designed as follows:

[0108] in, For the first Time derivative of the consistency state variable of the inverter; For the first The neighbor set of the inverter in the communication topology; The adjacency weights of the communication topology; For the neighbors Delay The state variables after that, For inverter and Communication delay between them; This is the uniformity gain coefficient; For reference tracking gain coefficient; For the reference adjustment ratio at the district level, by Calculated.

[0109] The convergence condition of a consistent control law is: when hour, For all For this to be valid, a communication topology diagram is required. Connected and Satisfy the time delay stability condition:

[0110] in, The maximum node degree in the communication topology; This represents the maximum communication delay.

[0111] Consistent control laws under saturation constraints The following steps are performed, corresponding to the inverter-level irradiance margin factor mapping constraints established in step 33. The actual power adjustment of the inverter is from Restore. Then calculate the output command of the inverter:

[0112] in, To be issued to the first The active power output command of the inverter (the power adjustment amount is reduced downward, so the actual output command is the current output minus the adjustment amount).

[0113] The update cycle of the sub-second layer is (Sub-second level), each inverter according to The inverter periodically updates the consistency control law and executes the output command. The actual output of the inverter... The data is fed back to the area edge controller, which then updates the real-time value of the irradiance margin factor accordingly. The data is then reported to the feeder controller. The feeder controller aggregates the data reported by each transformer area and calculates the actual power adjustment of the feeder. The report is then submitted to the county-level dispatch terminal.

[0114] in, For the first Taiwan inverter The actual contribution at any given moment; For the first One feeder in The actual power adjustment at any given time.

[0115] S5. Cloud-shading spatial correlation modeling and robust power allocation.

[0116] A spatial correlation coefficient matrix is ​​established based on the geographical distance of the transformer area and the cloud movement speed. An ellipsoidal uncertainty set is constructed, and the robust power sharing coefficient that satisfies the frequency deviation constraint is solved. Figure 4 As shown, the specific steps include: S51. Establish a spatial correlation coefficient matrix of photovoltaic power output in adjacent transformer substations.

[0117] Set up a county with a total Each transformer station participates in AGC adjustment. Taiwan The geographical distance between them is The cloud cluster is moving at a speed of The direction of cloud movement and the area Pointing to the station area The angle between the directions is Define the station area. Taiwan Spatial correlation coefficient of photovoltaic output between for:

[0118] in, Taiwan District Taiwan exist The spatial correlation coefficient of photovoltaic output at time t, with a value range of [0,1]; Taiwan District Taiwan Geographical distance, in km; for The speed of cloud movement at any given time, expressed in km / h; The direction of cloud movement and the area Pointing to the station area The angle between directions; For spatial attenuation scale parameters; The time delay attenuation scale parameter is defined by the first exponential term, which characterizes the attenuation of force correlation with spatial distance, and the second exponential term, which characterizes the time delay propagation characteristics of the cloud cover effect along the direction of cloud movement.

[0119] Constructing the spatial correlation coefficient matrix , its first Element is diagonal elements :

[0120] in, for The spatial correlation coefficient matrix at time points, The total number of stations participating in AGC regulation.

[0121] S52. Construct an ellipsoidal set of the most unfavorable spatial correlation fluctuation uncertainty.

[0122] Define the photovoltaic power output fluctuation vector for each distribution area. ,in, Taiwan District exist The output fluctuation at any given time (the deviation from the predicted value). Based on the spatial correlation coefficient matrix. and the standard deviation of power output fluctuation in each distribution area Construct the fluctuation covariance matrix :

[0123] in, for The output fluctuation covariance matrix at any given time; This is a diagonal matrix composed of the standard deviations of the power output fluctuations of each transformer area. Taiwan District exist The standard deviation of power output fluctuation at any given time can be obtained based on historical irradiance data and photovoltaic capacity estimation of the distribution area.

[0124] Define ellipsoidal uncertainty set for:

[0125] in, for An ellipsoidal set of uncertainties at time points; The power output fluctuation vector for each transformer substation; For uncertain budget parameters, control the degree of robust conservatism. The larger the value, the wider the range of uncertainties it covers.

[0126] S53. Establish a robust power sharing coefficient optimization model.

[0127] Using frequency deviation constraints as hard constraints, we solve for the robust power sharing coefficient that ensures the frequency deviation meets the limit requirements even under the most unfavorable space correlation fluctuation scenario. .

[0128] Robust power sharing coefficient is a minimax robust optimization problem:

[0129] The constraints are:

[0130] in, This represents the robust power allocation coefficient vector for the entire county. Taiwan District The apportionment coefficient; Taiwan District The power adjustment reference value; Taiwan District The output fluctuation; This is a regional control deviation signal; Taiwan District The adjustable capacity upper bound is defined. The first set of constraints ensures that the allocation coefficient is normalized; the second set of constraints ensures that the allocation amount of each transformer area does not exceed the adjustable capacity upper bound under the most unfavorable fluctuation scenario.

[0131] By utilizing the duality property of the uncertain set of the ellipsoid, the above minimax robust optimization problem is transformed into a deterministic second-order cone programming problem:

[0132] in, As an auxiliary variable, it represents the maximum frequency deviation proxy amount under the most unfavorable scenario; This is the vector of robust power amortization coefficients; Covariance matrix Cholesky decomposition factor; A vector composed of power adjustment reference values ​​for each transformer area; Taiwan District The apportionment coefficient; For the first One standard basis vector; Let be the Euclidean norm. This second-order cone programming problem can be solved efficiently using a commercial solver.

[0133] The solution obtained This is embedded in the process of distributing the feeder MPC output to each distribution area in step 42. Among these distribution areas... apportionment coefficient under county-wide numbering With the feeder-transformer area dual subscript in step 42 The correspondence is as follows: when the substation area Corresponding to the feeder line number When the platform area is in a certain area, This robust allocation coefficient ensures that the distribution of feeder power adjustment to each transformer area still meets the frequency deviation constraint even under the most unfavorable spatial correlation fluctuation scenario.

[0134] S6. Abnormal Degradation Strategy and Federal Assisted Calibration.

[0135] A tiered degradation strategy is designed for abnormal scenarios, and the irradiance margin factor prediction model is periodically calibrated using a federated learning framework. Specifically, the following steps are included: S61. Design a graded degradation strategy for three types of abnormal scenarios.

[0136] The three types of abnormal scenarios include communication interruption, sudden changes in irradiance, and inverter failure to operate. The tiered degradation strategies for each abnormal scenario specifically include: ① County-level communication outage downgrade.

[0137] When the communication link between the county-level dispatch center and a feeder controller is interrupted for more than a threshold time. At this time, the feeder will be downgraded to local MPC autonomous mode. The downgrade conditions are:

[0138] in, The current moment; This is the moment when the county-level instruction was successfully received for the last time. A threshold time is set for determining communication interruption. After degradation, the feeder adjusts its target based on the last received power. As a reference value for the local MPC, rolling scheduling continues according to the local irradiance margin factor constraint until communication is restored.

[0139] ②Irradiation mutation downgrade.

[0140] When the rate of change of irradiation in step 22 When this occurs, a fast reset of the irradiance margin factor is triggered. The fast reset performs the following operations: The predicted irradiance margin factor is... Replace with real-time measurement values Following step 23, recalculate the upper bound of the adjustable capacity at each level and send a margin update signal to the upper layer, triggering the second-level MPC layer to resolve the updated constraints in the next rolling step.

[0141] ③ Inverter fails to operate and is downgraded.

[0142] When the zone edge controller detects the first If the actual output deviation of the inverter exceeds the allowable range, the inverter is deemed to have failed to operate.

[0143] in, For the first The actual output of the inverter; This refers to the issued output command; The relative threshold for determining failure to operate. The failure to operate inverter is determined from the consistency topology diagram. Remove from the list and adjust its capacity. Set to zero, and the remaining inverters redistribute the power adjustment tasks of the transformer substations according to the updated consistency topology.

[0144] S62. Design a periodic calibration mechanism for the irradiance margin factor prediction model based on federated learning.

[0145] Each substation maintains a local irradiance margin factor prediction model. ,in, These are the model parameters. The calibration period is... (Slow timescale, usually in hours).

[0146] During each calibration cycle, each transformer substation uses local historical data to calculate the gradient. ,Will Uploaded to the county-level dispatch center.

[0147] in, For the first feeder line number Gradient parameters of the local model in each transformer area; For local loss functions; These are the parameters for the local model.

[0148] County-level dispatch centers execute federated aggregation:

[0149] in, These are global model parameters; The learning rate; For the first feeder line number The number of inverters in each distribution zone is used as the aggregation weight; This refers to the total number of inverters in all participating areas within the county.

[0150] The aggregated global model parameters are distributed to each station area, and each station area uses them. Update the local model.

[0151] The calibrated prediction model is used in step 22 to predict the irradiance margin factor. The calculation.

[0152] S7. Execution of the overall closed-loop scheduling process.

[0153] A complete closed-loop scheduling process is formed, from the main grid area control deviation signal to the power output adjustment of the transformer area inverter. The specific process is as follows: S71. Complete closed-loop signal flow description.

[0154] The overall scheduling process forms the following closed loop: (1) The main network AGC main station is based on a periodic Send regional control deviation signal To each county.

[0155] (2) County-level dispatch terminal receiving The power adjustment targets for each county are calculated based on the weighted bounded decomposition method of the irradiance margin factor in step 41. This is converted into a reference value for power adjustment of each feeder. The information is then distributed to each feeder controller.

[0156] (3) Each feeder controller operates in a periodic manner. Perform MPC rolling optimization in step 42 and output the power adjustment amount for each feeder. The distribution formula at the end of step 42 is combined with the robust allocation coefficient in step 53. Calculate the power adjustment reference values ​​for each transformer area. The data is then sent to the edge controller of the distribution area.

[0157] (4) Each area edge controller periodically Execute the consistency control law in step 43, and Distribute to each inverter and output inverter output commands. .

[0158] (5) Each inverter performs limiting control, and the actual output is... Feedback is sent to the area edge controller. The area edge controller updates the real-time value of the irradiance margin factor. The data is then reported to the feeder controller, which aggregates the data and reports it to the county-level dispatch terminal, completing the feedback loop.

[0159] S72. Definition of interface variables between three-level architectures.

[0160] The interface variables, data directions, and update cycles between each layer are summarized below: Minute-level to second-level distribution: County-level power adjustment targets County-level adjustable capacity upper limit prediction value The update cycle is .

[0161] Reporting from the second-level layer to the minute-level layer: Actual power adjustment of each feeder Real-time upper limit of adjustable capacity for each feeder The update cycle is .

[0162] Power adjustment reference values ​​for each distribution area are issued from the second-level layer to the sub-second-level layer. Adjustable capacity upper limit of each transformer area The update cycle is .

[0163] Sub-second level reports to second level: actual aggregated power output of each distribution area Aggregated real-time values ​​of irradiance margin factors for each distribution area The update cycle is .

Claims

1. A hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor, characterized in that: Includes the following steps: S1. County-level power distribution network topology modeling and photovoltaic cluster division: Establish the four-level object mapping relationship of county-level power distribution network, namely county level, feeder level, transformer area level and inverter level, and construct the photovoltaic aggregated output model of each level; S2. Definition and dynamic calculation of irradiance margin factor: Define the irradiance margin factor of a single inverter and its adjustable capacity upper limit, design a dynamic update mechanism including short-term irradiance prediction correction and rapid reset of irradiance mutation, and perform multi-level aggregation calculation to obtain the adjustable capacity upper limit of the substation area, feeder and county level. S3. Mapping of Irradiation Margin Factor to Level 3 AGC Constraints: The irradiation margin factor is mapped to the capacity constraint of the minute-level instruction decomposition layer, the upper and lower bound constraints of the power adjustment of the second-level model prediction control layer, and the tracking saturation boundary of the sub-second-level station balance layer, respectively. S4. Three-level time-scale coordinated AGC architecture execution: At the minute-level layer, the main network AGC signal is received and weighted and decomposed to the county level according to the irradiance margin factor; at the second-level layer, the county-level target is decomposed to the feeder level through model prediction control rolling optimization. At the sub-second level, the feeder target is distributed to the inverter through consistency control; S5. Cloud shading spatial correlation modeling and robust power allocation: Based on the geographical distance of the transformer area and the cloud movement speed, a spatial correlation coefficient matrix is ​​established, an ellipsoidal uncertainty set is constructed, and the robust power allocation coefficient that satisfies the frequency deviation constraint is solved. S6. Anomaly Degradation Strategy and Federated Auxiliary Calibration: A graded degradation strategy is designed for anomaly scenarios, and the irradiance margin factor prediction model is periodically calibrated using a federated learning framework.

2. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor as described in claim 1, characterized in that: In step S2, the irradiance margin factor of a single inverter Defined as: in, For the first feeder line number The first district Taiwan inverter The irradiance margin factor at time t has a value range of [0,1]. For the first feeder line number The first in the district Taiwanese photovoltaic inverters The maximum power output at any given moment; For the first feeder line number The first in the district Taiwanese photovoltaic inverters The actual contribution at any given moment; The rated active power capacity of the inverter; Define the upper limit of the adjustable capacity of a single inverter. for: 。 3. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor according to claim 2, characterized in that: The dynamic update mechanism in step S2 specifically includes: Real-time value update: Real-time value of irradiance margin factor Updated periodically according to the definition; Predicted values ​​updated: Predicted values ​​for the irradiance margin factor This is obtained by introducing a short-term irradiation prediction correction term, based on the predicted maximum power point output. The calculation is as follows: in, To predict the time step; Irradiation mutation detection and fast reset: Defining the rate of irradiation change for: in, for The rate of change of irradiance at any given time; For the first feeder line number The first district Location of Taiwan inverter The radiation intensity at a given time, expressed in W / m². ; The sampling interval; when When an irradiation mutation is detected, a fast reset is triggered: the current time is used directly. The real-time measured values ​​replace the predicted values ​​to update the irradiance margin factor, and a margin update signal is sent to the upper layer. The threshold for determining irradiation mutations; The multi-level aggregation calculation method in step 2 specifically includes: District-level aggregation: Upper limit of adjustable capacity at the district level The sum of the upper bounds of the adjustable capacity of each inverter, and the substation-level irradiance margin factor. Calculated using a capacity-weighted average; Feeder-level aggregation: Upper limit of adjustable capacity at the feeder level It is the sum of the upper limits of the adjustable capacity of each subordinate transformer station; County-level aggregation: Upper limit of adjustable capacity at the county level It is the sum of the upper limits of the adjustable capacity of each feeder under its jurisdiction; The predicted values ​​of the upper bounds of adjustable capacity at each level are aggregated using the same method, and the irradiance margin factor is predicted based on each inverter. Obtained through calculation.

4. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor as described in claim 1, characterized in that: In step S3, the three-level mapping specifically includes: Minute-level mapping: County-level power adjustment meets requirements ,in, To allocate to the county The amount of power adjustment; For the county exist The upper bound of the adjustable capacity is predicted based on the predicted irradiance margin factor. Define county Irradiation margin factor weighted decomposition weights for: In the formula, To allocate to the county The weighting coefficients are determined according to the proportion of the upper limit of the adjustable capacity forecast of each county to the total. This is the sum of the predicted upper bounds of the adjustable capacity for all counties involved in the adjustment process; Second-level mapping: Feeder power adjustment meets requirements ,in, For the first The feeder in the model predictive control Power adjustment amount per prediction step; The rolling step size for model predictive control; For the first One feeder in The upper bound of the adjustable capacity at any given time is calculated by aggregating the predicted irradiance margin factors of each substation under the feeder line. Sub-second mapping: Inverter power adjustment meets... ,in, For the first Taiwan inverter The amount of power adjustment at any given time; For this inverter in The real-time upper limit of the adjustable capacity at any given time.

5. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor according to claim 4, characterized in that: In step S4: The minute-level layer employs a bounded decomposition method weighted by the irradiation margin factor: ,in, To allocate to the county The power adjustment target; The weights are assigned to the irradiation margin factor defined in step 3. The regional control deviation signal issued by the main network AGC; The second-level layer aims to minimize the frequency deviation and uses the irradiance margin factor as a rolling constraint to establish a model predictive control optimization problem. The sub-second level layer employs a consistency control law with communication delay compensation to distribute power adjustment reference values ​​to each inverter.

6. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor according to claim 5, characterized in that: The bounded decomposition method of irradiation margin factor weighting in step 4 also satisfies the following constraints: Among them, the equality constraint guarantees the total power adjustment. Regional control deviation signal issued by the main network Balance; inequality constraints ensure that the allocation to each county does not exceed the upper bound of its adjustable capacity prediction value. ; when season And report the shortfall to the main network dispatcher; In the model predictive control optimization problem established at the second-level layer in step 4: Define the time domain length of the model predictive control prediction as: Step, control the time domain length as Step, rolling step length is The model predictive control objective function is: in, For the county In the The power adjustment reference value for each prediction step is expanded by minute-level instructions using linear interpolation or step hold. For all within the county M The feeder in the future The sum of power adjustments at each moment; For the first The increment of the power adjustment of the feeder; For tracking deviation weighting matrix; For the first The feeder in the future The power adjustment increment for each control step; To control the incremental weighting matrix; Denotes the weighted quadratic norm; The constraints include: Among them, the first set of constraints is the irradiance margin factor mapping constraint established in step 3, which ensures that the power adjustment of each feeder does not exceed the upper bound prediction value of the adjustable capacity. For the first The maximum power change rate constraint of a single feeder within a single model prediction control step; The sub-second-level real-time balancing of station consistency in step 4 adopts a consistency control law with communication delay compensation: in, For the first Time derivative of the consistency state variable of the inverter; For the first The neighbor set of the inverter in the communication topology; The adjacency weights of the communication topology; For the neighbors Delay The state variables after that, For inverter and Communication delay between them; This is the uniformity gain coefficient; For reference tracking gain coefficient; For the reference adjustment ratio at the district level, by Calculated; The convergence condition of the consistency control law is: when hour, This holds true for all inverters, provided that the communication topology is connected and Satisfy the time delay stability condition: in, The maximum node degree in the communication topology; This represents the maximum communication delay.

7. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor according to claim 1, characterized in that: In step S5, the ellipsoidal uncertainty set is defined as follows: in, for An ellipsoidal set of uncertainties at time points; The power output fluctuation vector for each transformer substation; for The output fluctuation covariance matrix at time t, This is a diagonal matrix composed of the standard deviations of the power output fluctuations of each transformer area. for The spatial correlation coefficient matrix at time points; For uncertain budget parameters; The total number of stations participating in AGC regulation; The robust power amortization coefficient is a minimax robust optimization problem: The constraints are: in, Taiwan District The apportionment coefficient; Taiwan District The power adjustment reference value; Taiwan District The output fluctuation; This is a regional control deviation signal; Taiwan District The upper limit of the adjustable capacity.

8. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor according to claim 7, characterized in that: The elements in the spatial correlation coefficient matrix are established using the following formula: in, Taiwan District Taiwan exist The spatial correlation coefficient of photovoltaic output at time t, with a value range of [0,1]; Taiwan District Taiwan Geographical distance, in km; for The speed of cloud movement at any given time, expressed in km / h; The direction of cloud movement and the area Pointing to the station area The angle between directions; For spatial attenuation scale parameters; This is the time delay attenuation scale parameter; The robust power sharing coefficient is solved by utilizing the duality of the ellipsoidal uncertainty set, transforming the minimax robust optimization problem into a deterministic second-order cone programming problem: The constraints include: in, As an auxiliary variable, it represents the maximum frequency deviation proxy amount under the most unfavorable scenario; This is the vector of robust power amortization coefficients; Covariance matrix Cholesky decomposition factor; A vector composed of power adjustment reference values ​​for each transformer area; Taiwan District The apportionment coefficient; For the first One standard basis vector; It is the Euclidean norm.

9. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor according to claim 1, characterized in that: In step S6, the abnormal scenarios include communication interruption, sudden irradiation changes, and inverter failure to operate. The graded degradation strategy for each abnormal scenario specifically includes: County-level communication interruption degradation: When the communication link between the county-level dispatch center and the feeder controller is interrupted for more than a threshold time, the feeder is degraded to the local model predictive control autonomous mode; Irradiation mutation downgrade: When the rate of irradiation change exceeds the threshold for irradiation mutation judgment, the irradiation margin factor is quickly reset. Inverter failure to operate and degradation: When the area edge controller detects that the actual output deviation of the inverter exceeds the allowable range, it determines that the inverter has failed to operate and removes it from the consistency topology; The federally assisted calibration mechanism in step S6 includes: Each substation maintains a local irradiance margin factor prediction model. Calculate the local gradient within each calibration period. ,in, For the first feeder line number Gradient parameters of the local model in each transformer area; For local loss functions; For local model parameters; Upload the gradient to the county-level scheduling center for federated aggregation: in, These are global model parameters; The learning rate; For the first feeder line number The number of inverters in each distribution zone is used as the aggregation weight; The total number of inverters in all participating districts within the county; The aggregated global model parameters are distributed to each station area to update the local model.

10. The hierarchical AGC optimization scheduling method for county-level distributed photovoltaic clusters based on irradiance margin factor according to claim 1, characterized in that: Step S6 is followed by step S7: Execution of the overall closed-loop scheduling process: forming a complete closed-loop scheduling process from the main grid area control deviation signal to the power output adjustment of the transformer area inverter, the specific process is as follows: The mainnet AGC main site operates on a cyclical basis. Send regional control deviation signal To each county; County-level dispatch terminal receiving And calculate the power adjustment targets for each county. This is then converted into power adjustment reference values ​​for each feeder and sent to each feeder controller; Each feeder controller operates in a periodic manner. The model predictive control rolling optimization is executed, the power adjustment amount of each feeder is output, and the reference value of power adjustment for each transformer area is calculated by combining the robust amortization coefficient and sent to the transformer area edge controller; Each area edge controller periodically The consistency control law is executed to distribute the reference value to each inverter and output the power command; Each inverter performs limiting control and feeds back the actual output to the distribution area edge controller. The distribution area edge controller updates the real-time value of the irradiance margin factor and reports it level by level, forming a complete closed-loop scheduling.

Images