A virtual power plant multi-time scale adaptive aggregation method based on grid overload

Abstract

The application discloses a kind of virtual power plant multi-time scale adaptive aggregation methods based on grid overload, it is related to power grid resource scheduling technical field, the gist is that: extracting the physical operator in the physical mechanism equation of various distributed resources, generating the grey-box aggregation model of various distributed resources;Two-stage stochastic programming model and distribution robust optimization model are constructed, and the benchmark scheduling plan is generated based on long stationary time scale scheduling scheme and short scale standby scheduling scheme;In the intra-day execution stage, when the aggregated power deviation or system frequency rate of change exceeds the preset safety threshold, then obtain the multi-dimensional measurement data of regional power grid within the preset period;The multi-dimensional measurement data is input into the deep learning prediction model to obtain the renewable energy output prediction sequence and regional load demand prediction sequence;A sparse regularization correction objective function is constructed in combination with the grey-box aggregation model;The sparse regularization correction objective function is solved to obtain the corrected benchmark scheduling plan.

Description

Technical Field

[0001] This invention relates to the field of power grid dispatching technology, specifically to a multi-timescale adaptive aggregation method for virtual power plants based on heavy power grid overload. Background Technology

[0002] With the construction of new power systems, the penetration rate of flexible resources such as distributed wind power, photovoltaics, electric vehicles (EVs), and energy storage systems is continuously increasing. At the same time, the rapid growth of electricity load leads to severe overload risks for the main and distribution networks during peak hours. Distributed power generation, as a technology that can aggregate multiple resources to participate in grid dispatch, is considered an effective way to solve the problem of heavy overload.

[0003] A method for controlling overload in a distribution network is disclosed in Chinese patent publication CN113872217A. This method compares the total electricity load and the total generation load with overload limits. If the total electricity load exceeds the overload limit and the distributed power source is operating under power limitation, the power generation limit of the distributed power source is lifted. If the total generation load exceeds the overload limit, the proportion of the charging facility load to the total electricity load is compared with both a lower limit and an upper limit for the proportion of the charging facility load. If the proportion of the charging facility load to the total electricity load is less than the lower limit, the real-time electricity price for the charging facility is reduced. This method utilizes the flexibility of electric vehicles and the controllability of distributed power sources to adjust electricity consumption and charging behavior in the distribution network.

[0004] However, the distributed multi-timescale adaptive aggregation scheme for grid overload provided by the above technical solutions still has the following drawbacks:

[0005] (1) The scheme is based on a rule-based or simplified physical model. Although the calculation is simple, it ignores the complex nonlinear dynamic characteristics of heterogeneous resources (such as battery aging and wind turbine aerodynamic characteristics), resulting in insufficient accuracy of the aggregation model in the wide frequency domain or under large disturbances.

[0006] (2) This scheme mainly relies on preset overload limits and upper and lower limits of proportion for judgment, lacks the ability to adapt to environmental changes (such as sudden weather changes or changes in user habits), and has poor robustness.

[0007] (3) This scheme often triggers control only after an overload occurs, resulting in a significant response lag; while fixed-cycle optimization calculations waste computing resources when the system is running smoothly.

[0008] Therefore, the present invention aims to provide a multi-timescale adaptive aggregation method for virtual power plants based on grid overload, in order to solve the aforementioned problems. Summary of the Invention

[0009] The technical problem this invention aims to solve is the lack of adaptive adjustment capability of existing technologies to environmental changes and the neglect of the complex nonlinear dynamic characteristics within heterogeneous resources. The goal is to provide a multi-timescale adaptive aggregation method for virtual power plants based on heavy grid overload. This invention employs robust-stochastic hybrid programming and parameter feedback evolution to effectively reduce the conservatism of the model and improve the economic efficiency of the virtual power plant while ensuring system safety, based on the adjustment requirements of different time scales. Through the coordinated optimization of long-stationary time scales and short-fluctuating time scales, both economic efficiency and safety are considered. Simultaneously, an event-triggered mechanism is adopted based on the real-time operating status and disturbance conditions of the system to achieve millisecond-level rapid response to sudden events. Through the dual drive of state and error, it ensures that the system only invests computing resources when necessary.

[0010] This invention is achieved through the following technical solution:

[0011] A multi-timescale adaptive aggregation method for virtual power plants based on grid overload, the method includes:

[0012] Physical operators are extracted from the physical mechanism equations of various distributed resources, and based on the historical operation data of various distributed resources, a gray box aggregation model of various distributed resources is generated under the constraints of the physical operators using a symbolic regression algorithm.

[0013] In the current phase, with the goal of minimizing expected operating costs, a two-stage stochastic programming model is constructed based on pre-built constraints. The two-stage stochastic programming model is then solved to obtain a long-term stable time-scale scheduling scheme. With the goal of minimizing reserve capacity when fluctuations are greatest in various scenarios, a sub-Brussels bar optimization model is constructed. Under the constraints of the long-term stable time-scale scheduling scheme, the sub-Brussels bar optimization model is solved to obtain a short-term reserve scheduling scheme. A baseline scheduling plan is generated based on the long-term stable time-scale scheduling scheme and the short-term reserve scheduling scheme.

[0014] During the intraday execution phase, the aggregated power deviation after the distributed resource execution benchmark scheduling plan is obtained, as well as the system frequency change rate at the common coupling point of the power grid; when the aggregated power deviation or the system frequency change rate exceeds the preset safety threshold, multi-dimensional measurement data of the regional power grid within the preset time period is obtained.

[0015] Multidimensional measurement data are input into a pre-trained deep learning prediction model to obtain renewable energy output prediction sequences and regional load demand prediction sequences. Based on the distributed resource output prediction sequences and regional load demand prediction sequences, a sparse regularization correction objective function is constructed by combining a gray box aggregation model. The sparse regularization correction objective function is solved to obtain the corrected baseline scheduling plan.

[0016] Furthermore, the expected minimum operating cost includes minimizing the day-ahead baseline scheduling cost and minimizing the adjustment penalty cost under various scenarios;

[0017] The pre-constructed constraints include power balance constraints, unit output and ramping constraints, energy storage charging and discharging constraints, and capacity constraints.

[0018] Furthermore, with the objective of minimizing standby capacity during periods of maximum fluctuation in various scenarios, a sub-Bruker optimization model is constructed, specifically as follows:

[0019] We construct a Wasserstein distance fuzzy set centered on empirical probability distribution using historical operational data, and based on the Wasserstein distance fuzzy set, we construct a sub-Bruker optimization model with the objective of minimizing the backup capacity when the fluctuations are greatest in various scenarios.

[0020] Furthermore, the method also includes:

[0021] Obtain the intraday operational deviation dataset from the previous scheduling cycle, and extract historical deviation distribution characteristic indicators based on the intraday operational deviation dataset to serve as deviation conditional risk value.

[0022] Based on the deviation condition risk value, the radius of the Wasserstein distance fuzzy set in the next scheduling cycle of the split-Bruker optimization model is adaptively adjusted and optimized.

[0023] Furthermore, based on the deviation conditional risk value, the radius of the Wasserstein distance fuzzy set of the sub-Bruker optimization model is adaptively adjusted and optimized in the next scheduling cycle, specifically as follows: In the formula, Indicates the first The Wasserstein distance radius of the fuzzy set during the scheduling period. Indicates the first The Wasserstein distance radius of the fuzzy set during the scheduling period. Indicates a smooth learning rate. This represents the adjustment coefficient. This represents the value at risk of deviation conditions.

[0024] Furthermore, based on the distributed resource output prediction sequence and the regional load demand prediction sequence, a sparse regularization correction objective function is constructed using a gray-box aggregation model, specifically as follows: In the formula, This represents the total system-level power regulation required to eliminate the current heavy overload or frequency over-limit. Indicates the first A gray-box aggregation model for distributed resources. This represents the baseline scheduling plan before the revision. This indicates the modification of the scheduling plan vector. Represents the sparsity penalty coefficient. express Norm.

[0025] This invention also provides a virtual power plant multi-timescale adaptive aggregation system based on grid heavy overload, which is used in the virtual power plant multi-timescale adaptive aggregation method based on grid heavy overload described in any one of the above claims, the system comprising:

[0026] The aggregation model generation module is used to extract physical operators from the physical mechanism equations of various distributed resources, and based on the historical operation data of various distributed resources, generate gray box aggregation models of various distributed resources under the constraints of the physical operators using a symbolic regression algorithm.

[0027] The scheduling plan generation module is used to construct a two-stage stochastic programming model in the day-ahead phase, with the goal of minimizing expected operating costs and combining pre-built constraints, and solve the two-stage stochastic programming model to obtain a long-term stable time-scale scheduling scheme; with the goal of minimizing reserve capacity when various scenarios have the greatest fluctuations, a sub-Bruker optimization model is constructed, and the sub-Bruker optimization model is solved under the constraints of the long-term stable time-scale scheduling scheme to obtain a short-term reserve scheduling scheme; a baseline scheduling plan is generated based on the long-term stable time-scale scheduling scheme and the short-term reserve scheduling scheme.

[0028] The execution deviation generation module is used to obtain the aggregated power deviation after the distributed resource execution benchmark scheduling plan and the system frequency change rate of the grid common coupling point during the intraday execution phase; when the aggregated power deviation or the system frequency change rate exceeds the preset safety threshold, it obtains multi-dimensional measurement data of the regional power grid within a preset time period.

[0029] The scheduling plan correction module is used to input multidimensional measurement data into a pre-trained deep learning prediction model to obtain renewable energy output prediction sequences and regional load demand prediction sequences. Based on the distributed resource output prediction sequences and regional load demand prediction sequences, a sparse regularization correction objective function is constructed by combining a gray box aggregation model. The sparse regularization correction objective function is solved to obtain the corrected baseline scheduling plan.

[0030] The present invention also provides a computer device, including a system memory and a processor, wherein the system memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0031] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0032] The present invention also provides a computer program product containing instructions that, when executed by a cluster of computer devices, cause the cluster of computer devices to perform the method described in any of the preceding claims.

[0033] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0034] In this invention, based on the adjustment requirements at different time scales, a robust-stochastic hybrid programming and parameter feedback evolution are adopted to effectively reduce the conservatism of the model and improve the economic efficiency of the virtual power plant while ensuring system safety. Through the coordinated optimization of long-stable time scales and short-fluctuating time scales, both economy and safety are taken into account. At the same time, based on the real-time operating status and disturbance conditions of the system, an event triggering mechanism is adopted to achieve millisecond-level rapid response to sudden events. Through the dual drive of state and error, it is ensured that the system only invests computing resources when necessary. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0036] Figure 1 This is a flowchart illustrating a multi-timescale adaptive aggregation method for virtual power plants based on grid overload in this embodiment.

[0037] Figure 2 This is a schematic diagram of the module connection of a virtual power plant multi-timescale adaptive aggregation system based on grid overload in this embodiment;

[0038] Figure 3 This is a schematic diagram of the structure of a computer device in this embodiment. Detailed Implementation

[0039] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0040] In this disclosure, unless otherwise stated, the use of terms such as "first," "second," etc., to describe various elements is not intended to limit the positional, temporal, or importance relationships of these elements; such terms are merely used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of that element, while in other cases, based on the context, they may refer to different instances.

[0041] The terminology used in the description of the various examples in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context explicitly indicates otherwise, an element may be one or more unless the number of elements is specifically limited. Furthermore, the term "and / or" as used in this disclosure covers any one of the listed items and all possible combinations thereof.

[0042] Example 1

[0043] See Figure 1 , Figure 1 A flowchart illustrating a multi-timescale adaptive aggregation method for virtual power plants based on grid overload is shown, wherein the method includes:

[0044] S1: Extract the physical operators from the physical mechanism equations of various distributed resources, and based on the historical operation data of various distributed resources, use the symbolic regression algorithm to generate gray box aggregation models of various distributed resources under the constraints of the physical operators;

[0045] It should be noted that in this embodiment, the distributed resources include wind turbine systems, energy storage systems, and electric vehicles. In other embodiments, other distributed resources may also be included, and no further restrictions are imposed here. Meanwhile, historical operating data includes equipment output data, issued control command data, equipment's own physical status data, external environment and grid boundary data, etc. In other embodiments, other operating data may also be set, and no further restrictions are imposed here.

[0046] Specifically, in this embodiment, step S1 aims to construct a heterogeneous resource physical guidance gray box model library. The specific process is as follows: First, based on the physical mechanism equations of various distributed resources, such as second-order cone relaxation models and electrochemical models, the core physical operators, such as integrals, differentials, and nonlinear saturation elements, are extracted from the physical mechanism equations of various distributed resources. Finally, using the symbolic regression algorithm, under the constraints of the physical operators, the gray box aggregation models corresponding to various distributed resources are automatically searched and generated from the historical operation data of various distributed resources, thereby achieving a high-fidelity reduced-order representation of the physical characteristics of heterogeneous resources and forming a heterogeneous resource gray box model library.

[0047] For example, in this embodiment, taking an energy storage system as an example, the gray box aggregation model constructed is as follows: In the formula Indicates the first Aggregate output power of the instantaneous energy storage system, kW; Indicates the first The charging and discharging control commands (continuous variables or discrete states between 0 and 1) are constantly sent to the energy storage system. Indicates the first The state of charge of the energy storage system at any time reflects the current remaining capacity of the battery; Indicates the grid frequency deviation, integral term This represents the cumulative response of the system to frequency fluctuations (simulated integral operator). Indicates the first Interactive power at the common coupling point of the power grid at any given time; This represents the hyperbolic tangent function, which signifies the nonlinear saturation stage and is used to simulate the power saturation characteristics of energy storage devices when approaching rated power or charge / discharge limits. These represent the parameters of the gray box model, which are undetermined coefficients automatically identified by the symbolic regression algorithm based on historical operating data.

[0048] S2: In the day-ahead phase, with the goal of minimizing expected operating costs, a two-stage stochastic programming model is constructed based on pre-built constraints. The two-stage stochastic programming model is then solved to obtain a long-term stable time-scale scheduling scheme. With the goal of minimizing reserve capacity when fluctuations are greatest in various scenarios, a sub-Bruker optimization model is constructed. The sub-Bruker optimization model is then solved under the constraints of the long-term stable time-scale scheduling scheme to obtain a short-term reserve scheduling scheme. A baseline scheduling plan is generated based on the long-term stable time-scale scheduling scheme and the short-term reserve scheduling scheme.

[0049] Specifically, in this embodiment, step S2 aims to establish a two-stage stochastic programming model for hourly energy balance needs during the day-ahead phase, and generate a long-term stable timescale scheduling scheme; for minute-level adjustment needs, a distributed bar optimization (DRO) model is established, and a short-term backup scheduling scheme that complements the long-term stable timescale scheduling scheme is generated, as follows:

[0050] In the day-ahead phase, unit start-up / shutdown and base generation are used as decision variables for the first phase, with the goal of minimizing the day-ahead baseline dispatch cost. Multiple probabilistic scenarios involving renewable energy output (wind and solar) and load uncertainty are used as decision variables for the second phase, with the goal of minimizing the regulation penalty cost under each scenario. The first and second phases are linked by coupling constraints: the actual output (the sum of baseline output and reserve regulation) in any scenario during the second phase must be constrained by the unit start-up / shutdown status determined in the first phase and must not exceed the upper and lower limits of the equipment's physical output. Then, combining power balance constraints, unit output and ramp-up constraints, and energy storage charging / discharging and capacity constraints, a two-stage stochastic programming model is constructed. The objective function of this two-stage stochastic programming model is defined as: In the formula, The first phase is the day-ahead baseline dispatch cost, which includes unit start-up and shutdown, and basic generation costs. This indicates the second phase in the scene. The adjustment and penalty costs include the costs of wind and solar curtailment, load shedding, and reserve call-up. For the scene The probability of occurrence; the two-stage stochastic programming model is transformed into a standard mixed integer linear programming (MILP) model for solution, and a long-term stationary time-scale scheduling scheme is formulated by solving the MILP model;

[0051] Simultaneously, a Wasserstein distance fuzzy set centered on an empirical probability distribution is constructed using historical operational data. Based on this Wasserstein distance fuzzy set, a sub-Bruker optimization model is constructed with the objective of minimizing reserve capacity when fluctuations are greatest in various scenarios. The objective function of the sub-Bruker optimization model is defined as follows: In the formula, Represents the decision variable for reserve capacity. This indicates that the cost function generated by the backup is invoked. The random variable representing short-term high-frequency fluctuations is used. Finally, the above infinite-dimensional minimax problem is transformed into a deterministic convex optimization model using strong duality theory. The model is then solved under the hard boundary constraints of unit start-up and shutdown and basic output provided by the long-stationary time-scale scheduling scheme to obtain the short-scale standby scheduling scheme. The solved segment-scale standby scheduling scheme is then added to the long-stationary time-scale scheduling scheme to obtain the baseline scheduling plan.

[0052] It should be noted that, in this embodiment, the Wasserstein distance fuzzy set is a probability distribution space jointly constructed by the empirical probability distribution (the "center" based on historical data statistics), the fuzzy set radius R (which determines the "size" of the level of protection and conservatism) as the boundary, the Wasserstein distance (the "scale" that measures the difference in distribution), and the support set (the "absolute boundary" of the real physical limit). It enables the virtual power plant to prevent the worst possible wind and solar disturbances within a reasonable range and at the most economical cost.

[0053] Specifically, in this embodiment, a two-stage stochastic programming model is used to handle the long-term power balance problem. By pre-smoothing the dual uncertainties of source and load, the economic optimization of the virtual power plant aggregation system in a statistical sense is ensured, and a margin for adjustment is reserved for subsequent intraday short-term high-frequency fluctuations. The split-bulb optimization model and the two-stage stochastic programming model complement each other on long and short scales to cope with the rapid random fluctuations of renewable energy on a minute-by-minute scale or the sudden changes in output caused by extreme weather. This ensures that the system can still have sufficient reserve capacity to support the system in extreme and severe scenarios with decreased prediction accuracy and violent fluctuations, thereby ensuring the absolute operational safety of the power grid.

[0054] Meanwhile, although a long-term stable timescale scheduling scheme at the hour level was formulated through a two-stage stochastic programming model to determine unit start-up and shutdown and base output, this model is based on probabilistic scenarios to handle the uncertainty of source loads. Its time granularity is insufficient to cover the rapid random fluctuations of renewable energy at the minute level and the sudden changes in output under extreme weather conditions. Therefore, on the baseline plan boundary determined by the two-stage stochastic programming model, a sub-Blu-bar optimization model is further established to address the minute-level high-frequency fluctuations. The worst distribution of short-term uncertainty is characterized by Wasserstein fuzzy sets, and the reserve capacity is determined. The two form a long-term and short-term complementary architecture of long-term power balance and short-term reserve security. The two-stage stochastic programming model provides hard boundary constraints for unit start-up and shutdown and base output for the sub-Blu-bar optimization model, while the sub-Blu-bar optimization model supplements the long-term stable timescale baseline plan of the two-stage stochastic programming model with a safety margin under minute-level extreme fluctuations.

[0055] S3: During the intraday execution phase, acquire the aggregated power deviation after the distributed resource execution baseline scheduling plan, as well as the system frequency change rate at the common coupling point of the power grid; when the aggregated power deviation or the system frequency change rate exceeds the preset safety threshold, acquire the multi-dimensional measurement data of the regional power grid within the preset time period;

[0056] It should be noted that in this embodiment, the multidimensional measurement data includes environmental meteorological data, historical power output sequence, node voltage and frequency characteristics. Other measurement data may also be included in other embodiments, and no further restrictions are imposed here.

[0057] Specifically, in this embodiment, step S3 aims to overcome the problems of wasted computing resources and response lag caused by traditional fixed-time-step rolling correction, and to construct an event-triggered detection mechanism based on state-error dual-drive, as follows:

[0058] During the intraday execution phase, the first step is to acquire all distributed resources at the current moment. The actual aggregated power collected is used to calculate the aggregated power deviation using the planned aggregated power in the baseline scheduling plan. Specifically: In the formula, Indicates the current moment The deviation in polymerization power, This indicates that the baseline scheduling plan is scheduled to be completed at the current time. The planned aggregate power, Indicates the current moment The system collects and aggregates the actual power from various distributed resources in real time, and then monitors the rate of change of the system frequency at the common coupling point of the power grid through a phase-locked loop (PLL). Finally, the deviation in polymerization power or the rate of change in system frequency are judged, and the specific judgment rules are as follows: In the formula, This indicates the preset power deviation tolerance safety threshold. Indicates the safety threshold for the system's rate of change of frequency; when When this occurs, it indicates that the system is on its normal operating trajectory and continues to maintain the day-ahead baseline scheduling plan without performing invalid calculations; when When this occurs, it indicates that a sudden disturbance or accumulated error exceeding the limit has occurred in the system, and the regional power grid is immediately acquired at the trigger time. The multidimensional measurement data within the previously preset time period triggers subsequent intraday short-term correction calculations.

[0059] S4: Input multidimensional measurement data into a pre-trained deep learning prediction model to obtain renewable energy output prediction sequences and regional load demand prediction sequences; based on the distributed resource output prediction sequences and regional load demand prediction sequences, construct a sparse regularization correction objective function in conjunction with a gray box aggregation model; solve the sparse regularization correction objective function to obtain the corrected baseline scheduling plan.

[0060] It should be noted that in this embodiment, the deep learning prediction model adopts a deep learning prediction model based on the spatiotemporal attention mechanism, such as the spatiotemporal graph convolutional network STGCN or the long short-term memory network LSTM. In other embodiments, other prediction models may also be used, which will not be elaborated on here.

[0061] Specifically, in this embodiment, step S4 aims to modify the baseline scheduling plan, as follows:

[0062] First, the collected multidimensional measurement data is input into a pre-trained deep learning prediction model based on a spatiotemporal attention mechanism to predict and output a renewable energy output forecast sequence and a regional load demand forecast sequence within the future prediction time domain (e.g., the next 15 minutes to 1 hour). Then, using the gray-box aggregation model of various distributed resources, and the high-precision prediction boundary composed of the renewable energy output forecast sequence and the regional load demand forecast sequence, a model with... The sparse regularization correction objective function for the norm penalty term is as follows: In the formula, This represents the total system-level power regulation required to eliminate the current heavy overload or frequency over-limit. Indicates the first A gray-box aggregation model for distributed resources. This represents the baseline scheduling plan before the revision. This indicates the modification of the scheduling plan vector. Represents the sparsity penalty coefficient. express Norm; finally, the objective function for sparse regularization correction is solved to obtain the corrected baseline scheduling plan.

[0063] In this embodiment, to achieve closed-loop feedback optimization between the day-ahead baseline plan and the actual intraday operation, the method further includes:

[0064] Obtain the intraday operational deviation dataset from the previous scheduling cycle, and extract historical deviation distribution characteristic indicators based on the intraday operational deviation dataset to serve as deviation conditional risk value.

[0065] Based on the deviation condition risk value, the radius of the Wasserstein distance fuzzy set in the next scheduling cycle of the split-Bruker optimization model is adaptively adjusted and optimized.

[0066] Specifically, in this embodiment, the intraday operational deviation dataset of the previous scheduling period is first obtained. ,in For a moment The absolute deviation between the actual output of renewable energy and the day-ahead forecast output in the baseline dispatch plan is analyzed. Historical deviation distribution characteristics are extracted through statistical analysis and used as the deviation conditional risk value. Then, based on the extracted deviation conditional risk value, an adaptive adjustment rule is constructed to dynamically update the fuzzy set radius in the sub-Bruker optimization model, specifically: In the formula, Indicates the first The Wasserstein distance radius of the fuzzy set during the scheduling period. Indicates the first The Wasserstein distance radius of the fuzzy set during the scheduling period. Indicates a smooth learning rate. This represents the adjustment coefficient. Value at risk (VAT) representing the conditional risk of deviation; when the historical intraday deviation risk... When the size is small, the rule-driven mechanism is adaptively adjusted. Fuzzy set radius in scheduling cycle Adaptive shrinkage.

[0067] It should be noted that in this embodiment, when the system assesses that the recent actual operating error (CVaR) is small and in a stable state, the adaptive adjustment rule will actively reduce the radius of the Wasserstein fuzzy set in the DRO model (i.e., lower the hypothetical protection level for extreme and severe scenarios). This significantly reduces the reserve capacity threshold required by the system. As a result, the equipment capacity that was forcibly withheld and idled in the day-ahead baseline plan for "just in case" can be unbound and "released". The virtual power plant can then directly invest these redundant capacities into daily power generation or market transactions. Thus, while maintaining the bottom line of safety, it perfectly resolves the capacity waste and economic losses caused by excessive conservatism in traditional robust optimization.

[0068] For example, to verify the feasibility of the method provided in this embodiment, an operating scenario and specific implementation process are provided as follows:

[0069] In a certain region, a virtual power plant aggregates 500 electric vehicles (EVs) and needs to participate in the day-ahead dispatch and intraday frequency response of the power grid. The data comes from the vehicle BMS system and charging pile measurement data, and there are problems such as user behavior uncertainty and battery parameter time variability.

[0070] First, historical data such as SOC, charging and discharging power, and user travel habits of the EV cluster are collected, and outliers are removed; then, an operator library containing physical mechanisms is set up. The aggregate power of the EV cluster is obtained by using a symbolic regression algorithm for evolution. With control signals The analytical relationships, such as those containing differential operators to reflect the dynamic charge and discharge rates of the battery, include... Operators are used to reflect charge / discharge saturation characteristics: in These are the undetermined model parameters generated by symbolic regression identification;

[0071] Then, for hourly energy balance requirements, a two-stage stochastic programming model (MILP) is constructed to formulate a baseline charge-discharge plan; for minute-level adjustment requirements, a distributed bar optimization (DRO) model is constructed to determine reserve capacity; simultaneously, based on the historical intraday operating deviations from the previous week (extracted... Indicators), using parameter feedback evolution formula Dynamically adjust the radius of uncertainty set of the DRO model This successfully reduced the redundancy of standby capacity by 15%, improving the economic efficiency of EV clusters participating in scheduling.

[0072] Finally, the measurement data is collected in real time to calculate the polymerization power deviation. And monitor the rate of change of grid frequency through a phase-locked loop (PLL). Set power threshold and frequency change rate safety threshold If at a certain moment a large number of EVs suddenly leave the grid, causing the aggregate power deviation to exceed 5%, the discrimination function... An output of 1 immediately triggers the event mechanism. The system then wakes up the STGCN / LSTM-based ultra-short-term forecasting module to reconstruct the EV available adjustment margin for the next 15 minutes; subsequently, based on... A sparse relaxation regularization algorithm for norm penalty terms can solve the modified objective function within 200ms. New correction commands are issued only to some key EV nodes, enabling a millisecond-level rapid response to sudden disturbances.

[0073] Example 2

[0074] See Figure 2 The present invention also provides a virtual power plant multi-timescale adaptive aggregation system based on grid heavy overload, which is used in the virtual power plant multi-timescale adaptive aggregation method based on grid heavy overload described in any one of the above claims, the system comprising:

[0075] The aggregation model generation module 100 is used to extract physical operators from the physical mechanism equations of various distributed resources, and based on the historical operation data of various distributed resources, generate gray box aggregation models of various distributed resources under the constraints of the physical operators using a symbolic regression algorithm.

[0076] The scheduling plan generation module 200 is used to construct a two-stage stochastic programming model in the day-ahead phase, with the goal of minimizing expected operating costs and combining pre-built constraints, and solve the two-stage stochastic programming model to obtain a long-term stable time-scale scheduling scheme; with the goal of minimizing reserve capacity when various scenarios have the greatest fluctuations, a sub-Brussels bar optimization model is constructed, and the sub-Brussels bar optimization model is solved under the constraints of the long-term stable time-scale scheduling scheme to obtain a short-term reserve scheduling scheme; a baseline scheduling plan is generated based on the long-term stable time-scale scheduling scheme and the short-term reserve scheduling scheme.

[0077] The execution deviation generation module 300 is used to obtain the aggregated power deviation after the distributed resource execution benchmark scheduling plan and the system frequency change rate of the grid common coupling point during the intraday execution phase; when the aggregated power deviation or the system frequency change rate exceeds the preset safety threshold, it obtains the multi-dimensional measurement data of the regional power grid within the preset time period.

[0078] The scheduling plan correction module 400 is used to input multidimensional measurement data into a pre-trained deep learning prediction model to obtain a renewable energy output prediction sequence and a regional load demand prediction sequence; based on the distributed resource output prediction sequence and the regional load demand prediction sequence, a sparse regularization correction objective function is constructed by combining a gray box aggregation model; the sparse regularization correction objective function is solved to obtain the corrected baseline scheduling plan.

[0079] It should be noted that the modules in the system of Embodiment 2 correspond to the steps in the method of Embodiment 1. The steps in the method of Embodiment 1 have been described in detail in Embodiment 1, and the module content in the system will not be described in detail in this Embodiment 2.

[0080] Example 3

[0081] This embodiment also provides a computer device, including a system memory 1005 and a processor 1001. The system memory 1005 stores a computer program, and the processor 1001 executes the computer program to implement the steps of any of the methods described above.

[0082] It should be noted that the processor 1001 is used to execute the steps in the above method embodiments according to the instructions in the program code. Alternatively, when the processor 1001 executes the computer program, it implements the functions of each module / unit in the above system / device embodiments.

[0083] Specifically, in this embodiment, the computer program can be divided into one or more modules / units, which are stored in the system memory 1005 and executed by the processor 1001 to complete this application. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.

[0084] The terminal device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor 1001 and a system memory 1005. Those skilled in the art will understand that this does not constitute a limitation on the terminal device; it may include more or fewer components than shown in the figures, or a combination of certain components, or different components. For example, the terminal device may also include an input / output device 1003, a network access device 1002, a bus 1006, etc.

[0085] The processor 1001 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0086] System memory 1005 can be an internal storage unit of the terminal device, such as a hard drive or RAM. System memory 1005 can also be a storage device 1004 of the terminal device, such as an external hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, or FlashCard. Furthermore, system memory 1005 can include both internal storage units and storage device 1004. System memory 1005 is used to store computer programs and other programs and data required by the terminal device. System memory 1005 can also be used to temporarily store data that has been output or will be output.

[0087] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0088] Example 4

[0089] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0090] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), registers, hard disks, optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof, or any other form of computer-readable storage medium in the art.

[0091] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside within an application-specific integrated circuit (ASIC). In embodiments of the invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device.

[0092] Example 5

[0093] This embodiment also provides a computer program product containing instructions that, when executed by a cluster of computer devices, cause the cluster of computer devices to perform the method described in Embodiment 1.

[0094] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for adaptive aggregation of virtual power plants across multiple time scales based on heavy grid overload, characterized in that the method... include: Physical operators are extracted from the physical mechanism equations of various distributed resources, and based on the historical operation data of various distributed resources, a gray box aggregation model of various distributed resources is generated under the constraints of the physical operators using a symbolic regression algorithm. In the current phase, with the goal of minimizing expected operating costs, a two-stage stochastic programming model is constructed based on pre-built constraints. The two-stage stochastic programming model is then solved to obtain a long-term stable time-scale scheduling scheme. With the goal of minimizing reserve capacity when fluctuations are greatest in various scenarios, a sub-Brussels bar optimization model is constructed. The sub-Brussels bar optimization model is then solved under the constraints of the long-term stable time-scale scheduling scheme to obtain a short-term reserve scheduling scheme. A baseline scheduling plan is generated based on long-stable time-scale scheduling schemes and short-scale backup scheduling schemes. During the intraday execution phase, the aggregated power deviation after the distributed resource execution baseline scheduling plan is obtained, as well as the system frequency change rate at the grid common coupling point; When the aggregated power deviation or the system frequency change rate exceeds the preset safety threshold, multi-dimensional measurement data of the regional power grid within the preset time period are acquired. Multidimensional measurement data are input into a pre-trained deep learning prediction model to obtain renewable energy output prediction sequences and regional load demand prediction sequences; based on the distributed resource output prediction sequences and regional load demand prediction sequences, a sparse regularization correction objective function is constructed by combining a gray box aggregation model. Solve the sparse regularization correction objective function to obtain the corrected baseline scheduling plan.

2. The method for adaptive aggregation of virtual power plants across multiple time scales based on grid overload as described in claim 1, characterized in that, The minimum expected operating cost includes minimizing the day-ahead baseline scheduling cost and minimizing the adjustment penalty cost under various scenarios; The pre-constructed constraints include power balance constraints, unit output and ramping constraints, energy storage charging and discharging constraints, and capacity constraints.

3. The method for adaptive aggregation of virtual power plants across multiple time scales based on grid overload as described in claim 1, characterized in that, With the objective of minimizing standby capacity during periods of greatest fluctuation across various scenarios, a sub-Bruker optimization model is constructed, specifically as follows: We construct a Wasserstein distance fuzzy set centered on empirical probability distribution using historical operational data, and based on the Wasserstein distance fuzzy set, we construct a sub-Bruker optimization model with the objective of minimizing the backup capacity when the fluctuations are greatest in various scenarios.

4. The method for adaptive aggregation of virtual power plants across multiple time scales based on grid overload as described in claim 3, characterized in that, The method also includes: Obtain the intraday operational deviation dataset from the previous scheduling cycle, and extract historical deviation distribution characteristic indicators based on the intraday operational deviation dataset to serve as deviation conditional risk value. Based on the deviation condition risk value, the radius of the Wasserstein distance fuzzy set in the next scheduling cycle of the split-Bruker optimization model is adaptively adjusted and optimized.

5. The method for adaptive aggregation of virtual power plants across multiple time scales based on grid overload as described in claim 3, characterized in that, Based on the deviation conditional risk value, the radius of the Wasserstein distance fuzzy set of the split-bar optimization model is adaptively adjusted and optimized in the next scheduling cycle, specifically as follows: In the formula, Indicates the first The Wasserstein distance radius of the fuzzy set during the scheduling period. Indicates the first The Wasserstein distance radius of the fuzzy set during the scheduling period. Indicates a smooth learning rate. This represents the adjustment coefficient. This represents the value at risk of deviation conditions.

6. The method for adaptive aggregation of virtual power plants across multiple time scales based on grid overload as described in claim 1, characterized in that, Based on the distributed resource output prediction sequence and the regional load demand prediction sequence, a sparse regularization correction objective function is constructed by combining the gray box aggregation model, specifically as follows: In the formula, This represents the total system-level power regulation required to eliminate the current heavy overload or frequency over-limit. Indicates the first A gray-box aggregation model for distributed resources. This represents the baseline scheduling plan before the revision. This indicates the modification of the scheduling plan vector. Represents the sparsity penalty coefficient. express Norm.

7. A virtual power plant multi-timescale adaptive aggregation system based on grid overload, characterized in that, This system is used in the virtual power plant multi-timescale adaptive aggregation method based on grid overload as described in any one of claims 1-6, wherein the system comprises: The aggregation model generation module is used to extract physical operators from the physical mechanism equations of various distributed resources, and based on the historical operation data of various distributed resources, generate gray box aggregation models of various distributed resources under the constraints of the physical operators using a symbolic regression algorithm. The scheduling plan generation module is used to construct a two-stage stochastic programming model in the day-ahead phase, with the goal of minimizing expected operating costs and combining pre-built constraints, and solve the two-stage stochastic programming model to obtain a long-term stable time-scale scheduling scheme; with the goal of minimizing reserve capacity when various scenarios have the greatest fluctuations, a sub-Bruker optimization model is constructed, and the sub-Bruker optimization model is solved under the constraints of the long-term stable time-scale scheduling scheme to obtain a short-term reserve scheduling scheme; a baseline scheduling plan is generated based on the long-term stable time-scale scheduling scheme and the short-term reserve scheduling scheme. The execution deviation generation module is used to obtain the aggregated power deviation after the distributed resource execution benchmark scheduling plan and the system frequency change rate of the grid common coupling point during the intraday execution phase; when the aggregated power deviation or the system frequency change rate exceeds the preset safety threshold, it obtains multi-dimensional measurement data of the regional power grid within a preset time period. The scheduling plan correction module is used to input multidimensional measurement data into a pre-trained deep learning prediction model to obtain renewable energy output prediction sequences and regional load demand prediction sequences. Based on the distributed resource output prediction sequences and regional load demand prediction sequences, a sparse regularization correction objective function is constructed by combining a gray box aggregation model. The sparse regularization correction objective function is solved to obtain the corrected baseline scheduling plan.

8. A computer device comprising a system memory and a processor, wherein the system memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1 to 6.

10. A computer program product containing instructions, characterized in that, When the instruction is executed by a cluster of computer devices, the cluster of computer devices causes the cluster of computer devices to perform the method as described in any one of claims 1 to 6.

Images