Virtual power plant deaggregation method of projection enhanced distributed aggregated gradient tracking

Abstract

The application belongs to the field of power grid dispatching, and specifically discloses a virtual power plant disaggregation method based on projection-enhanced distributed aggregated gradient tracking. The application constructs a global optimization problem aiming at minimizing the sum of local operation costs of each distributed energy source and tracking the aggregated power instruction, ensuring the global optimality of the dispatching scheme. In the distributed iteration, each distributed energy source updates the output value based on the neighbor variable in parallel, strictly satisfies the local feasible region constraint through the projection operation, and guarantees the engineering feasibility; meanwhile, the aggregated variable and the gradient tracking variable are updated, realizing the distributed estimation and propagation of the global gradient information, so that the local output adjustment tracks the global optimal direction. When the iteration converges, the output values of each distributed energy source are output, so that the disaggregation result meets all local operation limits and approximates the global cost optimum. Compared with the prior art, the application improves the applicability and reliability of the virtual power plant disaggregation, and the whole process does not need to introduce optimization software or solver, thereby guaranteeing the practical engineering application.

Description

Technical Field

[0001] This application belongs to the field of power grid dispatching technology, specifically the field of intelligent dispatching system technology, and more specifically, to a virtual power plant de-aggregation method with projection-enhanced distributed aggregation gradient tracking. Background Technology

[0002] The operation of a virtual power plant consists of two parts: aggregation and deaggregation. Aggregation integrates a large number of distributed energy sources within the virtual power plant into a single entity, enabling them to participate in grid operation and thus reducing the complexity of distributed energy management. Deaggregation, on the other hand, involves receiving the total aggregated output and distributing it to the individual distributed energy sources.

[0003] Currently, virtual power plant de-aggregation typically employs centralized scheduling, aiming to maximize the total profit of the virtual power plant while satisfying the output constraints of individual distributed energy sources and system operational constraints. However, in practical applications, centralized optimization lacks consideration for the performance and operational differences of distributed energy sources, resulting in problems such as local infeasibility and high computational load. Distributed scheduling, on the other hand, can only achieve the optimal output of each distributed energy source, but cannot achieve global optimization for the entire virtual power plant, and it cannot satisfy the condition that the output of each distributed energy source equals the total aggregated output.

[0004] In addition, the software performance of edge terminals in real-world engineering scenarios is limited, making it impossible to install large-scale and complex optimization software and solvers. These factors pose challenges to existing solution aggregation schemes. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application aims to provide a projection-enhanced distributed aggregation gradient tracking method for virtual power plant de-aggregation, which addresses the technical problem that current virtual power plant de-aggregation methods struggle to balance feasibility and globally optimal scheduling, thus limiting the applicability of de-aggregation results in engineering applications.

[0006] The first aspect of this application relates to a projection-enhanced distributed aggregation gradient tracking method for de-aggregating a virtual power plant, comprising: step S10, obtaining the aggregation power command of the virtual power plant, and constructing a global optimization objective based on the local operating cost of each distributed energy source and the aggregation power command; the global optimization objective is to minimize the sum of the local operating costs of each distributed energy source and to make the sum of the output values ​​of each distributed energy source track the aggregation power command; step S20, in response to the global optimization objective, initializing the output value of each distributed energy source, the aggregation variable, and the gradient tracking variable; wherein, the aggregation variable is the local estimate of the sum of the output values ​​of the current distributed energy source to the virtual power plant; the gradient tracking variable is the gradient of the local operating cost of the current distributed energy source with respect to the aggregation variable. Local estimated value; Step S30: Based on the aggregated variables and gradient tracking variables obtained from the communication and interaction between the distributed energy sources, execute in parallel in a distributed iterative manner: determine the output update direction according to the current output value and the current gradient tracking variable, and then update the output value that satisfies the local feasible region constraint projection according to the current output value and the output update direction; update the local aggregated variable according to the updated output value and the aggregated variables of other distributed energy sources; update the local gradient tracking variable according to the updated output value, the updated aggregated variable and the gradient tracking variables of other distributed energy sources; Step S40: When the output value and aggregated variable of each distributed energy source converge, output the current output value of each distributed energy source to achieve de-aggregation; otherwise, return to step S30.

[0007] In one implementation, the local operating cost is the sum of the generation cost or electricity utility function of the distributed energy source and the system balance penalty term for the deviation between the aggregated variable and the aggregated power command.

[0008] In one embodiment, step S20 includes: taking the actual output value of each distributed energy source before receiving the aggregated power command as the initial value of the local output value; taking the initial output value as the initial value of the local aggregated variable; and taking the gradient of the local operating cost with respect to the initial aggregated variable as the initial value of the local gradient tracking variable.

[0009] In one embodiment, the output update direction is determined based on the current output value and the current gradient tracking variable. Then, the output value that satisfies the local feasible region constraint projection is updated based on the current output value and the output update direction. Specifically, based on the gradient descent principle, the current output value is moved by a preset step along an output update direction to obtain an intermediate update value. The output update direction is determined by the gradient of the local operating cost with respect to the current output value and a correction term based on the gradient tracking variable. The intermediate update value is then projected onto the local feasible region to obtain the updated output value.

[0010] In one embodiment, the local aggregate variable is updated based on the updated output value and the aggregate variables of other distributed energy sources. Specifically, the aggregate variables from other distributed energy sources are weighted and summed according to a preset communication weight, and the change in aggregate contribution caused by the update of its own output value is added to update the local aggregate variable. The local gradient tracking variable is updated based on the updated output value, the updated aggregate variable, and the gradient tracking variables of other distributed energy sources. Specifically, the gradient tracking variables from other distributed energy sources are weighted and summed according to a preset communication weight, and the change in the gradient of its own local operating cost with respect to the aggregate variable before and after the iteration is added to update the local gradient tracking variable.

[0011] In one embodiment, convergence is defined as follows: the change in the output value of all distributed energy sources in adjacent iterations is less than a first preset threshold, and the deviation between the aggregated variable and the aggregated power command of all distributed energy sources is less than a second preset threshold.

[0012] In one embodiment, the communication topology corresponding to the communication interaction between each distributed energy source is a strongly connected graph, and the communication weight between any two distributed energy sources satisfies a double random matrix.

[0013] In a second aspect, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0014] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0015] Fourthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0016] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0017] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art:

[0018] This application employs a projection-enhanced distributed aggregated gradient tracking method. First, it constructs a global optimization problem with the goal of minimizing the sum of the local operating costs of each distributed energy source and tracking aggregated power commands, thus mathematically ensuring the global optimality of the scheduling scheme. During the distributed iteration process, each distributed energy source updates its output value in parallel based on neighbor variables obtained through communication interaction and strictly satisfies the local feasible region constraints through projection operations. This directly guarantees the engineering feasibility of the solution. At the same time, through the coordinated updating of aggregated variables and gradient tracking variables, the distributed estimation and propagation of global gradient information are realized, enabling local output adjustments to effectively track the global optimal direction.

[0019] Finally, upon iteration convergence, the output values ​​of each distributed energy source are output, ensuring that the de-aggregation result approximates the global cost optimum while satisfying all local operational constraints. This solves the problem of existing technologies struggling to balance feasibility and globally optimal scheduling, significantly improving the applicability and reliability of virtual power plant de-aggregation in complex engineering environments compared to centralized or simple distributed methods. Furthermore, this method does not require the introduction of any optimization software or solver during all iterative calculations, guaranteeing its application in practical engineering. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the virtual power plant de-aggregation method with projection-enhanced distributed aggregation gradient tracking provided in an embodiment of this application.

[0021] Figure 2 This is a conceptual diagram of virtual power plant operation and depolymerization provided in the embodiments of this application;

[0022] Figure 3 This is a communication topology diagram between distributed energy sources provided in the embodiments of this application;

[0023] Figure 4 This is a diagram illustrating the iterative output values ​​of various distributed energy sources in the examples provided in this application's embodiments;

[0024] Figure 5 This is a graph showing the aggregation variable as a function of iteration error in an example provided in this application embodiment;

[0025] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0027] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.

[0028] In this application, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0029] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0030] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0031] Currently, the de-aggregation operation of virtual power plants traditionally relies on centralized scheduling strategies. The goal is to maximize the overall profit of the virtual power plant while adhering to the output limits of each distributed energy unit and the overall system operational constraints. However, in actual deployment, this centralized optimization approach often ignores the individual differences in performance and operating status among different distributed energy sources, leading to insufficient local feasibility and high computational complexity. Conversely, while decentralized scheduling can achieve locally optimal output for each distributed energy unit, it is difficult to guarantee the global optimality of the virtual power plant, nor can it ensure that the sum of the outputs of each unit accurately matches the total aggregated output demand.

[0032] Based on this, this application proposes a projection-enhanced distributed aggregation gradient tracking method for virtual power plant de-aggregation. Please refer to... Figure 1 , Figure 1 This is a schematic flowchart of the virtual power plant de-aggregation method with projection-enhanced distributed aggregation gradient tracking provided in the embodiments of this application.

[0033] In this embodiment, the method includes steps S10 to S40.

[0034] Step S10: Obtain the aggregated power command of the virtual power plant, and construct a global optimization target based on the local operating cost of each distributed energy source and the aggregated power command.

[0035] Step S20: In response to the global optimization objective, initialize the output value, aggregation variable, and gradient tracking variable of each distributed energy source.

[0036] Step S30: Based on the aggregated variables and gradient tracking variables obtained from the communication and interaction between the distributed energy sources, the following steps are executed in parallel using a distributed iterative approach: Determine the output update direction based on the current output value and the current gradient tracking variable, and then update the output value that satisfies the projection of the local feasible region constraint based on the current output value and the output update direction; update the local aggregated variable based on the updated output value and the aggregated variables of other distributed energy sources; update the local gradient tracking variable based on the updated output value, the updated aggregated variable, and the gradient tracking variables of other distributed energy sources.

[0037] Step S40: When the output value and aggregation variable of each distributed energy source converge, output the current output value of each distributed energy source to achieve de-aggregation; otherwise, return to step S30.

[0038] First, please refer to Figure 2 , Figure 2 This is a conceptual diagram of the operation and de-aggregation of a virtual power plant provided in an embodiment of this application. The distributed energy sources aggregated in the virtual power plant include distributed photovoltaics, wind power generation units, energy storage systems, gas turbines, etc., and the sum of the output values ​​of all distributed energy sources must equal the total aggregated power of the virtual power plant.

[0039] Understandable, Figure 2 This paper describes a three-tiered collaborative architecture aimed at balancing centralized dispatch and distributed autonomy. Its core lies in achieving bidirectional information flow and coordinated optimization through a key hub: instructions from the dispatch center or electricity market are transmitted top-down, while heterogeneous distributed resources such as photovoltaics and energy storage provide feedback on their real-time operating parameters from the bottom up. The operator integrates global objectives and local information to generate coordination schemes and can issue de-aggregation signals to resource units to initiate their autonomous responses. This hybrid model of centralized guidance and decentralized execution, through bidirectional closed-loop communication, ensures that the sum of the outputs of each unit accurately matches the aggregated power required by the system. Theoretically, this approach can balance global optimization, resource diversity, and operational feasibility, effectively addressing the challenges faced by purely centralized or decentralized dispatching.

[0040] Specifically, in step S10, the aggregated power command of the virtual power plant refers to the total power target value issued by the superior power dispatch center or the power market, requiring the virtual power plant to transmit or absorb power from its point of common coupling to the grid within a specific time period. Specifically, it is usually a power curve that varies over time, representing the overall dispatch demand of the grid side for the virtual power plant aggregate. This command is the core input and overall constraint target for optimal dispatch or deaggregation calculations within the virtual power plant; the output plans of all internal distributed resources must ultimately satisfy this overall aggregated power command as a fundamental premise.

[0041] It is understandable that the application scenario of this application is precisely to de-aggregate power sources to enable them to output power, given the known aggregated power command of the virtual power plant. Therefore, how to obtain the aggregated power command of the virtual power plant is not limited in this application. It can be achieved through conventional methods such as direct issuance by the superior dispatching agency or participation in electricity market transactions, which will not be elaborated here.

[0042] It is worth noting that the core of step S10 lies in constructing a global optimization objective based on the local operating costs and aggregated power commands of each distributed energy source. Essentially, this step transforms an engineering economic problem into a mathematical optimization problem solvable on a computing device. Its purpose is to provide a clear and quantifiable evaluation criterion for subsequent solution algorithms, automatically selecting the optimal solution from all feasible distributed energy output schemes.

[0043] Specifically, the global optimization objective is typically constructed as an objective function. The main body of this objective function aims to minimize the sum of the local operating costs of each distributed energy source. Local operating cost is a key economic indicator, and its specific composition varies depending on the type of distributed energy source. For example, for fossil fuel power generation units such as gas turbines and diesel generators, the cost mainly includes fuel procurement costs, equipment operating losses, and emissions costs.

[0044] Simultaneously, this global optimization objective must satisfy a fundamental system balance constraint: ensuring that the sum of the output values ​​of all distributed energy sources tracks the aggregated power command. Mathematically, this tracking can be represented by a set of strict equality constraints or penalty terms. In most models, this is presented as an equality constraint, requiring that the total output of all distributed energy sources within a given scheduling period must be strictly equal to the aggregated power command value received by the virtual power plant from its superior during that period. This is the physical basis for ensuring that the virtual power plant, as a whole, reliably responds to external scheduling and maintains real-time power balance in the power grid.

[0045] Therefore, the global optimization objective is to minimize the sum of the local operating costs of each distributed energy source, and to ensure that the sum of the output values ​​of each distributed energy source tracks the aggregated power command. The specific expression for the global optimization objective function is as follows:

[0046] (1).

[0047] in, This means "defined as"; For the first The local operating cost function of the distributed energy source depends on the first... Output value of distributed energy sources and aggregate functions This is also to optimize both cost and output tracking simultaneously; This represents the total number of distributed energy resources in the virtual power plant. for The vector formed Indicates It is a vector composed of elements, and the meaning of the rest col(·) can be deduced in the same way. It is mainly used to unify the representation of numbers.

[0048] It is understandable that formula (1) expresses the idea of ​​minimizing the sum of the local operating costs of each distributed energy source.

[0049] It should be noted that aggregate functions ,in It is the first The contribution of each distributed energy source to the total aggregate power, and .

[0050] As explained above, the output of each distributed energy source is its contribution to the total aggregated power. This is because the aggregation function ultimately needs to approximate the total aggregated power, meaning the sum of the output values ​​of each distributed energy source also needs to approximate the total aggregated power. Here, the total aggregated power is essentially the aggregated power command value. Therefore, the aggregation function is expressed as above.

[0051] Furthermore, the local operating cost is expressed using a local operating cost function:

[0052] (2).

[0053] Specifically, For the first The generation cost or electricity utility function of a distributed energy source precisely quantifies the economics of its operation. The latter part of the summation is the system power balance penalty, which occurs if the total output of the virtual power plant does not equal the total aggregate power. This part is intended to guide the sum of the overall output of the virtual power plant in the optimization problem, that is, to ensure that the aggregate function equals the total aggregate power.

[0054] Understandable, This is the system balance penalty coefficient. It can be set to be the same for all units, or it can be set to different values ​​according to unit characteristics, such as reliability and response speed, so as to assign different adjustment responsibilities.

[0055] Therefore, formula (2) expresses the local operating cost as the generation cost or electricity utility function of distributed energy sources, and the sum of the system balance penalty term for the deviation between the aggregated variable and the aggregated power command. The core is to make the sum of the output values ​​of each distributed energy source track the connotation of the aggregated power command. The combination of formula (1) and formula (2) realizes the construction of the global optimization objective.

[0056] After determining the global optimization objective, step S20 needs to be executed. That is, in response to the objective, the core parameters of this application need to be initialized, namely: the output value of each distributed energy source, the aggregation variable, and the gradient tracking variable.

[0057] It should be noted that these three parameters are the core parameters of the distributed optimization method involved in this application because they together constitute the complete state space, information interaction basis, and convergence driving mechanism of the algorithm. They respectively represent the optimization object, global cognition, and collaborative guidance, and none of them can be omitted.

[0058] Among them, the output value is the direct object of optimization, and its initial value provides a feasible starting point for subsequent iterations. The aggregate variable is the local estimate of the sum of the output values ​​of the current distributed energy sources to the virtual power plant; the gradient tracking variable is the local estimate of the gradient of the local operating cost of the current distributed energy sources with respect to the aggregate variable.

[0059] Understandably, the output value represents the power generation or load capacity of each distributed energy source. The ultimate goal of the entire optimization process is to find a set of optimal output values ​​for all units, minimizing the total operating cost of the virtual power plant and accurately tracking aggregation commands while satisfying their respective constraints. The design and updating of all other parameters ultimately aim to guide the output values ​​to converge to this set of optimal values. Therefore, the output value is the fundamental goal of the algorithm, its very core.

[0060] Understandably, in centralized optimization, the central controller directly knows all output values ​​and calculates the sum. However, in a fully distributed architecture, no node possesses global information. The introduction of aggregate variables is precisely to address this core challenge. As a local estimate of the total global output for each node, iteratively updates aggregate variable information by nodes exchanging it only with other distributed energy sources, ensuring that the aggregate variables of all nodes eventually converge to the true global sum. This process enables each node to form a consistent understanding of the system's critical global state without sharing private data output values, providing the information foundation for distributed collaborative optimization.

[0061] Understandably, even if each node knows the globally estimated aggregate variables, ensuring that its decisions based on local costs lead to the global optimum, rather than a selfish local optimum, requires gradient tracking variables. Based on the local operating cost function, its gradient information with respect to the global aggregate variables contains the system balance requirements, guiding each node to adjust its output: when the total system output is insufficient, it encourages lower-cost units to generate more power; conversely, it reduces output. Therefore, gradient tracking variables are the coordination mechanism and driving force that keeps local decisions consistent with the global objective.

[0062] In one embodiment, step S20 includes: taking the actual output value of each distributed energy source before receiving the aggregated power command as the initial value of the local output value; taking the initial output value as the initial value of the local aggregated variable; and taking the gradient of the local operating cost with respect to the initial aggregated variable as the initial value of the local gradient tracking variable.

[0063] It is understandable that the actual output value of each distributed energy source just before receiving the aggregated power command is set as the initial value of its local optimization variable output value. This actual output value usually refers to the power value measured and reported in real time by local sensors, such as power transmitters and smart meters.

[0064] Understandably, since communication has not yet taken place and it is difficult to calculate the sum, the initial output value determined in the previous step can be directly assigned to the initial value of the local aggregation variable, and a closer iterative value can be obtained after one communication.

[0065] Understandably, based on the initial aggregation variables and the known aggregation power command, the gradient component related to system equilibrium in the local operating cost function is calculated. For the local operating cost function, i.e., the partial derivative of equation (2) with respect to the aggregation variables, this gradient component is: Then, substituting the above information, we use this calculation result as the initial value for the local gradient tracking variable.

[0066] Following the acquisition of initial values, an iterative process is required to obtain the convergence output values ​​of each distributed energy source. The main method employed in this application is to use the aggregated variables and gradient tracking variables obtained through communication and interaction between the distributed energy sources in step S30 to perform parallel calculations of multiple distributed energy sources in a distributed iterative manner.

[0067] It should be noted that the core of the distributed iterative optimization method is that each distributed energy source only exchanges limited information with its neighboring nodes in the communication network during the iteration, and updates its own variables based on local computation, thereby solving the global problem in a completely parallel manner.

[0068] It should be noted that each iteration, denoted as the k-th iteration, typically comprises two phases: information exchange and local computation. In the information exchange phase, each distributed energy source sends its local aggregated variables and gradient tracking variables from the previous iteration to its communication neighbors. Simultaneously, it receives the corresponding aggregated variables and gradient tracking variables from its neighbors. In the subsequent local computation phase, each distributed energy source fuses the received neighbor information with its own information, and independently and in parallel calculates the new round of local output value, aggregated variables, and gradient tracking variables according to a predetermined mathematical update rule. This process is repeated until the changes in all variables are less than the preset convergence tolerance, or the maximum number of iterations is reached. At this point, the output value of each output value is the desired converged output value. Mathematically, this involves using distributed algorithms, such as subgradient-based distributed optimization algorithms and distributed variants of the Alternating Direction Multiplier Method (ADMM), to decompose and solve centralized optimization problems.

[0069] Understandably, each distributed energy source's local controller executes its own variable update calculations synchronously or asynchronously within the same iteration cycle. This parallelism is reflected in the fact that once a node has collected the necessary neighbor information, its calculation process is completely independent and does not depend on the scheduling of the central node. This is achieved by deploying the same algorithm calculation program in each local controller, and the calculation core can be an embedded microprocessor, digital signal processor, or field-programmable gate array.

[0070] Specifically, communication can be achieved using various network topologies and protocols. For example, the communication network can be a point-to-point mesh network, a bus network, or a ring network. The communication protocol can be the standard TCP / IP protocol, IEC 61850 for power systems, DNP3.0, or a customized real-time communication protocol. The physical devices that implement this communication function include, but are not limited to: industrial communication modules, network switches, gateway devices, and corresponding communication cables or wireless access points installed on each distributed energy source side.

[0071] In one embodiment, the communication topology corresponding to the communication interaction between each distributed energy source is a strongly connected graph, and the communication weight between any two distributed energy sources satisfies a double random matrix.

[0072] Understanding this concept involves viewing each distributed energy source as a node in a graph, and the direct communication links between them as edges. A strongly connected graph is one where there is a path from any node to any other node. This means that information flow has no directional barriers or isolated points in the network. In engineering implementation, this requires that the physical or logical connections of the communication network ensure that all participants can ultimately exchange information directly or indirectly. For example, this can be achieved by deploying mesh networks, fully connected networks, or ensuring bidirectional connectivity in ring networks. Its core function is to guarantee the global coordination information required by the optimization algorithm, i.e., that aggregate variables and gradient tracking variables can be fully propagated and mixed throughout the network. If the topology is not strongly connected, it may form subgroups where information cannot be exchanged, preventing the entire system from achieving global consensus and collaborative optimization.

[0073] Understandingly, communication weights quantify the proportion of trust or allocation a node places in its own information and the information of its neighbors when updating its own variables. A double-random matrix is ​​a mathematical matrix whose elements represent the weights between nodes, satisfying two conditions: all elements are non-negative; the sum of the elements in any row is 1 (row random); and the sum of the elements in any column is also 1 (column random). In distributed iteration, each node i calculates the weighted sum of its own and its neighbor j's aggregated variable values ​​when updating its aggregated variable; the weights used constitute a double-random matrix. The deeper purpose of the double-random condition is to ensure that the total amount of information from all nodes is conserved during the information aggregation process of the entire network, preventing decay or expansion during iteration. This is the mathematical basis for driving variable values ​​to converge to the correct global average. This weight matrix can be pre-calculated and stored in the memory of the local controller, or it can be generated online by a distributed algorithm.

[0074] Specifically, let the communication weight matrix be... For one OK The column matrix corresponds to a strongly connected graph in the communication topology and is a doubly random matrix.

[0075] (3).

[0076] in, To meet the distributed energy needs of two communications and Communication weights with double randomness For distributed energy The set of all distributed energy sources for communication. Satisfying if... , ,otherwise .

[0077] Furthermore, the calculation process in step S30 can be expanded as follows:

[0078] Step S31: Determine the output update direction based on the current output value and the current gradient tracking variable, and then update the output value that satisfies the local feasible region constraint projection based on the current output value and the output update direction.

[0079] Understandably, this step aims to determine a new output plan for each distributed energy source that both follows global coordination signals and strictly adheres to its own operational limitations.

[0080] First, the output update direction is determined based on the current output value and the current gradient tracking variable, which essentially involves calculating an optimization direction. This direction is synthesized from two pieces of information: one is the gradient of the local cost function at the current output value, which indicates the direction in which the local output should be adjusted to reduce its own operating cost; the other is the current gradient tracking variable, which carries global gradient consensus information regarding the system power balance requirement after network coordination. Combining these two results in a comprehensive optimization direction that simultaneously considers individual economics and system balance. Subsequently, the output value that satisfies the projection of the local feasible region constraint is updated based on the current output value and the output update direction. This means that the current output value is initially adjusted along the calculated direction with a preset step size.

[0081] Then, this preliminary adjustment result is projected onto the local feasible region of the distributed energy source. Projection is a mathematical operation whose function is: if the preliminary adjustment result exceeds the range formed by physical constraints such as the upper and lower limits of the unit's technical output and the ramp rate, it is pulled back to the nearest feasible point within that range; if it does not exceed the limits, it remains unchanged.

[0082] This operation ensures that the output plan generated in each iteration is physically executable. The device that implements this step is the local controller of each distributed energy source. Its internal optimization calculation module is responsible for calculating the direction, and the constraint processing module performs the projection operation. The required data comes from locally stored cost parameters, real-time received coordination variables, and preset equipment operating constraints.

[0083] Understandably, the local feasible region is defined as follows: if the first... One distributed energy source is a micro gas turbine, and its local feasible region is... As shown below:

[0084] (4).

[0085] Specifically, This represents the upper and lower limits of the output of the micro gas turbine. This represents the upper and lower limits of the ramp-climbing range for this micro gas turbine.

[0086] Preferably, if the first Each distributed energy source is an energy storage unit, and its local feasible region is... As shown below:

[0087] (5).

[0088] Specifically, This represents the maximum charge and discharge power of the energy storage unit. These are the upper and lower limits of the energy storage capacity of this energy storage unit. This represents the initial energy stored in the energy storage unit.

[0089] Preferably, if the first Each distributed energy source is a distributed photovoltaic or wind power generation unit, and its local feasible area is... As shown below:

[0090] (6).

[0091] Specifically, This refers to the maximum output value of a distributed photovoltaic or wind power generation unit. This represents the maximum wind / solar curtailment rate.

[0092] In one embodiment, step S31 specifically involves: based on the gradient descent principle, moving the current output value along an output update direction by a preset step size to obtain an intermediate update value; wherein, the output update direction is determined by the gradient of the local operating cost with respect to the current output value and a correction term based on a gradient tracking variable; and projecting the intermediate update value onto the local feasible region to obtain the updated output value.

[0093] Based on the above, in the... In each iteration, each distributed energy source updates its own output value. The formula is as follows:

[0094] (7).

[0095] Specifically, Step size, To the local feasible domain The projection operation. The gradient calculation required in equation (7) , .

[0096] Understandably, the latter part is the gradient correction term, which consists of two parts: the first part is the gradient of the local cost objective function with respect to the output value, and the second part is the effect propagated back through the gradient of the aggregated information. The reason for this design is that the local cost objective function depends not only on the output value but also on the aggregated variables, and the aggregated variables depend on all output values. Therefore, the gradient needs to include the indirect influence of the aggregated variables on the output value, and then combine the gradient tracking variables to correct the direction. Equation (7) is essentially a gradient descent with gradient correction to update the decision variables.

[0097] Step S32: Update the local aggregate variable based on the updated output value and the aggregate variables of other distributed energy sources.

[0098] After the local output value is updated, each unit needs to adjust its estimate of the total system output accordingly, which is the aggregation variable. This is a process that combines information fusion and incremental injection.

[0099] Specifically, each distributed energy source obtains its previous round's aggregate variable estimates from its communication neighbors. Then, the local controller performs a weighted summation operation: it multiplies its own previous round's aggregate variable estimate by the corresponding values ​​of all its neighbors, each by a preset communication weight, and then adds them together. These weights form a double-random matrix, ensuring unbiased information propagation throughout the network. Next, the latest change in local output—the difference between the new output value after this iteration and the old output value before the iteration—is injected into the weighted summation result. Through this operation, the updated local aggregate variables not only incorporate the neighbors' estimates but also reflect the latest changes in its own state, thereby driving all units in the network to gradually converge their estimates of the global total output and approach the true total. This step relies on the communication module receiving neighbor data and the controller's arithmetic logic unit performing the weighted summation and addition operations.

[0100] In one embodiment, step S32 specifically involves: weighting and summing the aggregation variables from other distributed energy sources according to preset communication weights, and adding the change in aggregation contribution caused by the update of its own output value, in order to update the local aggregation variables.

[0101] Based on the above, in the... In each iteration, each distributed energy source updates its estimate of the aggregate variable by exchanging information with other distributed energy sources it communicates with, as shown in the following formula:

[0102] (8).

[0103] All parameters in equation (8) have been defined above and will not be repeated here.

[0104] Step S33: Update the local gradient tracking variable based on the updated output value, the updated aggregate variable, and the gradient tracking variables of other distributed energy sources.

[0105] The final step is to update the global gradient signal estimate used to coordinate the actions of each unit, i.e., the gradient tracking variable. This step is structurally similar to updating the aggregate variable. First, a weighted sum is taken of the gradient tracking variable from the previous iteration and the corresponding values ​​received from its neighbors. Simultaneously, a change in a local gradient component needs to be calculated and injected. This change is the difference between the system equilibrium penalty gradient recalculated based on the updated local aggregate variable and the old gradient value calculated based on the local aggregate variable before the update. By adding the weighted sum to this local increment, the updated gradient tracking variable can more accurately track the true gradient of the entire system's objective function with respect to the total output constraint, thus providing a more precise coordination direction for each unit in the next iteration.

[0106] In one embodiment, step S33 specifically involves: weighting and summing the gradient tracking variables from other distributed energy sources according to preset communication weights, and adding the change in the gradient of the local operating cost with respect to the aggregated variable before and after the iteration, in order to update the local gradient tracking variables.

[0107] Based on the above, in the... In each iteration, each distributed energy source updates its estimate of the gradient tracking variable by exchanging information with other distributed energy sources it communicates with, as shown in the following formula:

[0108] (9).

[0109] The required gradient calculation is as follows: .

[0110] Finally, the iterative process shown in this application can be integrated into equations (4) to (9).

[0111] In this application, convergence in step S40 means that the change in output value of all distributed energy sources in adjacent iterations is less than a first preset threshold, and the deviation between the aggregated variable and the aggregated power command of all distributed energy sources is less than a second preset threshold. That is, when both of the following preset conditions are met simultaneously, the distributed energy source de-aggregation problem converges, and each distributed energy source operates according to the desired output value:

[0112] (10).

[0113] Specifically, , This is the convergence threshold.

[0114] In this embodiment, by employing a projection-enhanced distributed aggregated gradient tracking method, a global optimization problem is first constructed with the goal of minimizing the sum of the local operating costs of each distributed energy source and tracking aggregated power commands. This mathematically ensures the global optimality of the scheduling scheme. During the distributed iteration process, each distributed energy source updates its output value in parallel based on the neighbor variables obtained through communication interaction and strictly satisfies the local feasible region constraints through projection operations. This directly guarantees the engineering feasibility of the solution. At the same time, through the coordinated updating of aggregated variables and gradient tracking variables, the distributed estimation and propagation of global gradient information are realized, enabling local output adjustments to effectively track the global optimal direction.

[0115] Finally, upon iteration convergence, the output values ​​of each distributed energy source are output, ensuring that the de-aggregation result approximates the global cost optimum while satisfying all local operational constraints. This solves the problem of existing technologies struggling to balance feasibility and globally optimal scheduling, significantly improving the applicability and reliability of virtual power plant de-aggregation in complex engineering environments compared to centralized or simple distributed methods. Furthermore, this method does not require the introduction of any optimization software or solver during all iterative calculations, guaranteeing its application in practical engineering.

[0116] Specifically, the beneficial effects of this embodiment will be further described below with reference to a specific application scenario. For a virtual power plant aggregating 20 distributed energy sources, including 2 small wind turbines, 3 micro gas turbines, 5 distributed photovoltaic systems, and 10 energy storage units, the communication links between the distributed energy sources are as follows: Figure 3 As shown, Figure 3 This is a communication topology diagram between distributed energy sources provided in the embodiments of this application. The initial output value of all distributed energy sources is set to the median of the upper and lower limits of the processing range, and the initial storage capacity of the energy storage unit is 50% of its maximum storage capacity. A convergence threshold is set. The power required for the demand response is 700kW.

[0117] Specifically, the data for the micro gas turbine, energy storage unit, and wind and solar power generation are shown in Tables 1, 2, and 3, respectively. A corresponding program was written in the computational software MATLAB R2024a to de-aggregate the virtual power plant. The computing device used was a Thinkbook 16 laptop with an Intel Core 5 220H (2.70 GHz) processor, 24GB of RAM, and running Windows 11 Professional operating system.

[0118] Table 1. Parameters of micro gas turbines:

[0119]

[0120] Table 2. Energy Storage Unit Parameters:

[0121]

[0122] Table 3. Wind and Solar Power Output Parameters:

[0123]

[0124] Please refer to the solution results for further information. Figure 4 and Figure 5 , Figure 4 This is a diagram showing the iterative output values ​​of various distributed energy sources in the example provided in this application embodiment. Figure 4 This paper illustrates that, driven by the distributed optimization algorithm, the virtual power plant, comprising 20 heterogeneous distributed energy sources including wind power, photovoltaics, gas turbines, and energy storage, dynamically adjusts its planned output values ​​with increasing iterations, ultimately converging to a stable set of optimal values ​​after approximately 123 iterations. The curves in the figure show that different energy sources exhibit different convergence behaviors based on their techno-economic characteristics: gas turbines, as controllable power sources, show the most significant output adjustment; photovoltaic and wind power outputs are relatively stable, reflecting their adherence to natural resource characteristics; while energy storage units flexibly switch between charging, negative and discharging, and positive states, playing a crucial role in power balancing and regulation. The output curves of all units rapidly stabilize from their initial dispersed state, intuitively demonstrating that the algorithm can effectively coordinate diverse resources and solve for the optimal scheduling scheme that satisfies the global objective in a distributed manner.

[0125] Figure 5 This is a diagram showing the aggregation variable as a function of iteration error in an example provided in this application embodiment. Figure 5 This demonstrates that the deviation between the total output of the virtual power plant and the 700kW aggregated power command decreases rapidly during the iteration process and eventually approaches zero. The power difference curve starts from an initial -356.00kW, experiences limited fluctuations during iteration, and converges to 0.00kW at a relatively fast rate, with the final convergence difference strictly meeting the preset threshold of 0.007kW. This proves that the algorithm can accurately achieve the core constraint of the sum of the output values ​​of each distributed energy source tracking the aggregated power command, ensuring the reliable response of the virtual power plant as a whole to external dispatch commands. Furthermore, the convergence process is smooth, without drastic oscillations, indicating that the algorithm has good numerical stability.

[0126] As can be seen, the solution time of this algorithm is only 0.08s. This is because the algorithm provided by this invention can converge to the global optimal solution at a linear convergence rate, with a fast response speed, and is suitable for the rapid scheduling scenario of virtual power plants participating in the real-time energy and ancillary services market.

[0127] Based on the methods in the above embodiments, please refer to Figure 6This application provides an electronic device that may include a processor, a communications interface, a memory, and a communication bus. The processor, communications interface, and memory communicate with each other via the communication bus. The processor can invoke logical instructions stored in the memory to execute the methods described in the above embodiments.

[0128] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0129] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0130] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0131] It is understood that the processor in the embodiments of this application 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, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0132] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the 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 the storage medium can reside in an ASIC.

[0133] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0134] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0135] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A virtual power plant de-aggregation method for projection-enhanced distributed aggregation gradient tracking, characterized in that, include: Step S10: Obtain the aggregated power command of the virtual power plant, and construct a global optimization target based on the local operating cost of each distributed energy source and the aggregated power command; The global optimization objective is to minimize the sum of the local operating costs of each distributed energy source and to make the sum of the output values ​​of each distributed energy source track the aggregated power command. Step S20: In response to the global optimization objective, initialize the output value, aggregation variable, and gradient tracking variable of each distributed energy source; Wherein, the aggregate variable is the local estimate of the sum of the output values ​​of the current distributed energy sources to the virtual power plant; the gradient tracking variable is the local estimate of the gradient of the local operating cost of the current distributed energy sources with respect to the aggregate variable; Step S30: Based on the aggregated variables and gradient tracking variables obtained from the communication and interaction between the distributed energy sources, the following steps are executed in parallel using a distributed iterative approach: the output update direction is determined according to the current output value and the current gradient tracking variable; then, the output value that satisfies the projection of the local feasible region constraint is updated according to the current output value and the output update direction; the local aggregated variable is updated according to the updated output value and the aggregated variables of other distributed energy sources; and the local gradient tracking variable is updated according to the updated output value, the updated aggregated variable, and the gradient tracking variables of other distributed energy sources. Step S40: When the output values ​​and aggregation variables of each distributed energy source converge, output the current output value of each distributed energy source to achieve de-aggregation; otherwise, return to step S30.

2. The virtual power plant de-aggregation method for projection-enhanced distributed aggregation gradient tracking as described in claim 1, characterized in that, The local operating cost is the sum of the generation cost or electricity utility function of the distributed energy source, and the system balance penalty term for the deviation between the aggregated variable and the aggregated power command.

3. The virtual power plant de-aggregation method for projection-enhanced distributed aggregation gradient tracking as described in claim 2, characterized in that, Step S20 includes: The actual output value of each distributed energy source before receiving the aggregated power command is used as the initial value of the local output value; Use the initial output value as the initial value for the local aggregation variable; Use the gradient of the local running cost with respect to the initial aggregate variable as the initial value of the local gradient tracking variable.

4. The virtual power plant de-aggregation method for projection-enhanced distributed aggregation gradient tracking as described in claim 1, characterized in that, The output update direction is determined based on the current output value and the current gradient tracking variable. Then, the output value that satisfies the projection of the local feasible region constraint is updated based on the current output value and the output update direction. Specifically: Based on the gradient descent principle, the current output value is moved by a preset step along an output update direction to obtain an intermediate update value; wherein, the output update direction is determined by the gradient of the local operating cost with respect to the current output value and a correction term based on the gradient tracking variable. The intermediate update value is projected onto the local feasible domain to obtain the updated output value.

5. The virtual power plant de-aggregation method for projection-enhanced distributed aggregation gradient tracking as described in claim 1, characterized in that, The local aggregate variable is updated based on the updated output value and the aggregate variables of other distributed energy sources. Specifically, the aggregate variables from other distributed energy sources are weighted and summed according to the preset communication weight, and the change in aggregate contribution caused by the update of its own output value is added to update the local aggregate variable. Based on the updated output value, the updated aggregate variable, and the gradient tracking variables of other distributed energy sources, the local gradient tracking variable is updated. Specifically, the gradient tracking variables from other distributed energy sources are weighted and summed according to preset communication weights, and the change in the gradient of the local operating cost with respect to the aggregate variable before and after the iteration is added to update the local gradient tracking variable.

6. The virtual power plant de-aggregation method for projection-enhanced distributed aggregation gradient tracking as described in claim 1, characterized in that, The convergence is as follows: The change in the output value of all distributed energy sources in adjacent iterations is less than the first preset threshold, and the deviation between the aggregated variable of all distributed energy sources and the aggregated power command is less than the second preset threshold.

7. The virtual power plant de-aggregation method for projection-enhanced distributed aggregation gradient tracking as described in claim 1, characterized in that, The communication topology corresponding to the communication interaction between the distributed energy sources is a strongly connected graph, and the communication weight between any two distributed energy sources satisfies a double random matrix.

8. An electronic device, characterized in that, Includes memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the electronic device to perform the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on an electronic device, the electronic device causes the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that: When the computer program or instructions are run on an electronic device, the electronic device causes the electronic device to perform the method as described in any one of claims 1 to 7.

Images