CN111105100A

CN111105100A - Neural network-based optimization method and system for multi-microgrid scheduling mechanism

Info

Publication number: CN111105100A
Application number: CN202010024491.9A
Authority: CN
Inventors: 余欣蓉; 邱革非; 金乐婷
Original assignee: Kunming University of Science and Technology
Current assignee: Kunming University of Science and Technology
Priority date: 2020-01-10
Filing date: 2020-01-10
Publication date: 2020-05-05
Anticipated expiration: 2040-01-10
Also published as: CN111105100B

Abstract

The invention discloses a neural network-based optimization method and system for a multi-microgrid scheduling mechanism, wherein the method comprises the following steps: constructing a first neural network, and mapping the neural network of the first neural network as an output scheduling function of a multi-microgrid scheduling mechanism; representing the energy type and the output power of the micro-grid by using a first input node, and representing an output scheduling result by using a first output node; configuring an optimization objective function according to the optimization objective; generating a micro-grid output training sample by using the known energy type and the known output power; according to the optimization objective function, training a first neural network by using a microgrid output training sample; and assigning the energy type and the output power of each micro-grid to a first input node of a first neural network, and calculating by using the first neural network to obtain an output scheduling result of each micro-grid. The technical scheme of the invention can solve the problem that a new scheduling mechanism is difficult to be optimized quickly when the power grid application scene is changed in the prior art.

Description

Neural network-based optimization method and system for multi-microgrid scheduling mechanism

Technical Field

The invention relates to a neural network-based optimization method and system for a multi-microgrid scheduling mechanism, and belongs to the technical field of microgrids.

Background

The mechanism optimization is a branch subject of a game theory and is combined with other theories to generate a plurality of branch subjects, wherein the multi-microgrid scheduling mechanism optimization is generally expressed as scheme optimization for scheduling each microgrid in real power grid production activities determined by adopting a class engineering method; generally, the optimization of the multi-microgrid scheduling mechanism aims to find an optimal output scheduling rule, so that all microgrid rationally pursue an individual benefit maximization behavior result under the rule, and the goal expected to be achieved by some mechanism optimizers can be achieved; for example: solution strategy balancing, individuality, incentive compatibility, budget balancing, allocation efficiency, profit maximization, social optimality, fairness, and/or collusion resistance.

Mechanism optimization in the electric power system is widely applied, and especially today that the power grid technology is gradually popularized, a large number of micro-grids can be connected into a power distribution network through various mechanisms, and certain benefits can be obtained while electric energy is supplemented. The micro-grid is a novel network structure and comprises a distributed power supply, a load, an energy storage system and the like. The micro-grid can be connected with an external power grid in a grid mode and can also be operated in an isolated mode. For the case of multiple micro-grids, optimization of the scheduling mechanism for multiple micro-grids is generally required.

Traditionally, the multi-microgrid scheduling mechanism is usually manually optimized by professional engineers, so that the optimization of the multi-microgrid scheduling mechanism requires hard game theory professional knowledge and acute electrical intuition, and is a complex and difficult task. However, each time an application scenario is newly added, the grid engineer needs to adjust the original incentive scheme and optimize the new incentive scheme suitable for the new application scenario. Due to the contradiction, the new excitation mechanism after the micro-grid application scene is changed is difficult to be optimized quickly.

Disclosure of Invention

The invention aims to solve the technical problem of providing a method and a system for optimizing a multi-microgrid scheduling mechanism based on a neural network, and aims to solve the problems that the multi-microgrid scheduling mechanism is complex and difficult to optimize in the prior art, and a new incentive mechanism is difficult to optimize quickly when an application scene changes.

To achieve the above object, according to a first aspect of the present invention, the present invention provides a method for optimizing a multi-microgrid scheduling mechanism based on a neural network, including:

constructing a first neural network, and mapping the neural network of the first neural network as an output scheduling function of a multi-microgrid scheduling mechanism, wherein the first neural network comprises a first input node and a first output node, and the neural network is mapped to a calculation relation from the first input node to the first output node;

representing the energy type and the output power of the micro-grid by using the first input node, wherein the first output node represents an output scheduling result calculated according to the output scheduling function;

configuring an optimization objective function of the first neural network according to an optimization objective of the multi-microgrid scheduling mechanism, wherein the optimization objective function is dependent on the first output node;

generating a microgrid output training sample for the first neural network training using a priori distributions of known energy types and known output powers in the first input node;

training the first neural network by using the microgrid output training sample according to the optimization objective function;

and when the training of the first neural network is finished, assigning the energy type and the output power of each micro-grid to a first input node of the first neural network, and calculating by using the first neural network to obtain an output scheduling result of each micro-grid.

Preferably, the first neural network is configured to compute a portion of a graph; the optimized target value of the objective function corresponds to a first computational graph node of the computational graph; the step of configuring an optimization objective function of the first neural network according to an optimization objective of a multi-microgrid scheduling mechanism includes:

configuring the first computational graph node and one or more second computational graph nodes on which the first computational graph node depends according to an optimization objective of the multi-microgrid scheduling mechanism; and

configuring a computational relationship between the first computational graph node and the one or more second computational graph nodes;

and the second calculation graph nodes comprise nodes corresponding to the sub-optimization targets and target parameter nodes.

Preferably, the optimization objective function includes a cost optimization objective function, and the step of configuring the optimization objective function of the first neural network according to the optimization objective of the multi-microgrid scheduling mechanism includes:

configuring the first computational graph node according to a comprehensive cost optimization objective of the cost optimization objective function;

configuring one or more second computing nodes depended by the first computing graph nodes according to the sub-cost optimization objectives of the cost optimization objective function and the cost optimization weight parameters corresponding to each sub-cost optimization objective, wherein the second computing nodes comprise sub-cost optimization objective corresponding nodes and cost optimization weight parameter nodes;

and configuring a computational relationship between the first computational graph node and the one or more second computational graph nodes according to the computational relationship of the cost optimization objective function.

Preferably, the first neural network further comprises a scheduling parameter node; the step of training the first neural network using the microgrid output training samples according to an optimization objective function comprises:

inputting the known energy type and the known output power of each microgrid to the first input node, and mapping according to the first neural network to obtain an output scheduling result;

updating the scheduling parameters corresponding to the scheduling parameter nodes so that the output scheduling result conforms to the optimized objective function;

updating the steps using all the known energy types and the known output power until the output scheduling results corresponding to the predetermined number of the known energy types and the known output power conform to the optimization objective function.

Preferably, the method for optimizing the multi-microgrid scheduling mechanism further comprises:

constructing a second neural network, and mapping the neural network of the second neural network as a calculation function of the output of each microgrid, wherein the second neural network comprises a third input node and a third output node, and the neural network is mapped into a calculation relation from the third input node to the third output node;

representing the energy type and the output power of each distributed power supply in each micro-grid by using the third input node, wherein the third output node represents the energy type and the output power of the micro-grid calculated according to the output calculation function;

and calculating the energy type and the output power of each micro-grid according to the output calculation function.

According to a second aspect of the present invention, the present invention further provides an optimization system of a multi-microgrid scheduling mechanism based on a neural network, including:

the first neural network construction module is used for constructing a first neural network, and mapping the neural network of the first neural network as an output scheduling function of a multi-microgrid scheduling mechanism, wherein the first neural network comprises a first input node and a first output node, and the neural network is mapped into a calculation relation from the first input node to the first output node;

an optimization objective function configuration module, configured to use the first input node to represent an energy type and an output power of a microgrid, where the first output node represents an output scheduling result calculated according to the output scheduling function, and configure an optimization objective function of the first neural network according to an optimization objective of the multi-microgrid scheduling mechanism, where the optimization objective function depends on the first output node;

a training sample generation module, configured to generate a microgrid output training sample for the first neural network training using a priori distributions of known energy types and known output powers in the first input node;

the neural network training module is used for training the first neural network by using the microgrid output training sample according to the optimization objective function;

and the output scheduling result calculating module is used for assigning the energy type and the output power of each micro-grid to a first input node of the first neural network when the first neural network is trained, and calculating the output scheduling result of each micro-grid by using the first neural network.

Preferably, the first neural network is configured to compute a portion of a graph; the optimized target value of the objective function corresponds to a first computational graph node of the computational graph; the optimization objective function configuration module comprises:

the calculation graph node configuration sub-module is used for configuring the first calculation graph node and configuring one or more second calculation graph nodes on which the first calculation graph node depends according to an optimization target of the multi-microgrid scheduling mechanism; and

a computation relationship configuration submodule for configuring computation relationships between the first computation graph node and the one or more second computation graph nodes;

Preferably, the first neural network further comprises a scheduling parameter node; the neural network training module comprises:

the node input submodule is used for inputting the known energy type and the known output power of each microgrid to the first input node, and obtaining an output scheduling result according to the mapping of the first neural network;

the scheduling parameter updating submodule is used for updating the scheduling parameters corresponding to the scheduling parameter nodes so as to enable the output scheduling result to accord with the optimization objective function;

and the cyclic updating submodule is used for updating the steps by using all the known energy types and the known output power until the output scheduling results corresponding to the predetermined number of the known energy types and the known output power conform to the optimization objective function.

Preferably, the optimization system of the multi-microgrid scheduling mechanism further comprises:

the second neural network construction module is used for constructing a second neural network, and mapping the neural network of the second neural network as a calculation function of the output of each microgrid, wherein the second neural network comprises a third input node and a third output node, and the neural network is mapped into a calculation relation from the third input node to the third output node;

the node configuration module is used for representing the energy type and the output power of each distributed power supply in each microgrid by using the third input node, and the third output node represents the energy type and the output power of the microgrid calculated according to the output calculation function;

and the micro-grid computing module is used for computing the energy type and the output power of each micro-grid according to the output computing function.

According to the technical scheme, a first neural network is constructed, and the neural network is mapped to be used as an output scheduling function in a multi-microgrid scheduling mechanism; the method comprises the steps of using a first input node in a first neural network as an energy type and output power of each micro-grid, using the first output node as an output scheduling result to configure an optimization objective function of the first neural network, further generating a training sample through the known energy type and output power, using the optimization objective function as a guide, using the training sample to train the neural network, and obtaining the output scheduling result of each micro-grid quickly. Various types of output scheduling functions can be configured through the first neural network, and then the multi-microgrid scheduling mechanism can be automatically optimized; the optimization method does not require a mechanism optimizer to have hard game theory professional knowledge and strong specificity. The optimization scheme of the multi-microgrid scheduling mechanism based on the neural network can be universally applied to most of practical electric power application scenes, so that the optimal multi-microgrid scheduling mechanism adaptive to the practical electric power application scenes is automatically and quickly optimized, and the optimal output scheduling result of the microgrid is automatically calculated through the neural network. The method and the device solve the problems that complex knowledge and high super level are required for excitation mechanism optimization, the universality is not high, and a new excitation mechanism is difficult to optimize quickly when an application scene is changed in the prior art.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.

Fig. 1 is a schematic flowchart of an optimization method of a multi-microgrid scheduling mechanism based on a neural network according to an embodiment of the present invention;

FIG. 2-1 is a schematic structural diagram of a first neural network provided in an embodiment of the present invention;

FIG. 2-2 is a schematic structural diagram of a second neural network provided in an embodiment of the present invention;

FIG. 3 is a flowchart illustrating a first method for optimizing the configuration of an objective function according to the embodiment shown in FIG. 1;

FIG. 4 is a flowchart illustrating a second method for optimizing the configuration of an objective function according to the embodiment shown in FIG. 1;

FIG. 5 is a flowchart illustrating a method for configuring a cost-optimized objective function according to the embodiment shown in FIG. 1;

FIG. 6 is a schematic flow chart diagram illustrating a neural network training method provided in the embodiment shown in FIG. 1;

fig. 7 is a schematic flowchart of a second optimization method for a neural network-based multi-microgrid scheduling mechanism according to an embodiment of the present invention;

fig. 8 is a schematic structural diagram of an optimization system of a multi-microgrid scheduling mechanism based on a neural network according to an embodiment of the present invention;

FIG. 9 is a block diagram illustrating an optimized objective function configuration module according to the embodiment shown in FIG. 8;

FIG. 10 is a schematic structural diagram of a neural network training module provided in the embodiment shown in FIG. 8;

fig. 11 is a schematic structural diagram of an optimization system of a second neural network-based multi-microgrid scheduling mechanism according to an embodiment of the present invention;

fig. 12 is a schematic structural diagram of an optimization system of a third neural network-based multi-microgrid scheduling mechanism according to an embodiment of the present invention.

The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.

Detailed Description

The invention is further described with reference to the following drawings and detailed description.

It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

In the prior art, the traditional multi-microgrid scheduling mechanism optimization needs high professional knowledge and acute economic intuition, and is a complex and difficult task, and each time an application scene is newly added, an economic engineer needs to pertinently adjust an original excitation mechanism and optimize a new excitation mechanism, so that the method can be suitable for the excitation mechanism of the new application scene, and the new excitation mechanism is difficult to be rapidly optimized.

In order to solve the above problems, the following embodiments of the present application optimize a multi-microgrid scheduling mechanism by using a neural network technology, so as to quickly and automatically optimize an optimal multi-microgrid scheduling mechanism suitable for various economic application scenarios.

First, in order to clearly and clearly illustrate the neural network-based multi-microgrid scheduling mechanism optimization scheme provided in the following embodiments of the present application, the terms related to the multi-microgrid scheduling mechanism need to be explained as follows:

(1) and a micro-grid: the system is also called a microgrid and refers to a small power generation and distribution system which is composed of distributed power supplies, energy storage devices, energy conversion devices, loads, monitoring and protection devices and the like. The micro-grid aims to realize flexible and efficient application of the distributed power supply, so that the grid connection problem of the distributed power supplies with large quantity and various forms is solved. The development and extension of the micro-grid can fully promote the large-scale access of distributed power sources and renewable energy sources, realize the high-reliability supply of various energy source types of loads, and is an effective mode for realizing an active power distribution network, so that the traditional power grid is transited to a smart power grid.

(2) And an output scheduling result: the output scheduling result is a scheduling mode of the upper-layer scheduling center for the output of all or part of the micro-grid which can be controlled. For example, the output scheduling result can determine the output or input required power of each microgrid for the scheduling center. The set of all possible contribution schedule results is denoted by X. The outcome of a particular achievable contribution schedule is denoted by x. In most cases, the output scheduling result is an individual assignment result x of each micro-grid i by the scheduling center_iComposition, i.e. x ═ x₁,x₂,…,x_n). Each x_iIt may also be a vector comprising a plurality of individual contribution schedule results.

(3) And energy type: the type of energy source depends on the type of power source of each distributed power source in the microgrid. Generally, a large amount of clean energy is generated in a micro-grid, such as wind power generation, photovoltaic power generation, water conservancy power generation and the like, and corresponding energy types are wind power, photoelectricity and hydropower. The energy type is denoted t in this application. In most cases, the energy type is determined by the energy type t of the distributed power source i in each microgrid_iComposition, i.e. t ═ t (t)₁,t₂,…,t_n) Each t of_iIt may also be a vector comprising a plurality of type parameters, the set of all possible energy types being denoted T.

(4) And output power: the output power refers to the output of the micro-grid, including the actual output and the output adjusted after subsequent dispatching by the dispatching center. Although the output power is a variable that is always changing dynamically, it can be considered as an observable constant when analyzed. Whether there are some variables that can be considered as output power depends on the application scenario of such variables, the output power being denoted by γ. Gamma can exchange power gamma global and individual output power gamma of each micro-grid i through a public connection point_iAnd (4) forming. I.e. gamma-gamma (γ)^globalγ,γ^individual) Wherein γ is^individual＝(γ₁,γ₂,…,γ_n) And i is 1,2, …, and n is a vector formed by the individual output powers of all the micro grids. Gamma ray^global,(γ₁,γ₂,…,γ_n) It may still be a vector itself. All possible values of the output power are taken together denoted as Γ.

(5) And an output scheduling function: output scheduling function

Is a function of the energy type and/or output power, and this selection function maps to a scheduling result x ═ f (t, γ) for each set of possible values (t, γ). In most cases, f is assigned a function f by the individual to each microgrid_iComposition, i.e. f () ═ f₁(.),f₂(.),…,f_n(.)), where x_i＝f_i(t, γ) is the result of individual assignment to the microgrid i. In some application scenarios, the argument of the output scheduling function may not include γ, i.e., x ═ f (t).

In combination with the above related terms of the multi-microgrid scheduling mechanism, the following describes in detail an optimization scheme of the multi-microgrid scheduling mechanism based on a neural network provided by the embodiments of the present application. According to the optimization scheme of each multi-microgrid scheduling mechanism, a neural network is used as an approximation of an output scheduling function, and the neural network is trained according to sample data which is distributed a priori and generated randomly so as to obtain optimal parameters; and using the neural network configured with the optimal parameters as an approximation of the output scheduling function to calculate the allocation result of the economic resources.

Referring to fig. 1 in detail, fig. 1 is a schematic flowchart of an optimization method of a multi-microgrid scheduling mechanism based on a neural network according to an embodiment of the present invention. As shown in fig. 1, the optimization method of the neural network-based multi-microgrid scheduling mechanism includes the following steps:

s110: constructing a first neural network, and mapping the neural network of the first neural network as an output scheduling function in a multi-microgrid scheduling mechanism, wherein the first neural network comprises nodes and edges, and the nodes comprise first input nodes and first output nodes; edges represent computational relationships. The neural network is mapped to a computational relationship from a first input node to a first output node.

And combining a specific application scene of the micro-grid, using the first input node as the output power and the energy type of the micro-grid in the multi-micro-grid scheduling mechanism, and using the first output node as an output scheduling result calculated according to the output scheduling function.

The multi-microgrid scheduling mechanism provided by the embodiment of the application is applied to a scene of a multi-microgrid, wherein the types of the multi-microgrid comprise wind power, hydropower, photovoltaic power generation and the like, and the output power of the multi-microgrid comprises parameters such as wind power output, photovoltaic output, gas turbine output, storage battery output and load. Therefore, the energy type and the output power of the micro-grid represented by the first input node can be calculated through the neural network to obtain an accurate output scheduling result, and the output of each micro-grid is adjusted according to the output scheduling result.

Specifically, as shown in fig. 2-1 and 2-2, the first neural network may be configured as part of a computational graph. For example, fig. 2-1 schematically shows a computational graph, of which the first neural network 2100A may be configured, according to an embodiment of the present method. The computation graph is composed of nodes, and edges among the nodes represent computation relations among the nodes. The nodes in the neural network include a first input node, a first output node, a parameter node, and a hidden layer node. For example, in the neural network 2100A shown in fig. 2-1, the

nodes

2101 and 2103 are first input nodes, the node 2201 is a first output node, and the

nodes

2301 and 2315 are hidden layer nodes. Each node may represent a tensor, which may be of any order, for example, a zeroth order tensor is a scalar, a first order tensor is a vector, a second order tensor is a matrix, a third order tensor is a list composed of matrices, and the like; alternatively, the tensor can be implemented as a multidimensional array. An m-order tensor has m dimensions, each dimension j having a length l of a positive integer_jTensors of one or more orders, the shape of which is m elements of length per dimensionA tuple of elements. Each dimension of the tensor is m-1 dimension tensor with the length number of the dimension. The tensor comprises one or more scalars, and each scalar comprised in the tensor is defined as an element of the tensor. In the embodiments of the present application, for the sake of simplicity, in a case where no ambiguity is caused, a node and a tensor represented by the node are not distinguished, and both the node and the tensor represented by the node are represented by the same mathematical symbol. The computational relationship between the nodes is a computational relationship between the tensors represented by the nodes, for example, for the case where the nodes In and Out, b represents a first order tensor and W represents a second order tensor, there may be the following computational relationship between In and Out: out is W × In; wherein, the node Out is obtained by matrix multiplication of a matrix W and a vector In; or the following calculated relationship exists: out is W multiplied by In + b, and the calculation relationship shows that the node Out is obtained by matrix multiplication of a matrix W and a vector In and addition of a vector b; or there is a calculated relationship as follows: out is σ (W × In + b), where σ may be a sigmoid function or ReLu function, and the calculation relationship indicates that node Out is a vector obtained by performing matrix multiplication on matrix W and vector In and adding a vector b to obtain elements at a position corresponding to vector (W × In + b) and performing function σ transformation; when the computational graph nodes have the above-described form of computational relationship, W is referred to as a weight, b is referred to as a bias, and a function σ is referred to as an activation function. A node Out having a calculation relationship Out ═ σ (W × In + b) is also referred to as a neuron. There may also be other types of computational relationships between nodes In and Out. For example: the scalar node Out is a scalar formed by taking Out one element from the first-order tensor node In; the first order tensor node Out is a first order vector of length s formed by stacking (stack) a plurality of scalar nodes In1, …, Ins; the scalar nodes Out1, …, Outs may be s scalars formed by unstacking (unstack) first order tensor nodes In of length s. For another example: the first order tensor node Out may be a longer first order tensor node formed by the concatenation (concat) of two first order tensor nodes In1 and In2, and so on. In the above calculation relationship, the node Out is referred to as being dependent on the node In, or the node In is dependent on the node Out.

In the first neural network, a node which is not depended on by any node in the neural network is called a first output node, and in this embodiment, the first output node represents an output scheduling result calculated by the microgrid according to the output scheduling function. Nodes that do not depend on any node in the neural network fall into two categories: one is a first input node, and the value of the node is acquired by a user; in an embodiment of the application, the first input node represents the energy type and output power of the microgrid. The other type is a parameter node, and the node stores parameters (such as the parameters W and b) included in the neural network calculation relationship. Such parametric nodes are part of the neural network map, in the present embodiment as coefficients in the contribution scheduling function. The parameter nodes comprise trainable parameter nodes and invariant parameter nodes; the values of the invariant parameter nodes are kept unchanged in the training process of the neural network, and the values of the trainable parameter nodes are continuously updated in the training process of the neural network. In the embodiment of the application, the neural network is continuously trained through the energy type and the output power, and the value of the trainable parameter node of the neural network is updated.

Note that the nodes of the neural network include an optimization objective function further composed of certain calculation relations among the first output nodes, as shown in fig. 2-1, the optimization objective function may also be configured to include a node Obj in the calculation graph of the neural network, and the first output nodes are depended on by the nodes representing the optimization objective function; in the present application, however, the first neural network is determined to not include nodes representing the optimization objective function, the reliance of the first output node on the nodes representing the optimization objective function does not affect the independence of the first output node on any node in the neural network. For example, in FIG. 2-1, the neural network is determined to not include a portion 2100A of the computational graph that represents nodes of the optimization objective function; a first output node x (2201) is depended on by a node Obj representing an optimization objective function, without affecting that the first output node is not depended on by any node in the neural network 2100A.

The computational relationship of each node in the first neural network constitutes a composite computational relationship between a first output node and a first input node of the first neural network, also referred to as a first input node of the first neural networkA neural network mapping to a first output node. The neural network mapping was used in the middle century of this application as the contribution scheduling function of the multi-microgrid scheduling mechanism. In this neural network map, other nodes in the neural network other than the first output node, the first input node, and the parameter node are referred to as hidden nodes, and the hidden nodes store intermediate calculation results of the neural network map. Training the neural network is to continuously optimize the values of all or part of the parameter nodes so that the neural network mapping satisfies a certain condition. This step uses neural network mapping as the aforementioned contribution scheduling function

That is, the computational relationship of the output force scheduling function f () is considered to be consistent with the composite computational relationship between the first output node and the first input node of the neural network. For any value of the f (·) argument, the value of the argument may be assigned to a first input node of the neural network, and forward propagated through the neural network to obtain a value of a first output node of the corresponding neural network, which is the value of f (·) corresponding to the argument.

The following explains a process related to configuring a neural network according to various embodiments of the present application with reference to fig. 2-1 to 2-2.

As shown in fig. 2-1, the neural network is configured as follows: the first input node of the neural network comprises a node representing output power in the microgrid and/or a node representing an energy type. For example, in the neural network described in fig. 2-1, three first input nodes may be included, respectively node t (2101) representing the energy type, node γ individual (2102) representing the individual output power of the respective participants, and node γ global (2103) representing the common connection point interaction power. Note that when the energy type t of each participant_iWhen all are scalars, the node t (2101) is a one-dimensional tensor; when the energy type t of each participant_iWhen both are vectors, node t (2101) is a two-dimensional tensor. The same is true for node γ individual (2102) representing individual output powers; similarly, the node gamma global (2103) representing PCC interaction power of the public connection pointWhich may be a scalar or a vector.

In addition, as shown in fig. 2-2, the neural network is further configured to: the first output node of the neural network represents a scheduling result of the output of each microgrid. For example, in the neural network 2100A shown in fig. 2-1, a first output node is set as the node 2201 representing the force schedule result x. Note that individual contribution scheduling result x for each microgrid_iWhich may be a scalar or a vector.

The neural network is further configured to: the neural network includes a plurality of hidden nodes. The hidden nodes may be connected to each other according to the above calculation relationship Out ═ σ (W × In + b), for example. For example, in the neural network of FIG. 2-1, the nodes 2301-2315 are hidden nodes.

In some embodiments, because there are multiple micro-grids, the energy type node t may be de-stacked₁,t₂,…,t_n(ii) a Node gamma after individual output power node gamma indevivual is de-stacked₁,γ₂,…,γ_nOr node x after de-stacking of output scheduling result node x₁,x₂,…,x_nAs the first input node or the first output node of the neural network, respectively, which does not affect the implementation of the method. For example, in the neural network 2100B of FIGS. 2-2, node t after unstacking node t (2101) may be performed₁,t₂,…,t_n(3101-₁,γ₂,…,γ_n(3201-3204) is used as a first input node of the first neural network, and the node x after the original output scheduling result node x (2201) is de-piled₁,x₂,…,x_n(3401-3404) as a first output node of the neural network.

In other embodiments, the neural network is comprised of a plurality of sub-neural networks, each representing a contribution schedule for each microgrid. For example, in some embodiments, the force scheduling function f () may be set (f)₁(t₁,γ₁,γ),f₂(t₂,γ₂,γ),...,f_n(t_n,γ_nγ)), wherein the output scheduling result x for the microgrid i_i＝f_i(t_i,γ_iγ) depends only on the energy type of the microgrid, the individual output power of the microgrid and the point of common connection interaction power. At this time, the first input node may be used as t_i、γ_iAnd gamma, the first output node is x_iAs a neural network mapping of the sub-neural network of_i(.). And taking a composite neural network formed by n sub neural networks as the neural network. For example, in fig. 2-2, a composite neural network 2100B composed of n such sub-neural networks 3501-3503 is taken as the neural network.

For example, in fig. 2-2, the n sub-neural networks 3501-3503 may have the same structure and share the same set of parameter nodes to ensure that the incentive mechanism is the same for different participants. For example, different nodes of the neural network can be divided into a plurality of layers, each layer comprises a plurality of nodes, different layers can be connected through convolution, and parameter nodes in a convolution kernel are shared among different nodes of the same layer.

After the first neural network is obtained, for any given set of specific values of output power and energy type, the specific values can be assigned to a first input node of the neural network, and the values of the first output node obtained by the first neural network through forward propagation can be used as a configuration result for each microgrid.

S120: and configuring an optimization objective function of the first neural network according to an optimization objective of the multi-microgrid scheduling mechanism. The optimization objective function may be represented in the first neural network using a computational graph node.

This step configures an optimization objective function of the first neural network according to one or more of a first input node, a first output node, and a mechanism cost optimization objective of the first neural network.

Referring specifically to fig. 3, the neural network includes a calculation relationship in addition to each node such as the first input node, the first output node, and the hidden node. As shown in fig. 3, the above step S120 shown in fig. 1: the method for configuring the optimization objective function of the neural network according to the cost optimization objective of the multi-microgrid scheduling mechanism specifically comprises the following steps:

s210: the method comprises the steps of configuring a first computational graph node according to an optimization objective of a multi-microgrid scheduling mechanism, and configuring one or more second computational graph nodes on which the first computational graph node depends.

S220: and configuring the calculation relationship between the first calculation graph node and the one or more second calculation graph nodes according to the optimization target of the multi-microgrid scheduling mechanism. And the second calculation graph nodes comprise nodes corresponding to the sub-optimization targets and target parameter nodes.

Specifically, the calculation relationship between the calculation graph nodes Obj and the nodes on which the calculation graph nodes Obj depend is configured according to the cost optimization objective of the multi-microgrid scheduling mechanism. The optimization objectives may include: solution strategy balancing, individuality, incentive compatibility, budget balancing, allocation efficiency, profit maximization, cost minimization, social optimality, fairness, and collusion resistance. Each corresponding to a set of equality, inequality, or extremum targets. According to the above definitions and notations, the cost optimization objective of the multi-microgrid scheduling mechanism can be further defined as follows, and an equality, inequality or extremum objective corresponding to the cost optimization objective can be written.

In addition, the calculation relationship between the calculation graph node Obj and the node on which it depends is obtained according to the optimization objective function of the neural network. The optimization objective function comprises a cost optimization objective function, and thus, the optimization objective function for configuring the neural network comprises the steps of respectively configuring each sub-cost optimization objective in the cost optimization objectives of the multi-microgrid scheduling mechanism, configuring each sub-cost optimization objective representative node as the sub-optimization objective function for the cost optimization objective, and combining the sub-optimization objective functions into a global optimization objective function as the optimization objective function.

Specifically, the step of configuring the optimization objective function of the neural network according to the optimization objective of the multi-microgrid scheduling mechanism is shown in fig. 4, where the step of configuring the optimization objective function of the neural network includes:

s310: and decomposing the optimization target of the multi-microgrid scheduling mechanism into a plurality of sub-optimization targets.

S320: and configuring a plurality of sub-optimization objective functions which correspond to the sub-optimization objectives one to one.

S330: and sequentially combining the plurality of sub-optimization objective functions into a global optimization objective function of the neural network.

In other embodiments, other methods may also be adopted to combine the sub-optimization objective functions into a global optimization objective function, for example, a maximum value of each sub-optimization objective function is adopted as the global optimization objective function, which is not described herein again.

Specifically, the optimization objective function includes a cost optimization objective function, as shown in fig. 5, the above step S120 shown in fig. 1: configuring an optimization objective function of the first neural network according to an optimization objective of a multi-microgrid scheduling mechanism, which specifically comprises the following steps:

s410: and configuring the first computational graph node according to the comprehensive cost optimization objective of the cost optimization objective function.

S420: and configuring one or more second computing nodes depended by the first computing graph node according to the sub-cost optimization objectives of the cost optimization objective function and the cost optimization weight parameter corresponding to each sub-cost optimization objective, wherein the second computing nodes comprise sub-cost optimization objective corresponding nodes and cost optimization weight parameter nodes.

S430: and configuring the calculation relationship between the first calculation graph node and the one or more second calculation graph nodes according to the calculation relationship of the cost optimization objective function.

For example, cost minimization is taken as an optimization target, and the first computational graph node is configured to represent the total composite cost Wr of the multi-microgrid system; the second computational graph node is configured to run a maintenance cost Wr1, a gas turbine power generation cost Wr2, an environmental pollution cost Wr3, and a microgrid trading cost Wr 4. Accordingly, the overall cost for the minimized multi-microgrid system is

I.e. first computation graph node and second computationThe computational relationships between graph nodes. Each second computation graph node depends on the first output node and other nodes, so that the actual size of the second computation graph node can be obtained by configuring the computation relationship between the second computation graph node and other nodes in the first neural network, and specifically, the operation and maintenance cost of the microgrid i is

Wherein, p1, p2, p3 and p4 are the output of photovoltaic power, wind power, gas turbine and accumulator respectively, and k1 to k4 are the coefficients of the operation and maintenance cost respectively.

In addition, in the embodiment shown in fig. 1, the method for optimizing the multi-microgrid scheduling mechanism further includes:

s130: a microgrid contribution training sample for training a first neural network is generated using a known energy type and a known output power in a first input node. Because the known energy type and the known output power are both known, and the corresponding output scheduling result is also known, the first neural network is trained through the known data, the neural network mapping of the first neural network can be continuously optimized, and the trainable parameters with high accuracy are obtained.

S140: and training the first neural network by using the microgrid output training sample according to the optimization objective function.

Referring to fig. 6, as shown in fig. 6, the training steps are as follows, step S140: according to the optimization objective function, training a neural network by using a microgrid output training sample specifically comprises the following steps:

s510: and inputting the known energy type and the known output power of each micro-grid into a first input node, and mapping according to the neural network to obtain an output scheduling result.

S520: and updating the scheduling parameters corresponding to the scheduling parameter nodes so that the output scheduling result conforms to the optimization objective function.

S530: and updating the steps by using all known energy types and known output powers until the output scheduling results corresponding to the predetermined number of known energy types and known output powers conform to the optimization objective function.

Specifically, the microgrid output training sample is used for continuously training the neural network, so that the output scheduling result approaches the output scheduling result corresponding to the optimization objective function, the value of trainable parameters in the neural network is continuously updated, the trainable parameters of the first neural network trained according to the microgrid output training sample are adjusted, and the output scheduling result output by the first output node accords with the optimization objective function. Wherein the values of all trainable parameters may be updated using a gradient descent method. Of course, Adam's algorithm, random gradient descent method, momentum gradient descent method, small batch gradient descent method, etc. may be used. Such neural network training algorithms are described in detail in the published materials for neural networks and will not be described in detail here.

In addition, the method for optimizing the multi-microgrid scheduling mechanism shown in fig. 1 further includes the following steps:

s150: and assigning the energy type and the output power of each micro-grid to a first input node of a first neural network, and calculating by using the first neural network to obtain the output scheduling result of each micro-grid.

The output scheduling result reflects the scheduling condition of the scheduling center for outputting the output of each microgrid, so that each microgrid can distribute the output electric energy according to the output scheduling result.

In addition, each microgrid may include a plurality of distributed power sources, the type and output power of each distributed power source are not consistent, and at this time, the distributed power sources need to be optimized to obtain the energy type and output power of the microgrid suitable for the needs of the first neural network. Referring to fig. 7, fig. 7 is a schematic flow chart of an optimization method of a multi-microgrid scheduling mechanism based on a neural network according to a second embodiment of the present invention, where the optimization method of the multi-microgrid scheduling mechanism further includes, in addition to the steps shown in fig. 1:

s610: and constructing a second neural network, and mapping the neural network of the second neural network as a calculation function of the output of each microgrid, wherein the second neural network comprises a third input node and a third output node, and the neural network is mapped into a calculation relation from the third input node to the third output node.

S620: and representing the energy type and the output power of each distributed power supply in each micro-grid by using the third input node, and representing the energy type and the output power of the micro-grid calculated according to the output calculation function by using the third output node.

S630: and calculating the energy type and the output power of each micro-grid according to the output calculation function.

According to the technical scheme, the second neural network is constructed, then the energy type and the output power of each distributed power supply in the microgrid are used as third input nodes, the accurate energy type and the accurate output power of the microgrid can be obtained through the corresponding output calculation function, and therefore the accurate energy type and the accurate output power can be input into the first neural network and used as the value of the first input node for input calculation, and an accurate output scheduling result is obtained.

The structure of the second neural network is similar to that of the first neural network, and can be seen in fig. 2-1 and 2-2, which are not described in detail herein.

In addition, corresponding to the optimization method of the multi-microgrid scheduling mechanism based on the neural network provided by the above embodiments of the present application, the present application also provides an embodiment of an optimization system of the multi-microgrid scheduling mechanism based on the neural network. Since the technical idea is consistent with the above method, the related technical effects are not described in detail.

Referring to fig. 8 in particular, fig. 8 is a schematic structural diagram of an optimization system of a multi-microgrid scheduling mechanism based on a neural network according to an embodiment of the present invention. As shown in fig. 8, the system for optimizing the multi-microgrid scheduling mechanism includes:

a first neural network constructing module 1010, configured to construct a first neural network, and map the neural network of the first neural network as an output scheduling function of a multi-microgrid scheduling mechanism, where the first neural network includes a first input node and a first output node, and the neural network is mapped as a calculation relationship from the first input node to the first output node;

an optimization objective function configuration module 1020, configured to use the first input node to represent an energy type and an output power of a microgrid, where the first output node represents an output scheduling result calculated according to the output scheduling function, and configure an optimization objective function of the first neural network according to an optimization objective of the multi-microgrid scheduling mechanism, where the optimization objective function depends on the first output node;

a training sample generating module 1030, configured to generate a microgrid output training sample for the first neural network training using a priori distributions of known energy types and known output powers in the first input node;

the neural network training module 1040 is configured to train the first neural network by using the microgrid output training sample according to the optimized objective function;

and the output scheduling result calculating module 1050 is configured to assign the energy type and the output power of each microgrid to a first input node of the first neural network when the first neural network is trained, and calculate the output scheduling result of each microgrid by using the first neural network.

Wherein the first neural network is configured to compute a portion of the graph; the optimized target value of the objective function corresponds to a first computational graph node of the computational graph; as shown in fig. 9, the optimization objective function configuration module 1020 includes:

a computation graph node configuration submodule 1021, configured to configure the first computation graph node and configure one or more second computation graph nodes on which the first computation graph node depends according to an optimization objective of the multi-microgrid scheduling mechanism; and

a computation relationship configuration submodule 1022, configured to configure computation relationships between the first computation graph node and the one or more second computation graph nodes;

As shown in fig. 10, the first neural network further includes a scheduling parameter node; the neural network training module 1040 includes:

the node input submodule 1041 is configured to input the known energy type and the known output power of each microgrid to the first input node, and obtain an output scheduling result according to the first neural network mapping;

the scheduling parameter updating submodule 1042 is configured to update the scheduling parameter corresponding to the scheduling parameter node, so that the output scheduling result conforms to the optimized objective function;

and a cyclic update submodule 1043, configured to update the foregoing steps using all the known energy types and known output powers until a predetermined number of output scheduling results corresponding to the known energy types and the known output powers meet the optimization objective function.

As shown in fig. 11, the optimization system of the multi-microgrid scheduling mechanism provided in this embodiment further includes, in addition to the modules shown in fig. 8:

a second neural network construction module 1100, configured to construct a second neural network, and map the neural network of the second neural network as a calculation function of the output of each microgrid, where the second neural network includes a third input node and a third output node, and the neural network is mapped as a calculation relationship from the third input node to the third output node;

the node configuration module 1110 is configured to use the third input node to represent the energy type and the output power of each distributed power source in each microgrid, and the third output node represents the energy type and the output power of the microgrid calculated according to the output calculation function;

and the microgrid calculation module 1120 is used for calculating the energy type and the output power of each microgrid according to the output calculation function.

In addition, as shown in fig. 12, an embodiment of the present invention further provides an optimization system for a multi-microgrid scheduling mechanism based on a neural network, including:

the memory 1004, the processor 1001, and the program for determining the multi-microgrid scheduling mechanism optimization method, which is stored in the memory 1004 and is executable on the processor 1001, implement the steps of the method for optimizing the multi-microgrid scheduling mechanism provided in any of the above embodiments when the program for determining the multi-microgrid scheduling mechanism optimization method is executed by the processor 1001.

In addition, the present invention also claims a computer-readable storage medium, where the computer-readable storage medium stores a determination program for multi-microgrid scheduling mechanism optimization, and when the determination program for multi-microgrid scheduling mechanism optimization is executed by a processor, the steps of the optimization method for multi-microgrid scheduling mechanism provided in any one of the above embodiments are implemented.

The specific embodiment of the computer-readable storage medium of the present invention is substantially the same as the embodiments of the optimization method of the multi-microgrid scheduling mechanism described above, and details are not repeated herein.

In summary, according to the optimization scheme of the multi-microgrid scheduling mechanism provided by the technical scheme of the application, a first neural network is constructed and mapped to serve as an output scheduling function in the multi-microgrid scheduling mechanism; the method comprises the steps of using a first input node in a first neural network as output power and an energy type of a multi-microgrid scheduling mechanism, using the first output node as an output scheduling result to configure an optimization objective function of the neural network, further randomly generating a training sample through prior distribution of uncertain random variables, using the optimization objective function as a guide, using the training sample to train the neural network, and obtaining the output scheduling result of each microgrid quickly. By constructing the first neural network, the multi-microgrid scheduling mechanism can be automatically optimized; the optimization method does not require a mechanism optimizer to have hard game theory professional knowledge and strong specificity. The optimization scheme of the multi-microgrid scheduling mechanism based on the neural network can be universally applied to most of real power scenes, so that the optimal multi-microgrid scheduling mechanism adaptive to the real power scenes is automatically and quickly optimized, and the optimal output scheduling result of the microgrid is automatically calculated through the neural network. The scheme solves the problems that in the prior art, mechanism optimization needs complex knowledge and high super level, the universality is not available, and a new mechanism is difficult to optimize quickly when the application scene of the power system is changed.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The usage of the words first, second and third, etcetera do not indicate any ordering. These words may be interpreted as names.

While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims

1. A multi-microgrid scheduling mechanism optimization method based on a neural network is characterized by comprising the following steps:

step 1: constructing a first neural network, and mapping the neural network of the first neural network as an output scheduling function of a multi-microgrid scheduling mechanism, wherein the first neural network comprises a first input node and a first output node, and the neural network is mapped to a calculation relation from the first input node to the first output node;

step 2: representing the energy type and the output power of the micro-grid by using the first input node, wherein the first output node represents an output scheduling result calculated according to the output scheduling function;

step 3: configuring an optimization objective function of the first neural network according to an optimization objective of the multi-microgrid scheduling mechanism, wherein the optimization objective function is dependent on the first output node;

step 4: generating a microgrid output training sample for the first neural network training using a priori distributions of known energy types and known output powers in the first input node;

step 5: training the first neural network by using the microgrid output training sample according to the optimization objective function;

step 6: and when the training of the first neural network is finished, assigning the energy type and the output power of each micro-grid to a first input node of the first neural network, and calculating by using the first neural network to obtain an output scheduling result of each micro-grid.

2. The optimization method of the neural network-based multi-microgrid scheduling mechanism of claim 1, characterized in that: the first neural network is configured to compute a portion of a graph; the optimized target value of the objective function corresponds to a first computational graph node of the computational graph; the step of configuring an optimization objective function of the first neural network according to an optimization objective of a multi-microgrid scheduling mechanism includes:

3. The optimization method of the neural network-based multi-microgrid scheduling mechanism of claim 1, characterized in that: the step of configuring the optimization objective function of the first neural network according to the optimization objective of the multi-microgrid scheduling mechanism includes:

4. The optimization method of the neural network-based multi-microgrid scheduling mechanism of claim 1, characterized in that: the first neural network further comprises a scheduling parameter node; the step of training the first neural network using the microgrid output training samples according to an optimization objective function comprises:

inputting the known energy type and the known output power of each microgrid to the first input node, and mapping according to the neural network to obtain an output scheduling result;

5. The optimization method of the neural network-based multi-microgrid scheduling mechanism of claim 1, characterized in that:

6. A neural network-based optimization system for a multi-microgrid scheduling mechanism, comprising:

7. The optimization system of the neural network-based multi-microgrid scheduling mechanism of claim 6, characterized in that: the first neural network is configured to compute a portion of a graph; the optimized target value of the objective function corresponds to a first computational graph node of the computational graph; the optimization objective function configuration module comprises:

8. The optimization system for a neural network-based multi-microgrid scheduling mechanism of claim 7, characterized in that: the first neural network further comprises a scheduling parameter node; the neural network training module comprises:

9. The optimization system for a neural network-based multi-microgrid scheduling mechanism of claim 7, further comprising: