Power grid dispatching method and device, electronic equipment and readable storage medium

Abstract

The application relates to a power grid dispatching method and device, electronic equipment and a readable storage medium, and relates to the technical field of electric power. The method comprises the following steps: acquiring current predicted output data of a first node and a first current parameter value corresponding to a target parameter in a first dispatching model of the first node; the first dispatching model is a dispatching model locally deployed by a first electronic device; the first current parameter value is obtained by updating a first historical parameter value corresponding to the target parameter in the first dispatching model of the first node based on parameter gradient information of the first dispatching model and a second historical parameter value corresponding to a target parameter in a second dispatching model of a second node; inputting the current predicted output data and the first current parameter value into the first dispatching model to obtain target decision output data of the first node; and issuing the target decision output data to the first node so that the first node outputs according to the target decision output data. The method can improve the reliability of power grid dispatching.

Description

Technical Field

[0001] This application relates to the field of power technology, and in particular to a power grid dispatching method, apparatus, electronic device, computer-readable storage medium, and computer program product. Background Technology

[0002] With the development of new energy technologies, various new energy sources, represented by wind power and photovoltaics, are rapidly being integrated into the power system. While this transformation brings clean electricity to the power grid, it also introduces core challenges such as strong volatility and uncertainty, posing a severe test to traditional power grid dispatching models.

[0003] In related technologies, research on dispatching for power grids with a high proportion of renewable energy has mainly developed along two paths: centralized and distributed. When distributed dispatching methods are applied to power grids with a high proportion of renewable energy, the local dispatching models of each node in the power grid are often obtained through offline training. Their output decisions are difficult to adapt in real time to the strong fluctuations and uncertainties of renewable energy output, resulting in low reliability of power grid dispatching. Summary of the Invention

[0004] Therefore, it is necessary to provide a power grid dispatching method, apparatus, electronic equipment, computer-readable storage medium, and computer program product that can improve the reliability of power grid dispatching in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a power grid dispatching method, wherein the power grid includes multiple nodes, each node has a corresponding electronic device, and the electronic device corresponding to each node has a dispatching model locally deployed; the method is applied to a first electronic device corresponding to a first node, wherein the first node is any one of the multiple nodes; the method includes:

[0006] The first node is obtained by acquiring its current predicted output data and its first current parameter value corresponding to the target parameter in the first scheduling model. The first scheduling model is a scheduling model deployed locally by the first electronic device. The first current parameter value is obtained by updating the first historical parameter value of the target parameter in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model. The parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node. The second node is a neighbor node of the first node among the plurality of nodes, and the second scheduling model is a scheduling model deployed locally by the second electronic device corresponding to the second node.

[0007] The current predicted output data and the first current parameter value are input into the first scheduling model to obtain the target decision output data of the first node;

[0008] The target decision output data is sent to the first node so that the first node outputs power according to the target decision output data.

[0009] Secondly, this application also provides a power grid dispatching device, wherein the power grid includes multiple nodes, each node has a corresponding electronic device, and the electronic device corresponding to each node has a dispatching model locally deployed; the device includes:

[0010] The data acquisition module is used to acquire the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first scheduling model; the first node is any one of multiple nodes in the power grid, and the first scheduling model is the scheduling model locally deployed by the first electronic device corresponding to the first node; the first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model; the parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node; the second node is a neighboring node of the first node among the multiple nodes, and the second scheduling model is the scheduling model locally deployed by the second electronic device corresponding to the second node;

[0011] The decision calculation module is used to input the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node;

[0012] The decision-making distribution module is used to distribute the target decision output data to the first node, so that the first node outputs power according to the target decision output data.

[0013] Thirdly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0014] The process involves obtaining the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first scheduling model. The first node is any one of multiple nodes in the power grid, each node has a corresponding electronic device, and each electronic device corresponding to the node has a locally deployed scheduling model. The first scheduling model is the scheduling model locally deployed by the first electronic device corresponding to the first node. The first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model. The parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node. The second node is a neighboring node of the first node among the multiple nodes, and the second scheduling model is the scheduling model locally deployed by the second electronic device corresponding to the second node.

[0015] The current predicted output data and the first current parameter value are input into the first scheduling model to obtain the target decision output data of the first node;

[0016] The target decision output data is sent to the first node so that the first node outputs power according to the target decision output data.

[0017] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0018] The process involves obtaining the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first scheduling model. The first node is any one of multiple nodes in the power grid, each node has a corresponding electronic device, and each electronic device corresponding to the node has a locally deployed scheduling model. The first scheduling model is the scheduling model locally deployed by the first electronic device corresponding to the first node. The first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model. The parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node. The second node is a neighboring node of the first node among the multiple nodes, and the second scheduling model is the scheduling model locally deployed by the second electronic device corresponding to the second node.

[0019] The current predicted output data and the first current parameter value are input into the first scheduling model to obtain the target decision output data of the first node;

[0020] The target decision output data is sent to the first node so that the first node outputs power according to the target decision output data.

[0021] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0022] The process involves obtaining the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first scheduling model. The first node is any one of multiple nodes in the power grid, each node has a corresponding electronic device, and each electronic device corresponding to the node has a locally deployed scheduling model. The first scheduling model is the scheduling model locally deployed by the first electronic device corresponding to the first node. The first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model. The parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node. The second node is a neighboring node of the first node among the multiple nodes, and the second scheduling model is the scheduling model locally deployed by the second electronic device corresponding to the second node.

[0023] The current predicted output data and the first current parameter value are input into the first scheduling model to obtain the target decision output data of the first node;

[0024] The target decision output data is sent to the first node so that the first node outputs power according to the target decision output data.

[0025] The aforementioned power grid dispatching method, apparatus, electronic device, computer-readable storage medium, and computer program product are described. The power grid includes multiple nodes, each node having a corresponding electronic device, and each node's corresponding electronic device locally deploys a dispatching model. The method is applied to the first electronic device corresponding to the first node. Firstly, the first electronic device acquires the current predicted output data of the first node, and the first current parameter value of the target parameter corresponding to the first node in the first dispatching model. The first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first dispatching model based on the parameter gradient information of the first dispatching model and the second historical parameter value of the second node corresponding to the target parameter in the second dispatching model. The first node is any one of the multiple nodes in the power grid. In this scenario, each node has a corresponding electronic device, and each electronic device has a locally deployed scheduling model. The first scheduling model is the scheduling model locally deployed on the first electronic device corresponding to the first node. The parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node. The second node is a neighboring node of the first node among multiple nodes, and the second scheduling model is the scheduling model locally deployed on the second electronic device corresponding to the second node. Then, the first electronic device inputs the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node. Next, the first electronic device sends the target decision output data to the first node so that the first node outputs power according to the target decision output data. In this way, each time the first electronic device determines the target decision output data of the first node according to the first scheduling model, it calculates the first current parameter value of the first node corresponding to the target parameter in the first scheduling model. The first current parameter value is updated based on the parameter gradient information of the first scheduling model obtained from the historical actual output data and historical predicted output data of the first node, and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model. On the one hand, obtaining the first current parameter value based on the parameter gradient information of the first scheduling model obtained from the historical actual output data and historical predicted output data of the first node ensures that the first current parameter value can conform to the actual output pattern of the first node itself. On the other hand, obtaining the first current parameter value based on the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model achieves parameter coordination between adjacent nodes. Therefore, it ensures that the decision output data calculated by the first scheduling model can adapt to the strong fluctuation and uncertainty of new energy output in the power grid, thereby improving the reliability of power grid scheduling. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only one embodiment of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a diagram illustrating the application environment of a power grid dispatching method in one embodiment;

[0028] Figure 2 This is a flowchart illustrating a power grid dispatching method in one embodiment;

[0029] Figure 3 This is a flowchart illustrating the steps of updating the parameter value of the first node corresponding to the target parameter in the first scheduling model in one embodiment.

[0030] Figure 4 This is a flowchart illustrating the decision-making layer in one embodiment;

[0031] Figure 5 This is a flowchart illustrating the learning layer in one embodiment;

[0032] Figure 6 This is a structural block diagram of a power grid dispatching device in one embodiment;

[0033] Figure 7 This is a diagram of the internal structure of an electronic device in one embodiment. Detailed Implementation

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0035] It is understood that terms such as "first" and "second" in this application are used only to distinguish similar objects and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. The term "connection" in the embodiments of this application refers to various connection methods, such as direct or indirect connections, to achieve communication between devices; this application does not impose any limitations on this.

[0036] It is understandable that "at least one" refers to one or more, while "multiple" refers to two or more.

[0037] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising / including” or “having,” etc., specify the presence of the stated features, wholes, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0038] Before introducing the specific embodiments of this application, the technical terms involved in this application will be explained:

[0039] Nodes in a power grid: A power grid consists of multiple nodes, each with corresponding electronic equipment. Each node can be any one of the following: a generator node, an energy storage node, a wind power node, a photovoltaic node, or a pure load node. Each node's electronic equipment has a locally deployed scheduling model, which is used to calculate the node's decision-making output data. Each node's scheduling model includes its output cost sub-function, physical constraint penalty sub-function, and global constraint penalty sub-function. Furthermore, the target scheduling functions for different node types (generator nodes, energy storage nodes, wind power nodes, photovoltaic nodes, or pure load nodes) differ, but the sub-functions within the target scheduling function for each node type can all be divided into the node's output cost sub-function, physical constraint penalty sub-function, and global constraint penalty sub-function.

[0040] Node's neighboring nodes: For each node, the neighboring nodes that are adjacent to that node in the power grid are called that node's neighboring nodes.

[0041] Decision moment: refers to the moment when a decision is made regarding the output of a node and the decision output data of the node is calculated.

[0042] Update time: refers to the moment when the parameter value of the node corresponding to the target parameter of the scheduling model is updated.

[0043] Node output data: In this application, node output data includes three categories: predicted output data, decision-making output data, and actual output data. Predicted output data represents the prediction of future output data; decision-making output data refers to the output target within a future period; and actual output data refers to the actual output result. For example, predicted output data serves as the basis for calculating decision-making output data, and actual output data is the result of executing the decision-making output data.

[0044] Current forecast power output data: refers to the most recent forecast power output data.

[0045] Historical predicted power output data refers to historical predicted power output data. In some embodiments, historical predicted power output data includes current predicted power output data, that is, historical predicted power output data includes the most recently predicted power output data and the predicted power output data obtained earlier. In other embodiments, historical predicted power output data does not include the most recently predicted power output data, that is, historical predicted power output data includes predicted power output data prior to the most recently predicted power output data.

[0046] Current actual output data: refers to the most recent actual output data.

[0047] Historical actual output data: refers to historical actual output data. In some embodiments, historical actual output data includes current actual output data, that is, historical actual output data includes the actual output data of the most recent output and the actual output data of earlier outputs. In other embodiments, historical actual output data does not include the actual output data of the most recent output, that is, historical actual output data includes the actual output data before the actual output data of the most recent output.

[0048] The power grid dispatching method provided in this application embodiment can be applied to, for example... Figure 1 The power grid environment shown includes multiple nodes 102, each node 102 having a corresponding electronic device 104. Each electronic device 104 can communicate with its corresponding node 102 via a network. Each electronic device 104 corresponding to a node 102 has a locally deployed scheduling model for calculating the decision-making output data of the corresponding node 102. The electronic devices 104 can communicate with each other via the network. For example, each node 102 can be any one of a generator node, energy storage node, wind power node, photovoltaic node, or pure load node. For example, each electronic device 104 can be any one of a server, a terminal, or a system including servers and terminals. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart vehicle devices, projection devices, etc. Portable wearable devices can be smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can include virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc.

[0049] For example, firstly, the first electronic device acquires the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first scheduling model; the first node is any one of the multiple nodes 102 in the power grid, and the first electronic device is the electronic device 104 corresponding to the first node; the first scheduling model is the scheduling model locally deployed by the first electronic device corresponding to the first node; the first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model; the parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node; the second node is a neighboring node of the first node among the multiple nodes, and the second scheduling model is the scheduling model locally deployed by the second electronic device corresponding to the second node; then, the first node inputs the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node; next, the first node sends the target decision output data to the first node so that the first node outputs power according to the target decision output data.

[0050] In one embodiment, such as Figure 2 As shown, a power grid dispatching method is provided, which can be applied to... Figure 1 The following description will be based on the first electronic device among the various electronic devices 104, wherein the first electronic device is as follows: Figure 1 The electronic device 104 corresponding to the first node in each of the nodes 102 shown, wherein the first node is any one of the multiple nodes 102, the method includes the following steps:

[0051] Step S202: Obtain the current predicted output data of the first node, and the first current parameter value of the first node corresponding to the target parameter in the first scheduling model; the first current parameter value is obtained by updating the first historical parameter value of the first node corresponding to the target parameter in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model; the parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node.

[0052] The first scheduling model is the scheduling model deployed locally on the first electronic device corresponding to the first node; the second scheduling model is the scheduling model deployed locally on the second electronic device corresponding to the second node, where the second node is a neighboring node of the first node.

[0053] Wherein, the first node corresponds to the first current parameter value of the target parameter in the first scheduling model, which means that at the current moment, the first node corresponds to the parameter value of the target parameter in the first scheduling model; the second node corresponds to the first current parameter value of the target parameter in the second scheduling model, which means that at the current moment, the second node corresponds to the parameter value of the target parameter in the second scheduling model.

[0054] Wherein, the first node corresponds to the first historical parameter value of the target parameter in the first scheduling model, which means that the first node corresponds to the parameter value of the target parameter in the first scheduling model before the first current parameter value; the second node corresponds to the second historical parameter value of the target parameter in the second scheduling model, which means that the second node corresponds to the parameter value of the target parameter in the second scheduling model before the second current parameter value.

[0055] Among them, the parameter gradient information of the first scheduling model is used to characterize the parameter value update direction of the target parameter in the first scheduling model.

[0056] The target parameters in the scheduling model may differ for nodes of different node types.

[0057] In this process, when any output decision event is triggered, the first electronic device makes a decision on the output of the first node and determines the current moment as the current decision moment. The output decision event includes at least the time difference between the current moment and the historical decision moment reaching a first preset time difference. For example, the first electronic device makes a decision on the output of the first node 5 minutes after the last decision.

[0058] In this step, when the first electronic device needs to decide the output of the first node, it obtains the current predicted output data of the first node at the current decision time, as well as the first current parameter value of the first node corresponding to the target parameter in the first scheduling model at the current decision time.

[0059] Step S204: Input the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node.

[0060] Among them, the target decision output data of the first node refers to the decision output data to be executed by the first node.

[0061] In this step, the first electronic device inputs the current predicted output data and the first current parameter value corresponding to the first node at the current decision time into the first scheduling model to obtain the target decision output data of the first node. For example, the electronic device substitutes the current predicted output data and the first current parameter value corresponding to the first node into the target scheduling function in the first scheduling model to solve for the decision variables in the target scheduling function, and uses the solution results of the decision variables as the target decision output data of the first node for the target time interval after the current decision time.

[0062] In practical applications, the decision variables for nodes of different node types differ.

[0063] Step S206: The target decision output data is sent to the first node so that the first node outputs power according to the target decision output data.

[0064] In this step, the first electronic device sends the obtained target decision output data to the first node, so that the first node can output power with the target decision output data as the output target within the target time interval after the current decision time.

[0065] In the aforementioned power grid dispatching method, firstly, the first electronic device acquires the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first dispatching model. The first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first dispatching model based on the parameter gradient information of the first dispatching model and the second historical parameter value of the second node corresponding to the target parameter in the second dispatching model. The first node is any one of multiple nodes in the power grid, and each node has a corresponding electronic device, and each electronic device corresponding to the node has a dispatching model deployed locally. The first dispatching model is the dispatching model deployed locally by the first electronic device corresponding to the first node. The parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node. The second node is a neighboring node of the first node among multiple nodes, and the second dispatching model is the dispatching model deployed locally by the second electronic device corresponding to the second node. Then, the first electronic device inputs the current predicted output data and the first current parameter value into the first dispatching model to obtain the target decision output data of the first node. Next, the first electronic device sends the target decision output data to the first node so that the first node outputs power according to the target decision output data. In this way, each time the first electronic device determines the target decision output data of the first node according to the first scheduling model, it calculates it using the first current parameter value of the first node corresponding to the target parameter in the first scheduling model. The first current parameter value is updated based on the parameter gradient information of the first scheduling model obtained from the historical actual output data and historical predicted output data of the first node, and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model. On the one hand, the first current parameter value based on the parameter gradient information of the first scheduling model obtained from the historical actual output data and historical predicted output data of the first node ensures that the first current parameter value can fit the actual output pattern of the first node itself. On the other hand, the first current parameter value based on the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model achieves parameter coordination between adjacent nodes. Therefore, it ensures that the decision output data calculated by the first scheduling model can adapt to the strong fluctuations and uncertainties of new energy output in the power grid, thereby improving the reliability of power grid scheduling.

[0066] In one embodiment, the scheduling model corresponding to different node types is explained:

[0067] For the generator node, the parameters are shown in Tables 1_1 and 1_2:

[0068] Table 1_1

[0069]

[0070] Table 1_2

[0071]

[0072] The economic scheduling subproblem corresponding to the generator node is shown in Equation 1, and its corresponding physical constraints are shown in Equation 2:

[0073] (Formula 1)

[0074] (Formula 2)

[0075] The ADMM (Alternating Direction Method of Multipliers) algorithm is introduced. Based on the economic scheduling subproblem of the generator node as shown in Equation 1 and the physical constraints as shown in Equation 2, the augmented Lagrangian function is obtained as shown in Equation 3. In practical applications, the augmented Lagrangian function shown in Equation 3 is used as the objective scheduling function in the scheduling model corresponding to the generator node. The decision variables of the generator node are solved by finding the minimum value of this augmented Lagrangian function.

[0076] (Formula 3)

[0077] See formula 3 for details. For generator nodes The output cost subfunction; For generator nodes The generator power limit penalty function, For generator nodes The generator ramp penalty function, For generator nodes The load shedding penalty function, For generator nodes Branch capacity penalty function; generator node The physical constraint penalty functions include the generator power limit penalty function, the generator ramp penalty function, the load shedding penalty function, and the branch capacity penalty function; For generator nodes Global constraint penalty subfunction based on ADMM.

[0078] In practical applications, for generator nodes In this regard, the target parameter is The target variables include Current projected power output data includes Decision variables (target decision output data) include .

[0079] For energy storage nodes, the parameters are shown in Tables 2_1 and 2_2:

[0080] Table 2_1

[0081]

[0082] Table 2_2

[0083]

[0084] The economic scheduling subproblem corresponding to the energy storage node is shown in Equation 4, and its corresponding physical constraints are shown in Equation 5:

[0085] (Formula 4)

[0086] (Formula 5)

[0087] The ADMM (Alternating Direction Method of Multipliers) algorithm is introduced. Based on the economic scheduling subproblem of the energy storage node as shown in Equation 4 and the physical constraints as shown in Equation 5, the augmented Lagrangian function is obtained as shown in Equation 6. In practical applications, the augmented Lagrangian function shown in Equation 6 is used as the target scheduling function in the scheduling model corresponding to the energy storage node. The decision variables of the energy storage node are solved by finding the minimum value of this augmented Lagrangian function.

[0088] (Formula 6)

[0089] See formula 6 for details. For energy storage nodes The output cost subfunction; For energy storage nodes The load shedding penalty function, For energy storage nodes The energy storage capacity limit penalty function, For energy storage nodes Branch capacity penalty function; energy storage node The physical constraint penalty functions include the load shedding penalty function, the energy storage capacity limit penalty function, and the branch capacity penalty function; For energy storage nodes Global constraint penalty subfunction based on ADMM.

[0090] In practical applications, for energy storage nodes In this regard, the target parameter is The target variables include Current projected power output data includes Decision variables (target decision output data) include .

[0091] For wind power nodes, the parameters are shown in Table 3:

[0092] Table 3

[0093]

[0094] The economic dispatch subproblem corresponding to the wind power node is shown in Equation 7, and its corresponding physical constraints are shown in Equation 8:

[0095] (Formula 7)

[0096] (Formula 8)

[0097] The ADMM (Alternating Direction Method of Multipliers) algorithm is introduced. Based on the economic scheduling subproblem of wind farm nodes as shown in Equation 7 and the physical constraints as shown in Equation 8, the augmented Lagrangian function is obtained as shown in Equation 9. In practical applications, the augmented Lagrangian function shown in Equation 9 is used as the target scheduling function in the scheduling model corresponding to the wind farm nodes. The decision variables of the wind farm nodes are solved by finding the minimum value of this augmented Lagrangian function.

[0098] (Formula 9)

[0099] See formula 9 for details. For wind power nodes The wind curtailment penalty function, For wind power nodes Load shedding penalty function; wind power node The physical constraint penalty sub-functions include the wind curtailment penalty function and the load shedding penalty function, wind power node The physical constraint penalty subfunction is used as the output cost subfunction; For wind power nodes Global constraint penalty subfunction based on ADMM.

[0100] In practical applications, for wind power nodes In this regard, the target parameter is The target variables include Current projected power output data includes Decision variables (target decision output data) include .

[0101] For photovoltaic nodes, the parameters are shown in Table 4:

[0102] Table 4

[0103]

[0104] The economic scheduling subproblem corresponding to the photovoltaic node is shown in Equation 10, and its corresponding physical constraints are shown in Equation 11:

[0105] (Formula 10)

[0106] (Formula 11)

[0107] The ADMM (Alternating Direction Method of Multipliers) algorithm is introduced. Based on the economic scheduling subproblem of photovoltaic nodes as shown in Equation 10 and the physical constraints as shown in Equation 11, the augmented Lagrangian function is obtained as shown in Equation 12. In practical applications, the augmented Lagrangian function shown in Equation 12 is used as the target scheduling function in the scheduling model corresponding to the photovoltaic nodes. The decision variables of the photovoltaic nodes are solved by finding the minimum value of this augmented Lagrangian function.

[0108] (Formula 12)

[0109] See formula 9 for details. For photovoltaic nodes The light-discarding penalty function, For photovoltaic nodes The load shedding penalty function, For photovoltaic nodes Branch capacity penalty function, photovoltaic node The physical constraint penalty subfunction is used as the output cost subfunction; photovoltaic node The physical constraint penalty functions include the wind curtailment penalty function, the load shedding penalty function, and the branch capacity penalty function; For photovoltaic nodes Global constraint penalty subfunction based on ADMM.

[0110] In practical applications, for photovoltaic nodes In this regard, the target parameter is The target variables include Current projected power output data includes Decision variables (target decision output data) include .

[0111] For pure load nodes, the parameters are shown in Table 5:

[0112] Table 5

[0113]

[0114] The economic scheduling subproblem corresponding to a pure load node is shown in Equation 13, and its corresponding physical constraints are shown in Equation 14.

[0115] (Formula 13)

[0116] (Formula 14)

[0117] The ADMM (Alternating Direction Method of Multipliers) algorithm is introduced. Based on the economic scheduling subproblem of pure load nodes as shown in Equation 13 and the physical constraints as shown in Equation 14, the augmented Lagrangian function is obtained as shown in Equation 15. In practical applications, the augmented Lagrangian function shown in Equation 15 is used as the target scheduling function in the scheduling model corresponding to the pure load nodes. The decision variables of the pure load nodes are solved by finding the minimum value of this augmented Lagrangian function.

[0118] (Formula 15)

[0119] See formula 9 for details. For pure load machine nodes The load shedding penalty function, For pure load machine nodes Branch capacity penalty function; pure load machine node The physical constraint penalty subfunctions include the load shedding penalty function and the branch capacity penalty function, and the pure load node The physical constraint penalty subfunction is used as the output cost subfunction; For pure load machine nodes Global constraint penalty subfunction based on ADMM.

[0120] In practical applications, for pure load machine nodes In this regard, the target parameter is The target variables include Current projected power output data includes Decision variables (target decision output data) include .

[0121] In one embodiment, the first scheduling model includes a target scheduling function corresponding to the first node. The target scheduling function includes the output cost sub-function, the physical constraint penalty sub-function, and the global constraint penalty sub-function of the first node. The global constraint penalty sub-function includes target variables, which include the consistency variable of the ADMM corresponding to the first node, the scaling dual variable of the ADMM, and the dual variable.

[0122] In this embodiment, step S204 above, which inputs the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node, includes the following steps: initializing the variable value of the first node corresponding to the target variable to obtain the initial variable value of the first node corresponding to the target variable; inputting the current predicted output data, the first current parameter value, and the initial variable value into the first scheduling model to obtain the initial decision output data of the first node; sending the first initial branch power to the second electronic device and receiving the second initial branch power sent by the second electronic device; updating the initial variable value based on the first initial branch power and the second initial branch power to obtain the updated variable value; if the updated variable value meets the first preset condition, using the updated variable value as the target variable value of the first node corresponding to the target variable; and inputting the current predicted output data, the first current parameter value, and the target variable value into the first scheduling model to obtain the target decision output data of the first node.

[0123] Wherein, the first branch power is the branch power flowing from the first node to the second node, and the second branch power is the branch power flowing from the second node to the first node; the initial decision output data of the first node includes the first initial branch power, which is the first initial branch power in the initial decision output data; the second initial branch power is the second branch power in the second initial branch power of the second node calculated by the second electronic device.

[0124] In this embodiment, before invoking the first scheduling model, the first electronic device first initializes some parameters and some variables of the first scheduling model; for example, it initializes consistency variables. Scaling dual variables Dual variables , , , , , Original residuals Dual residuals Slack variable penalty coefficient And set according to the energy storage cycle life curve .

[0125] Then, the first electronic device substitutes the current predicted output data, the first current parameter value, and the information obtained from the initialization into the first scheduling model to calculate the initial decision output data of the first node. Initial decision output data of the first node Including the initial branch power flowing from the first node to the second node Next, the first electronic device will transfer the power of the first initial branch. Sending to the second electronic device, and receiving the second initial branch power from the second electronic device, which flows from the second node to the first node. And cache the power of the second initial branch locally. .

[0126] Next, the first electronic device is based on the first initial branch power. Second initial branch power Update the initial variable value of the target variable corresponding to the first node in the first scheduling model to obtain the updated variable value. For example, update the initial consistency variable corresponding to the first node. Initial scaling dual variables Initial dual variables Obtain the updated consistency variables Scaling dual variables after update Updated dual variables As shown in Formula 16:

[0127] (Formula 16)

[0128] Then, the first electronic device determines whether the updated variable value satisfies the original residual. and dual residuals As shown in Formula 17:

[0129] (Formula 17)

[0130] If the updated variable value satisfies Formula 17, the first electronic device will determine the updated variable value as the target variable value corresponding to the target variable of the first node, and substitute the current predicted output data, the first current parameter value, and the target variable value into the first scheduling model to calculate the target decision output data of the first node. .

[0131] If the updated variable value does not satisfy Formula 17, the first electronic device will use the updated variable value as the new initial variable value, and return the current predicted output data, the first current parameter value, and the information obtained from the above initialization, and substitute them into the first scheduling model to calculate the initial decision output data of the first node. The steps continue until the updated variable value satisfies Formula 17.

[0132] In this embodiment, based on the ADMM framework, an optimal balance is found between "distributed computing" and "convergence performance," which retains the privacy protection and parallel computing advantages of the distributed power grid architecture, while also achieving convergence efficiency and accuracy close to that of centralized algorithms.

[0133] In one embodiment, after receiving the second initial branch power sent by the second electronic device in the above steps, the following step is further included: if receiving the second initial branch power sent by the second electronic device fails, predict the second initial branch power based on the second branch power previously sent by the second electronic device, and obtain the second predicted branch power corresponding to the second node.

[0134] The above steps, based on the first initial branch power and the second initial branch power, update the initial variable values ​​to obtain the updated variable values, including the following steps: based on the first initial branch power and the second predicted branch power, update the initial variable values ​​to obtain the updated variable values.

[0135] The failure to receive the second initial branch power sent by the second electronic device means that the second initial branch power was not received within the first preset time period.

[0136] In this embodiment, if the first electronic device does not receive the second initial branch power sent by the second electronic device within a first preset time period (e.g., 50 milliseconds), it estimates the second initial branch power that should have been received based on the second branch power that the second electronic device has historically sent, and obtains the second predicted branch power corresponding to the second node. Then, the first electronic device updates the initial variable value with the second predicted branch power as the second initial branch power that should have been received, and obtains the updated variable value.

[0137] For example, the first electronic device estimates the second initial branch power it should receive according to Formula 18:

[0138] (Formula 18)

[0139] Specifically, when the first electronic device locally caches the second branch power transmitted by the second electronic device, it caches the second branch power in the order it is received, with the sequence number of the second branch power received later being larger. The sequence number of the second initial branch power that the first electronic device should have received; The power of the second predicted branch corresponding to the second node; This is the extrapolation coefficient.

[0140] This embodiment proposes a fault-tolerance mechanism for communication failures, allowing computation to continue using locally cached data even when nodes are temporarily disconnected. This ensures convergence and a feasible solution even in harsh environments with high communication interruption rates, significantly improving the availability and resilience of power grid dispatching. In practical applications, experiments demonstrate that with a 10% communication interruption rate, incorporating the proposed fault-tolerance mechanism into the ADMM framework reduces the number of iterations required for the objective variable to converge compared to traditional synchronous ADMM and asynchronous ADMM frameworks without fault tolerance. Furthermore, at different communication interruption rates (0%, 5%, 10%, 20%, 30%), the fewer iterations are required when the interruption rate is high, demonstrating a clear advantage.

[0141] In one embodiment, after updating the initial variable value based on the first initial branch power and the second initial branch power to obtain the updated variable value, the method further includes the following steps: if the updated variable value does not meet the first preset condition, the updated variable value is used as the new initial variable value, and the method returns to the step of inputting the current predicted output data, the first current parameter value and the initial variable value into the first scheduling model to obtain the initial decision output data of the first node, until the obtained updated variable value meets the first preset condition.

[0142] In this embodiment, referring to Formula 17, if the updated variable value does not satisfy Formula 17, the first electronic device will use the updated variable value as the new initial variable value, and return the current predicted output data, the first current parameter value, and the information obtained from the above initialization, and substitute them into the first scheduling model to calculate the initial decision output data of the first node. The steps continue until the updated variable value satisfies Formula 17.

[0143] In this embodiment, based on the ADMM framework, an optimal balance is found between "distributed computing" and "convergence performance," which retains the privacy protection and parallel computing advantages of the distributed power grid architecture, while also achieving convergence efficiency and accuracy close to that of centralized algorithms.

[0144] In one embodiment, such as Figure 3 As shown, the power grid dispatching method provided in this application further includes the following step: updating the parameter value of the first node corresponding to the target parameter in the first dispatching model.

[0145] Step S302: Based on the historical actual output data and historical predicted output data of the first node, determine the parameter gradient information of the target parameters in the first scheduling model.

[0146] Step S304: Send the first historical parameter value of the first node corresponding to the target parameter in the first scheduling model to the second electronic device, and receive the second historical parameter value sent by the second electronic device.

[0147] Step S306: Based on the parameter gradient information and the second historical parameter value, update the first historical parameter value to obtain the first updated parameter value corresponding to the first node.

[0148] Step S308: Send the first updated parameter value to the second electronic device, and receive the second updated parameter value sent by the second node.

[0149] Step S310: If the difference between the first updated parameter value and the second updated parameter value satisfies the second preset condition, the first updated parameter value is used as the first current parameter value of the first node corresponding to the target parameter in the first scheduling model.

[0150] Specifically, when any parameter update event is triggered, the first electronic device updates the parameter value of the target parameter corresponding to the first node in the first scheduling model and determines the current time as the current update time. The parameter update event includes at least the following: the time difference between the current time and the historical update time reaches a second preset time difference, the total decision error of each scheduling model meets a third preset condition, and the total output fluctuation of the new energy nodes among multiple nodes meets a fourth preset condition. For example, the first electronic device updates the parameter value corresponding to the first node when the interval between the last update and the last update is 15 minutes; or, the first electronic device updates the parameter value corresponding to the first node when the total decision error (the difference between the decision output data and the actual output data) of each scheduling model exceeds a first threshold (e.g., 15%) for a consecutive preset number of decision cycles (i.e., a consecutive preset number of decisions, such as three consecutive decisions); or, the first electronic device updates the parameter value corresponding to the first node when the total decision error (the difference between the decision output data and the actual output data) of each scheduling model exceeds a first threshold (e.g., 15%) for a consecutive preset number of decision cycles; ... three consecutive If the total output of new energy nodes among multiple nodes fluctuates by more than 15% within a second preset time period (e.g., 5 minutes), the parameter value corresponding to the first node will be updated. In practical applications, if the triggered parameter update event is that the time difference between the current time and the historical update time reaches the second preset time difference, the first electronic device will obtain and update the historical actual output data and historical predicted output data within the first historical time interval. The duration of the first historical time interval is 4 to 6 times the duration of the second preset time difference. If the triggered parameter update event is that the total decision error of each scheduling model meets the third preset condition, or the total output fluctuation of new energy nodes among multiple nodes meets the fourth preset condition, the first electronic device will obtain and update the historical actual output data and historical predicted output data within the second historical time interval. The duration of the second historical time interval is 2 to 3 times the duration of the second preset time difference.

[0151] The second preset condition refers to the parameter value after the first update. Compared with the second updated parameter value Parameter values ​​corresponding to the same target parameter tend to be consistent, for example, the parameter values ​​after the first update. Compared with the second updated parameter value The difference between parameter values ​​corresponding to the same target parameter is less than or equal to a preset difference threshold.

[0152] In this embodiment, the first electronic device updates parameter values ​​using a fixed scene-playback optimization method.

[0153] In step S302, when the first electronic device needs to update the parameter value of the first node corresponding to the target parameter in the first scheduling model, it acquires the historical actual output data and historical predicted output data of the first node before the current update time. Based on the historical actual output data and historical predicted output data of the first node, it calculates the parameter gradient information of the target parameter in the first scheduling model. For example, both the historical actual output data and the historical predicted output data include data from multiple historical time points; for each historical time point, the first electronic device updates the first historical parameter value of the first node corresponding to the target parameter in the first scheduling model. Substituting the historical predicted output data at that historical moment into the first scheduling model, the virtual decision output data of the first node at that historical moment is calculated. ,in, This refers to a historical moment; it is easy to understand that the parameter value of the first node at that historical moment, corresponding to the target parameter in the first scheduling model, is not necessarily the same as the first historical parameter value. Therefore, the virtual decision-making output data of the first node at that historical moment is not necessarily the same as the target decision-making output data of the first node at that historical moment; then, the first electronic device bases its decision on the first historical parameter value. And the output cost sub-function in the first scheduling model, to calculate the historical moment. Virtual decision-making output data Corresponding virtual output cost For example, the first historical parameter value and that historical moment Virtual decision-making output data Substituting into the first scheduling model, and obtaining the first node at that historical moment. Actual output cost Next, the first electronic device is based on this historical moment. Actual output cost and virtual output cost Calculate the first scheduling model at this historical moment The loss function is given, and based on this historical moment. The loss function is used to calculate the target parameters in the first scheduling model at that historical moment. The parameter gradient at each historical moment is used as the average value of the parameter gradients in the first electronic device as the parameter gradient information of the target parameter in the first scheduling model.

[0154] For example, the first electronic device calculates historical moments based on Formula 19. loss function :

[0155] (Formula 19)

[0156] The first electronic device, based on the target parameters in the first scheduling model of Formula 20, at this historical moment... Parameter gradient :

[0157] (Formula 20)

[0158] The first electronic device calculates the parameter gradient information of the target parameter in the first scheduling model based on formula 21. :

[0159] (Formula 21)

[0160] in, The number of historical moments.

[0161] In step S304, the first electronic device stores the first historical parameter value. Sending to the second electronic device, and receiving the second historical parameter value sent by the second electronic device, which corresponds to the target parameter in the second scheduling model for the second node. And cache the second historical parameter value locally. .

[0162] In step S306, the first electronic device, as shown in Formula 22, is based on parameter gradient information. Second historical parameter value Update the first historical parameter value We obtain the first updated parameter value corresponding to the first node:

[0163] (Formula 22)

[0164] in, This is the first updated parameter value corresponding to the first node; The learning rate; This represents the consensus strength coefficient. The weights are double random matrix weights.

[0165] In some embodiments, the first electronic device updates the consensus target parameter in the target parameters based on Formula 22, and updates the non-consensus target parameters based on the parameter gradient information of the non-consensus target parameters in the target parameters. Here, the consensus target parameter is the target parameter requiring consensus, and the non-consensus target parameter is the consensus parameter other than the consensus target parameter. In practical applications, the non-consensus target parameter is the load shedding penalty coefficient. The consensus target parameter is the load shedding penalty coefficient. The parameter values ​​of the other target parameters, and the load shedding penalty coefficient. Updated based on Formula 23:

[0166] (Formula 23)

[0167] in, The load shedding penalty factor The parameter gradient information.

[0168] In step S308, the first electronic device updates the first parameter value. Send to the second electronic device, and receive the second updated parameter value corresponding to the second node sent by the second node. .

[0169] In step S310, the electronic device displays the parameter values ​​after the first update. Compared with the second updated parameter value If the difference between them meets the second preset condition, the first updated parameter value will be... The first node corresponds to the first current parameter value of the target parameter in the first scheduling model. If the first updated parameter value... Compared with the second updated parameter value If the difference between them does not meet the second preset condition, then the second electronic device will update the first parameter value. As the new first historical parameter value The process involves returning historical actual output data and historical predicted output data based on the first node, and calculating the parameter gradient information of the target parameters in the first scheduling model, until the first updated parameter value is obtained. Compared with the second updated parameter value The difference between them satisfies the second preset condition.

[0170] In this embodiment, on the one hand, the parameter values ​​of the first scheduling model are updated based on the historical actual output data and historical predicted output data of the first node, so that the first current parameter value can fit the actual output pattern of the first node itself. On the other hand, the parameter values ​​of the second node are updated based on the second historical parameter value corresponding to the target parameter in the second scheduling model, realizing parameter coordination between adjacent nodes. Therefore, it is ensured that the decision output data calculated by the first scheduling model based on the first current parameter value can adapt to the strong fluctuation and strong uncertainty of the new energy output in the power grid, thereby improving the reliability of power grid scheduling.

[0171] In one embodiment, after receiving the second historical parameter value sent by the second electronic device in step S304 above, the following step is further included: if there is a communication delay in the second historical parameter value sent by the second electronic device, the second historical parameter value is compensated based on the second parameter value sent by the second electronic device in the past to obtain the second compensated parameter value corresponding to the second node.

[0172] Step S306 above, which updates the first historical parameter value based on parameter gradient information and the second historical parameter value to obtain the first updated parameter value corresponding to the first node, includes the following steps: updating the first historical parameter value based on parameter gradient information and the second compensation parameter value to obtain the first updated parameter value.

[0173] The second historical parameter value sent by the second electronic device carries a sending timestamp. After receiving the second historical parameter value sent by the second electronic device, the first electronic device compares the current time (the time when the second historical parameter value is received) with the sending timestamp carried by the second historical parameter value. When the time difference between the two reaches a third preset time difference, the first electronic device determines that there is a communication delay in the second historical parameter value sent by the second electronic device.

[0174] In this embodiment, when there is a communication delay in the second historical parameter value sent by the second electronic device, the first electronic device compensates for the second historical parameter value based on the locally cached second historical parameter value sent by the second electronic device to obtain the second compensated branch power corresponding to the second node; then, the first electronic device uses the second compensated branch power as the second historical parameter value to participate in updating the first historical parameter value, and updates the first historical parameter value to obtain the first updated parameter value.

[0175] For example, the first electronic device compensates for the second historical parameter value according to formula 24:

[0176] (Formula 24)

[0177] When the first electronic device locally caches the second parameter value sent by the second electronic device, it caches the second parameter values ​​in the order they are received, with later received second parameter values ​​having larger sequence numbers. The sequence number in which the first electronic device receives the second historical parameter value with a delay; For delay step size; This is the compensation coefficient.

[0178] In this embodiment, a fault-tolerance mechanism under communication delay is proposed, which allows the computation to continue using local cached data when there is communication delay at the node. This ensures that parameter updates can still converge even under poor communication conditions, thereby optimizing the parameter update mechanism and ensuring the reliability of parameter updates.

[0179] In one embodiment, after receiving the second updated parameter value sent by the second node in step S308 above, the method further includes the following steps: if the difference between the first updated parameter value and the second updated parameter value does not meet the second preset condition, the first updated parameter value is used as the new first historical parameter value, and the parameter gradient information of the target parameter in the first scheduling model is determined based on the historical actual output data and historical predicted output data of the first node, until the difference between the obtained first updated parameter value and the second updated parameter value meets the second preset condition.

[0180] The second preset condition refers to the parameter value after the first update. Compared with the second updated parameter value Parameter values ​​corresponding to the same target parameter tend to be consistent, for example, the parameter values ​​after the first update. Compared with the second updated parameter value The difference between parameter values ​​corresponding to the same target parameter is less than or equal to a preset difference threshold.

[0181] In this embodiment, if the first updated parameter value Compared with the second updated parameter value If the difference between them does not meet the second preset condition, then the second electronic device will update the first parameter value. As the new first historical parameter value The process involves returning historical actual output data and historical predicted output data based on the first node, and calculating the parameter gradient information of the target parameters in the first scheduling model, until the first updated parameter value is obtained. Compared with the second updated parameter value The difference between them satisfies the second preset condition.

[0182] In this embodiment, based on the ADMM framework, an optimal balance is found between "distributed computing" and "convergence performance," which retains the privacy protection and parallel computing advantages of the distributed power grid architecture, while also achieving convergence efficiency and accuracy close to that of centralized algorithms.

[0183] In one embodiment, the power grid dispatching method provided by this application further includes the following steps: when any output decision event is triggered, determining the target decision output data of the first node based on the first dispatching model.

[0184] The power grid dispatching method provided in this application further includes the following steps: when any parameter update event is triggered, update the first historical parameter value of the first node corresponding to the target parameter in the first dispatching model to obtain the first current parameter value of the first node corresponding to the target parameter in the first dispatching model.

[0185] Among them, the power output decision event includes at least the time difference between the current moment and the historical decision moment reaching a first preset time difference; for example, the first electronic device makes a power output decision on the first node when the interval between the last historical decision moment and 5 minutes has passed.

[0186] The parameter update event includes at least the following: the time difference between the current time and the historical update time reaches a second preset time difference, the total decision error of each scheduling model meets a third preset condition, and the total output fluctuation of the new energy nodes among multiple nodes meets a fourth preset condition. For example, the first electronic device updates the parameter value corresponding to the first node 15 minutes after the last update, or the first electronic device updates the parameter value corresponding to the first node when the total decision error (the difference between the decision output data and the actual output data) of each scheduling model exceeds a first threshold (e.g., 15%) for a consecutive preset number of decision cycles (i.e., a consecutive preset number of decisions, such as three consecutive decisions). Alternatively, the first electronic device updates the parameter value corresponding to the first node when the minute-level fluctuation of the total output of the new energy nodes among multiple nodes exceeds the second threshold (15%) within a second preset time period (e.g., 5 minutes).

[0187] Wherein, the first preset time difference is less than the second preset time difference; for example, the first preset time is 5 minutes and the second preset time difference is 15 minutes.

[0188] In this embodiment, when the first electronic device detects that any parameter update event is triggered, it makes a decision on the output of the first node by executing steps S202 to S206. When the first electronic device detects that any parameter update event is triggered, it updates the parameter value of the first node corresponding to the target parameter in the first scheduling model by executing steps S302 to S310.

[0189] In this embodiment, by setting parameter update events, the output decision and parameter update of the first node can be triggered on time and as needed.

[0190] In another specific embodiment, this application also provides a "decision-learning" dual-closed-loop decoupled distributed economic dispatch method for a renewable energy power grid. In this method, the renewable energy power grid includes generator nodes, energy storage nodes, wind power nodes, photovoltaic nodes, and pure load nodes. The output dispatch of the renewable energy power grid adopts distributed dispatch based on the ADMM framework. For example, each node deploys a corresponding agent, and each agent locally deploys a corresponding dispatch model. The dispatch models for generator nodes are shown in Tables 1_1 and 1_2, and Equations 1 to 3; the dispatch models for energy storage nodes are shown in Tables 2_1 and 2_2, and Equations 4 to 5; the dispatch models for wind power nodes are shown in Table 3, and Equations 7 to 9; the dispatch models for photovoltaic nodes are shown in Table 4, and Equations 10 to 12; and the dispatch models for pure load nodes are shown in Table 5, and Equations 13 to 15. For example, each node's agent includes two layers: a decision layer and a learning layer. The decision layer is used to call the locally deployed scheduling model based on the latest model parameters to calculate the decision output data of the node when making a scheduling decision. The learning layer is used to update the model parameters when the parameters are updated.

[0191] like Figure 4The diagram illustrates the decision-making process. For each agent, upon triggering any parameter update event, the agent obtains the current predicted output data of its node and the current parameter value of the target parameter in the locally deployed scheduling model corresponding to that node. The agent initializes the variable value of the target variable in the corresponding locally deployed scheduling model for that node, obtaining the initial variable value for that node's target variable. Based on the node's current predicted output data, current parameter value, and initial variable value, the agent calls the locally deployed scheduling model to obtain the node's initial decision output data. The agent then submits the node's initial decision... In the output data, the initial branch power flowing from this node to neighboring nodes is sent to the agent of the neighboring node, and the initial branch power flowing from the neighboring node to this node is received from the agent of the neighboring node. If the initial branch power received from the agent of the neighboring node is successful, the agent updates the initial variable value of this node based on the initial branch power of this node and the initial branch power of the neighboring node, thus obtaining the updated variable value of this node. For example, the process of the agent updating the initial variable value of this node is shown in Formula 16. If the initial branch power received from the agent of the neighboring node fails, the agent updates the initial variable value of this node based on the initial branch power of the neighboring node. The proxy estimates the initial branch power of neighboring nodes based on the historical branch power of the proxy, obtaining the estimated branch power of neighboring nodes. Based on the initial branch power of the current node and the estimated branch power of neighboring nodes, the proxy updates the initial variable value of the current node, obtaining the updated variable value of the current node. For example, the process of the proxy estimating the initial branch power of neighboring nodes is shown in Formula 18. If the updated variable value of the current node meets a first preset condition, the proxy uses the updated variable value of the current node as the target variable value corresponding to the target variable. If the updated variable value of the current node does not meet the first preset condition, the proxy... The updated variable value of a node is used as the new initial variable value for this node. The process of calling the locally deployed scheduling model based on the current predicted output data, current parameter value, and initial variable value of this node to obtain the initial decision output data of this node is repeated. For example, the first preset condition is shown in Formula 17. The agent calls the locally deployed scheduling model based on the current predicted output data, the first current parameter value, and the target variable value of this node to obtain the target decision output data of this node. The agent sends the target decision output data of this node to this node so that this node outputs power according to the corresponding target decision output data.

[0192] like Figure 5The diagram illustrates the learning layer's workflow. For each agent, upon triggering any parameter update event, the agent, based on its historical actual output data, historical predicted output data, and historical parameter values ​​corresponding to the target parameter in the first scheduling model, invokes the locally deployed scheduling model to determine the virtual decision output data for that node. The agent then determines the historical actual output cost corresponding to the node's historical actual output data, and the virtual output cost corresponding to the node's virtual decision output data. Based on these historical actual output costs and virtual output costs, the agent determines the target parameter in the locally deployed scheduling model. For example, the process of the agent calculating the parameter gradient information of this node is shown in Equations 19 to 21; the agent sends the historical parameter values ​​of this node to the agents of its neighboring nodes, and receives the historical parameter values ​​of the neighboring nodes sent by the agents of the neighboring nodes; in the absence of communication delay with neighboring nodes, the agent updates the historical parameter values ​​of this node based on the parameter gradient information of this node and the historical parameter values ​​of the neighboring nodes, and obtains the first updated parameter value of this node. For example, the process of the agent calculating the parameter gradient information of this node is shown in Equations 19 to 21; in the absence of communication delay with neighboring nodes, the agent updates the historical parameter values ​​of this node based on the parameter gradient information of this node and the historical parameter values ​​of the neighboring nodes, and obtains the first updated parameter value of this node. In the event of communication delay between nodes, the agent compensates for the historical parameter values ​​of neighboring nodes based on the historical parameter values ​​sent by the agent of neighboring nodes, obtaining the compensated parameter values ​​of neighboring nodes. Based on the parameter gradient information of the current node and the compensated parameter values ​​of neighboring nodes, the agent updates the historical parameter values ​​of the current node, obtaining the updated parameter values ​​of the current node. For example, the process of the agent compensating for the historical parameter values ​​of neighboring nodes is shown in Formula 24. The agent sends the updated parameter values ​​of the current node to the agent of the neighboring node and receives the updated parameter values ​​of the neighboring node sent by the agent of the neighboring node. If the difference between the updated parameter values ​​of the current node and the updated parameter values ​​of neighboring nodes meets the second preset condition, the updated parameter of the current node is used as the current parameter value of the target parameter of the current node corresponding to the first scheduling model. If the difference between the updated parameter values ​​of the current node and the updated parameter values ​​of neighboring nodes does not meet the second preset condition, the updated parameter of the current node is used as the new historical parameter value of the current node. The steps of calling the locally deployed scheduling model based on the historical actual output data, historical predicted output data, and the historical parameter values ​​of the target parameter of the current node corresponding to the first scheduling model to determine the virtual decision output data of the current node are repeated.

[0193] The method proposed in this embodiment has the following significant advantages:

[0194] First, it achieves a balance between the accuracy of the scheduling objective function parameters and the speed of decision-making: through the learning layer, the scheduling model can be continuously calibrated using the latest operational data, enabling the decision layer to always optimize rapidly based on the current "most accurate" understanding, fundamentally solving the long-term deviation problem caused by the fixed model, and realizing a leap from "static optimization" to "dynamic adaptive optimization".

[0195] Second, it possesses excellent communication robustness: the asynchronous fault-tolerant alternating direction multiplier method in the decision layer and the asynchronous collaborative learning mechanism in the outer layer both allow nodes to continue computation using local historical information even when they are temporarily disconnected. This ensures that even in harsh environments with a high communication interruption rate (such as a 10% interruption rate), it can still converge and provide a feasible solution, greatly improving availability and resilience.

[0196] Third, it efficiently handles complex physical constraints: by embedding complex physical constraints into the distributed optimization framework in the form of continuously differentiable penalty functions, and combining collaborative learning to dynamically adjust the coefficients of the penalty functions, it effectively handles multi-constraint optimization engineering problems in an online and fully distributed manner, and the resulting scheduling scheme is closer to physical reality.

[0197] Fourth, the scalability and privacy brought by the fully distributed architecture: no central controller is required, all nodes are equal in status, and only a small amount of coordination information is exchanged with neighbors to complete scheduling online. This perfectly fits the development trend of "distributed resources as the main body" in the new power system. This method supports "plug and play" of resources and strictly protects the commercial data privacy of each market player.

[0198] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0199] Based on the same inventive concept, this application also provides a power grid dispatching device for implementing the power grid dispatching method described above. The solution provided by this device is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more power grid dispatching device embodiments provided below can be found in the limitations of the power grid dispatching method described above, and will not be repeated here.

[0200] In one embodiment, such as Figure 6 As shown, a power grid dispatching device is provided. The power grid includes multiple nodes, each node has corresponding electronic equipment, and the electronic equipment corresponding to each node has a dispatching model locally deployed. The device includes a data acquisition module 602, a decision calculation module 604, and a decision distribution module 606, wherein:

[0201] The data acquisition module 602 is used to acquire the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first scheduling model; the first node is any one of the multiple nodes of the power grid, and the first scheduling model is the scheduling model of the first electronic device corresponding to the first node deployed locally; the first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model; the parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node; the second node is a neighboring node of the first node among the multiple nodes, and the second scheduling model is the scheduling model of the second electronic device corresponding to the second node deployed locally.

[0202] The decision calculation module 604 is used to input the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node.

[0203] The decision distribution module 606 is used to distribute the target decision output data to the first node so that the first node outputs power according to the target decision output data.

[0204] In one embodiment, the first scheduling model includes a target scheduling function corresponding to the first node. The target scheduling function includes the output cost sub-function of the first node, the physical constraint penalty sub-function, and the global constraint penalty sub-function; the global constraint penalty sub-function includes the target variable.

[0205] The decision calculation module 604 is further configured to initialize the variable values ​​of the first node corresponding to the target variable, thereby obtaining the initial variable values ​​of the first node corresponding to the target variable; input the current predicted output data, the first current parameter value, and the initial variable values ​​into the first scheduling model to obtain the initial decision output data of the first node; the initial decision output data includes the first initial branch power, which is the branch power flowing from the first node to the second node; send the first initial branch power to the second electronic device, and receive the second initial branch power sent by the second electronic device; the second initial branch power is the branch power flowing from the second node to the first node; update the initial variable values ​​based on the first initial branch power and the second initial branch power to obtain the updated variable values; if the updated variable values ​​meet the first preset condition, use the updated variable values ​​as the target variable values ​​of the first node corresponding to the target variable; input the current predicted output data, the first current parameter value, and the target variable values ​​into the first scheduling model to obtain the target decision output data of the first node.

[0206] In one embodiment, the decision calculation module 604 is further configured to, in the event that the second initial branch power sent by the second electronic device fails to be received, predict the second initial branch power based on the second branch power previously sent by the second electronic device, and obtain the second predicted branch power corresponding to the second node.

[0207] The decision calculation module 604 is also used to update the initial variable values ​​based on the first initial branch power and the second predicted branch power, so as to obtain the updated variable values.

[0208] In one embodiment, the decision calculation module 604 is further configured to, when the updated variable value does not meet the first preset condition, use the updated variable value as the new initial variable value, and return the step of inputting the current predicted output data, the first current parameter value and the initial variable value into the first scheduling model to obtain the initial decision output data of the first node, until the obtained updated variable value meets the first preset condition.

[0209] In one embodiment, the scheduling model device further includes a parameter update module, configured to: determine parameter gradient information of the target parameter in the first scheduling model based on the historical actual output data and historical predicted output data of the first node; send a first historical parameter value of the first node corresponding to the target parameter in the first scheduling model to a second electronic device, and receive a second historical parameter value sent by the second electronic device; update the first historical parameter value based on the parameter gradient information and the second historical parameter value to obtain a first updated parameter value corresponding to the first node; send the first updated parameter value to the second electronic device, and receive a second updated parameter value sent by the second node; and, if the difference between the first updated parameter value and the second updated parameter value satisfies a second preset condition, use the first updated parameter value as the first current parameter value of the first node corresponding to the target parameter in the first scheduling model.

[0210] In one embodiment, the parameter update module is further configured to, when there is a communication delay in the second historical parameter value sent by the second electronic device, compensate the second historical parameter value based on the second historical parameter value sent by the second electronic device to obtain the second compensated parameter value corresponding to the second node.

[0211] The parameter update module is also used to update the first historical parameter value based on the parameter gradient information and the second compensation parameter value, so as to obtain the first updated parameter value.

[0212] In one embodiment, the parameter update module is further configured to, when the difference between the first updated parameter value and the second updated parameter value does not meet the second preset condition, take the first updated parameter value as the new first historical parameter value, return the parameter gradient information of the target parameter in the first scheduling model based on the historical actual output data and historical predicted output data of the first node, until the difference between the obtained first updated parameter value and the second updated parameter value meets the second preset condition.

[0213] In one embodiment, the power grid dispatching device further includes a decision triggering module, which is used to determine the target decision output data of the first node based on the first dispatching model when any output decision event is triggered; the output decision event includes at least the time difference between the current time and the historical decision time reaching a first preset time difference.

[0214] The power grid dispatching device also includes an update triggering module, which is used to update the first historical parameter value of the target parameter corresponding to the first node in the first dispatching model when any parameter update event is triggered, so as to obtain the first current parameter value of the target parameter corresponding to the first node in the first dispatching model. The parameter update event includes at least the time difference between the current time and the historical update time reaching a second preset time difference, the total decision error of each dispatching model satisfying a third preset condition, and the total output fluctuation of the new energy nodes among multiple nodes satisfying a fourth preset condition.

[0215] Each module in the aforementioned power grid dispatching device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor within the electronic device, or stored in the memory of the electronic device as software, so that the processor can call and execute the corresponding operations of each module.

[0216] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows. Figure 7 As shown, this electronic device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores power grid data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a power grid dispatching method.

[0217] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0218] In one embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0219] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above-described method embodiments.

[0220] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0221] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0222] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0223] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A power grid dispatching method, characterized in that, The power grid includes multiple nodes, each node having a corresponding electronic device, and each electronic device corresponding to a node has a locally deployed scheduling model; the method is applied to a first electronic device corresponding to a first node, where the first node is any one of the multiple nodes; the method includes: The first node is obtained by acquiring its current predicted output data and its first current parameter value corresponding to the target parameter in the first scheduling model. The first scheduling model is a scheduling model deployed locally by the first electronic device. The first current parameter value is obtained by updating the first historical parameter value of the target parameter in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model. The parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node. The second node is a neighbor node of the first node among the plurality of nodes, and the second scheduling model is a scheduling model deployed locally by the second electronic device corresponding to the second node. The current predicted output data and the first current parameter value are input into the first scheduling model to obtain the target decision output data of the first node; The target decision output data is sent to the first node so that the first node outputs power according to the target decision output data.

2. The method according to claim 1, characterized in that, The first scheduling model includes a target scheduling function corresponding to the first node. The target scheduling function includes a power output cost sub-function, a physical constraint penalty sub-function, and a global constraint penalty sub-function for the first node. The global constraint penalty sub-function includes a target variable. The step of inputting the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node includes: The variable values ​​of the first node corresponding to the target variable are initialized to obtain the initial variable values ​​of the first node corresponding to the target variable; The current predicted output data, the first current parameter value, and the initial variable value are input into the first scheduling model to obtain the initial decision output data of the first node; the initial decision output data includes the first initial branch power, which is the branch power flowing from the first node to the second node; The first initial branch power is sent to the second electronic device, and the second initial branch power sent by the second electronic device is received; the second initial branch power is the branch power flowing from the second node to the first node; Based on the first initial branch power and the second initial branch power, the initial variable value is updated to obtain the updated variable value; If the updated variable value meets the first preset condition, the updated variable value is used as the target variable value of the first node corresponding to the target variable. The current predicted output data, the first current parameter value, and the target variable value are input into the first scheduling model to obtain the target decision output data of the first node.

3. The method according to claim 2, characterized in that, After receiving the second initial branch power sent by the second electronic device, the method further includes: In the event that the second initial branch power transmitted by the second electronic device fails to be received, the second initial branch power is predicted based on the second branch power transmitted by the second electronic device in the past, and the second predicted branch power corresponding to the second node is obtained. The step of updating the initial variable values ​​based on the first initial branch power and the second initial branch power to obtain updated variable values ​​includes: Based on the first initial branch power and the second predicted branch power, the initial variable value is updated to obtain the updated variable value.

4. The method according to claim 2, characterized in that, After updating the initial variable value based on the first initial branch power and the second initial branch power to obtain the updated variable value, the process further includes: If the updated variable value does not meet the first preset condition, the updated variable value is used as the new initial variable value, and the process of inputting the current predicted output data, the first current parameter value, and the initial variable value into the first scheduling model to obtain the initial decision output data of the first node is repeated until the obtained updated variable value meets the first preset condition.

5. The method according to claim 1, characterized in that, The method further includes: Based on the historical actual output data and historical predicted output data of the first node, the parameter gradient information of the target parameter in the first scheduling model is determined. Send the first historical parameter value of the first node corresponding to the target parameter in the first scheduling model to the second electronic device, and receive the second historical parameter value sent by the second electronic device; Based on the parameter gradient information and the second historical parameter value, the first historical parameter value is updated to obtain the first updated parameter value corresponding to the first node; Send the first updated parameter value to the second electronic device, and receive the second updated parameter value sent by the second node; If the difference between the first updated parameter value and the second updated parameter value satisfies the second preset condition, the first updated parameter value is used as the first current parameter value of the first node corresponding to the target parameter in the first scheduling model.

6. The method according to claim 5, characterized in that, After receiving the second historical parameter value sent by the second electronic device, the method further includes: If there is a communication delay in the second historical parameter value sent by the second electronic device, the second historical parameter value is compensated based on the second parameter value sent by the second electronic device in the past to obtain the second compensated parameter value corresponding to the second node. The step of updating the first historical parameter value based on the parameter gradient information and the second historical parameter value to obtain the first updated parameter value corresponding to the first node includes: Based on the parameter gradient information and the second compensation parameter value, the first historical parameter value is updated to obtain the first updated parameter value.

7. The method according to claim 5, characterized in that, After receiving the second updated parameter value sent by the second node, the method further includes: If the difference between the first updated parameter value and the second updated parameter value does not meet the second preset condition, the first updated parameter value is used as the new first historical parameter value, and the parameter gradient information of the target parameter in the first scheduling model is determined based on the historical actual output data and historical predicted output data of the first node, until the difference between the obtained first updated parameter value and the second updated parameter value meets the second preset condition.

8. The method according to any one of claims 1 to 7, characterized in that, The method further includes: When any output decision event is triggered, the target decision output data of the first node is determined based on the first scheduling model; the output decision event includes at least the time difference between the current time and the historical decision time reaching a first preset time difference. The method further includes: When any parameter update event is triggered, the first historical parameter value of the first node corresponding to the target parameter in the first scheduling model is updated to obtain the first current parameter value of the first node corresponding to the target parameter in the first scheduling model; the parameter update event includes at least the time difference between the current time and the historical update time reaching a second preset time difference, the total decision error of each scheduling model satisfying a third preset condition, and the total output fluctuation of the new energy nodes among the plurality of nodes satisfying a fourth preset condition.

9. A power grid dispatching device, characterized in that, The power grid includes multiple nodes, each node having corresponding electronic equipment, and each electronic equipment at each node has a locally deployed scheduling model; the device includes: The data acquisition module is used to acquire the current predicted output data of the first node and the first current parameter value of the target parameter corresponding to the first node in the first scheduling model; the first node is any one of multiple nodes in the power grid, and the first scheduling model is the scheduling model locally deployed by the first electronic device corresponding to the first node; the first current parameter value is obtained by updating the first historical parameter value of the target parameter corresponding to the first node in the first scheduling model based on the parameter gradient information of the first scheduling model and the second historical parameter value of the second node corresponding to the target parameter in the second scheduling model; the parameter gradient information is obtained based on the historical actual output data and historical predicted output data of the first node; the second node is a neighboring node of the first node among the multiple nodes, and the second scheduling model is the scheduling model locally deployed by the second electronic device corresponding to the second node; The decision calculation module is used to input the current predicted output data and the first current parameter value into the first scheduling model to obtain the target decision output data of the first node; The decision-making distribution module is used to distribute the target decision output data to the first node, so that the first node outputs power according to the target decision output data.

10. An electronic device comprising a memory and a processor, wherein the 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 8.

Images