A power dispatching method and related equipment for a virtual power plant

Abstract

The present invention relates to the field of data processing, and specifically provides a power dispatching method and related equipment for a virtual power plant, obtaining a distributed charging network map, including a first lamp pole unit and a second lamp pole unit; obtaining the basic implicit representation vector of the first lamp pole unit and the second lamp pole unit; processing the basic implicit representation vector of the first lamp pole unit and the second lamp pole unit based on a machine learning network, and determining the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network map; adjusting the machine learning network based on the real involvement relationship and the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network map, and determining the implicit representation vector of each lamp pole unit in the distributed charging network map based on the adjusted machine learning network. The present invention can provide more accurate representation information in the subsequent power dispatching process based on the implicit representation vector, so that the rationality and accuracy of power dispatching are higher.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a power dispatching method and related equipment for a virtual power plant. Background Art

[0002] With the continuous growth of energy demand and the development of power systems, virtual power plants, as an emerging concept of power supply and management, have gradually attracted widespread attention. Virtual power plants integrate distributed energy resources (such as distributed generation facilities, energy storage systems, etc.) to achieve flexible scheduling and optimized management of electricity, thereby improving the stability, reliability and efficiency of the power system.

[0003] In the architecture of virtual power plants, distributed charging networks are a key component. For example, in some urban areas, a large number of charging facilities are set up on both sides of the road. These charging facilities can not only charge electric vehicles and other equipment, but also can be integrated into a distributed charging network. There are complex relationships between the various charging points in this network (similar to the light pole units mentioned above), involving many aspects such as power transmission, distribution, and interaction with the external power grid.

[0004] Traditional power dispatching methods face many challenges when faced with such a complex distributed charging network. It is very difficult to accurately obtain the status information of each charging point and to perform effective feature representation, which leads to a lack of sufficiently accurate and comprehensive basis for power dispatching decisions, and it is easy to produce unreasonable dispatching results. Summary of the invention

[0005] In view of this, the present invention provides a power dispatching method and related equipment for a virtual power plant.

[0006] The technical solution of the present invention is achieved in this way:

[0007] In one aspect, the present invention provides a method for dispatching power of a virtual power plant, comprising:

[0008] Obtain a distributed charging network map, the distributed charging network map including a first lamp pole unit set and a second lamp pole unit set, the first lamp pole unit set and the second lamp pole unit set do not have an intersection, the first lamp pole unit set includes a first lamp pole unit, the second lamp pole unit set includes a second lamp pole unit, wherein the lamp pole unit is used to indicate data of charging lamp poles in a virtual power plant, the virtual power plant is composed of charging lamp poles arranged on both sides of a road; the distributed charging network map includes a real involvement relationship between the first lamp pole unit and the second lamp pole unit;

[0009] Obtaining a basic implicit representation vector of the first lamp pole unit and a basic implicit representation vector of the second lamp pole unit, wherein the basic implicit representation vector of the first lamp pole unit is obtained according to the involvement relationship between the lamp pole units in the first lamp pole unit set, and the basic implicit representation vector of the second lamp pole unit is obtained according to the involvement relationship between the lamp pole units in the second lamp pole unit set;

[0010] Performing feature mapping on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit based on a machine learning network to obtain an inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph;

[0011] Based on the real involvement relationship and the inferred involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, the machine learning network is calibrated to obtain a calibrated machine learning network;

[0012] Determine an implicit representation vector of each lamp pole unit in the distributed charging network graph based on a calibrated machine learning network;

[0013] Light pole unit power scheduling is performed based on the implicit representation vector of each light pole unit in the distributed charging network graph.

[0014] In a second aspect, the present invention provides a power dispatching device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor implements the steps in the above-mentioned method when executing the program.

[0015] The power dispatching method and related equipment of the virtual power plant provided by the present invention perform labeled training on the machine learning network based on the basic implicit representation vector of the first lamp pole unit and the second lamp pole unit, and the real involvement relationship between the first lamp pole unit and the second lamp pole unit. The implicit representation vector of the lamp pole unit in the distributed charging network map can be obtained based on the calibrated machine learning network, so that the implicit representation vector can more accurately represent the characteristics of the lamp pole unit in the distributed charging network map. In this way, based on the implicit representation vector, more accurate representation information can be provided in the subsequent power dispatching process, so that the rationality and accuracy of power dispatching are higher.

[0016] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present invention and, together with the specification, are used to explain the technical solutions of the present invention.

[0018] Figure 1 A schematic diagram of the implementation flow of a power dispatching method for a virtual power plant provided in an embodiment of the present invention.

[0019] Figure 2 A hardware entity schematic diagram of a power dispatching device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The embodiment of the present invention provides a method for dispatching power of a virtual power plant, which can be executed by a processor of a power dispatching device, wherein the power dispatching device can refer to a device with data processing capability, such as a server or a personal computer.

[0021] In the application scenario of the present invention, many rechargeable smart lamp poles are set on both sides of the road. These smart lamp poles form a distributed charging network, which can be used for their own use or be incorporated into the power grid to realize power dispatching. In the power dispatching method of the virtual power plant, graph theory technology can be used to take the relevant attribute data of the smart lamp poles as each graph unit in the graph, and establish the involvement relationship of each lamp pole. In the process, it is necessary to perform feature representation on each graph unit, and then use a general AI model to predict and infer the representation results to obtain feasible power dispatching decisions. It can be understood that the more accurate the feature representation of the graph unit, that is, the more accurate the data characterization result, directly affects the accuracy of the power dispatching decision. In the existing implementation method, random walk graph embedding is used for the feature representation of the graph unit. If the obtained node sequence has errors with the characteristics of the actual situation, it is easy to produce omissions and needs to be optimized.

[0022] Based on this, please refer to Figure 1 , is a flow chart of a power dispatching method for a virtual power plant provided by the present invention, the method comprising the following steps:

[0023] Step S100: Obtain a distributed charging network map, the distributed charging network map includes a first lamp pole unit set and a second lamp pole unit set, the first lamp pole unit set and the second lamp pole unit set do not have an intersection, the first lamp pole unit set includes a first lamp pole unit, the second lamp pole unit set includes a second lamp pole unit, wherein the lamp pole unit is used to indicate data of charging lamp poles in a virtual power plant, and the virtual power plant is composed of charging lamp poles arranged on both sides of the road; the distributed charging network map includes a real involvement relationship between the first lamp pole unit and the second lamp pole unit.

[0024] The first lamp pole unit set and the second lamp pole unit set are important parts of the distributed charging network graph. The two sets do not have an intersection, which means that the lamp pole units they contain are completely different. For example, in a specific area, suppose that the units corresponding to the lamp poles on one side of the road are classified as the first lamp pole unit set, and the units corresponding to the lamp poles on the other side of the road are classified as the second lamp pole unit set. The lamp pole unit here is actually a graph node in the graph, and each lamp pole unit contains data related to the charging lamp poles in the virtual power plant. These data may include the location information of the lamp pole (for example, expressed in geographic coordinates, such as the position of longitude x1 and latitude y1), the charging power range (such as the minimum charging power is P min1 , the maximum charging power is P max1 ), current power status (such as the remaining power is E1), connection status (whether it is normally connected to other light pole units or the power grid, which can be expressed as a Boolean value, such as True for connection and False for non-connection), etc.

[0025] The real involvement relationship refers to the actual correlation relationship between the first lamp pole unit and the second lamp pole unit in this distributed charging network map. This relationship may be based on various factors, such as physical connection, logical relationship of power transmission, etc. For example, if there is a possibility of power transmission between the first lamp pole unit and the second lamp pole unit in the topology of the power grid, then there is a certain involvement relationship. It may be that the power of the first lamp pole unit can be transmitted to the second lamp pole unit through a certain line, or they share a certain power regulation equipment, etc. This relationship is a real involvement relationship.

[0026] There are many ways for the power dispatching equipment to obtain the distributed charging network map. One way is to collect relevant data of each lamp pole through a sensor network, and then use the data to build a map. For example, a sensor is installed on each lamp pole, which can detect the power of the lamp pole, charging power and other data, and then transmit these data to the central processing unit. The power dispatching equipment can create a distributed charging network map based on these data using the construction algorithm of the graph data structure. Specifically, if the map is represented by an adjacency matrix, for the case of n lamp pole units, an n×n matrix A is created. If there is a direct relationship between lamp pole unit i and lamp pole unit j (such as a power transmission relationship), the values ​​of A[i][j] and A[j][i] can be set to values ​​representing the strength of this relationship (such as power transmission efficiency, etc.), and if there is no relationship, it is set to 0. Another way is to extract relevant data from an existing power equipment database to build a map. The database may already store information such as the location and connection relationship of the lamp poles. The power dispatching equipment can build a distributed charging network map based on this information.

[0027] Step S200: Obtain a basic implicit representation vector of the first lamp pole unit and a basic implicit representation vector of the second lamp pole unit, wherein the basic implicit representation vector of the first lamp pole unit is obtained based on the involvement relationship between the lamp pole units in the first lamp pole unit set, and the basic implicit representation vector of the second lamp pole unit is obtained based on the involvement relationship between the lamp pole units in the second lamp pole unit set.

[0028] The basic implicit representation vector is a pre-trained feature representation result, which can reflect the characteristics of the lamp pole unit in the distributed charging network map to a certain extent. For the basic implicit representation vector of the first lamp pole unit, it is obtained based on the involvement relationship between the lamp pole units in the first lamp pole unit set; similarly, the basic implicit representation vector of the second lamp pole unit is obtained based on the involvement relationship between the lamp pole units in the second lamp pole unit set.

[0029] The involvement relationship between lamp pole units contains a variety of information. For example, considering the power transmission aspect, it may involve factors such as efficiency and capacity limitation when power is transmitted from one lamp pole unit to another. Assume that there are lamp pole units A, B, and C belonging to the first lamp pole unit set, there is a power transmission line between lamp pole unit A and lamp pole unit B, and the transmission efficiency of the line is 80% (this data may be based on factors such as the material and length of the line), and the transmission efficiency between lamp pole unit B and lamp pole unit C is 70%. At the same time, the upper limit of the charging power of lamp pole unit A is 5 kilowatts, the upper limit of the charging power of lamp pole unit B is 4 kilowatts, and the upper limit of the charging power of lamp pole unit C is 3 kilowatts. This information about the upper limit of charging power and power transmission efficiency together constitutes the involvement relationship between lamp pole units.

[0030] There are many ways for the power dispatching equipment to obtain the basic implicit representation vector. One possible method is based on matrix decomposition technology. Suppose there is a matrix R representing the involvement relationship of the lamp pole units in the first lamp pole unit set, and its dimension is m×n (m represents the number of lamp pole units, and n represents the number of attributes related to the involvement relationship, such as the transmission efficiency and charging power upper limit mentioned above). The power dispatching equipment can decompose the matrix R into two matrices P and Q, so that R≈P×Q. The elements of the P matrix and the Q matrix here can be used as part of the basic implicit representation vector. For example, a row element of the P matrix may correspond to a part of the basic implicit representation vector of a lamp pole unit, and a column element of the Q matrix may be related to the feature representation of this lamp pole unit under different attribute dimensions.

[0031] Another technical means is the automatic encoding technology based on neural networks. The power dispatching equipment can build an automatic encoder and take the vector representing the relationship between the lamp pole units as input. For example, the upper limit of the charging power of a lamp pole unit and the power transmission efficiency with other lamp pole units form an input vector x. The autoencoder contains an encoder network and a decoder network. The encoder network maps the input vector x to a low-dimensional representation vector h, which can be used as an approximation of the basic implicit representation vector. The decoder network attempts to reconstruct the original input vector x from h. By minimizing the reconstruction error (such as the mean square error, the formula is MSE=1 / m∑(x-x')², where m is the number of samples, x is the original input vector, and x' is the reconstructed vector), the autoencoder can learn an effective basic implicit representation vector.

[0032] Step S300: Feature mapping is performed on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit based on the machine learning network to obtain the inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph.

[0033] A machine learning network is a specifically designed and trained computing model that processes input data and outputs the desired results. In this scenario, it is used to process the underlying implicit representation vectors of the lamp pole units to explore deeper relationships between the lamp pole units.

[0034] Feature mapping aims to transform the basic implicit representation vector from the original feature space to a new feature space that is more conducive to analyzing and mining relationships. For example, assuming that the basic implicit representation vector of the first light pole unit contains features such as location information, initial charging capacity, and connection stability with adjacent light pole units, these features may be messy in the original space and difficult to directly use to analyze the relationship between light pole units. Through feature mapping, the power dispatching equipment can transform these features into a new representation form that is more logical and relevant.

[0035] Taking a simple linear feature mapping as an example, assuming that the basic implicit representation vector of the first lamp pole unit is x=[x1, x2, x3], the linear mapping matrix used by the power dispatching device is A, and its dimension is 3×3 (the dimension here is determined according to the dimension of the input vector and the expected output dimension). Then the vector y after feature mapping can be obtained by the formula y=Ax. Here y is the representation in the new feature space, which may better reflect the potential characteristics of the first lamp pole unit in its relationship with the second lamp pole unit.

[0036] The reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph is obtained by further analyzing the results after the above feature mapping. This reasoning involvement relationship is a data-driven estimation of the potential relationship between lamp pole units. For example, in a distributed charging network, the first lamp pole unit and the second lamp pole unit may be physically far apart, but through feature mapping and subsequent reasoning, the power dispatching device may find that they are closely connected in the logical relationship of power dispatching. It may be because they both have a specific association with an intermediate power regulation device, or their charging demand patterns are complementary in the load balancing of the entire power grid.

[0037] Electric power dispatching equipment can use various types of machine learning networks to implement feature mapping. A common one is neural networks, especially multi-layer perceptrons (MLPs). In MLPs, the input layer receives the basic implicit representation vectors of the first lamp pole unit and the second lamp pole unit, and the hidden layer performs nonlinear transformations on the input through a series of neurons. Assuming that the basic implicit representation vector of the input first lamp pole unit is x1, the basic implicit representation vector of the second lamp pole unit is x2, the neuron activation function of the hidden layer is f, the weight matrix is ​​W, and the bias vector is b, then the output h of the hidden layer can be calculated by the formula h=f(W[x1;x2]+b) (here [x1; x2] means concatenating the two vectors into a longer vector). The output of this hidden layer is an intermediate result after feature mapping, and the final reasoning involvement relationship can be obtained through further processing.

[0038] Another implementation is a kernel-based machine learning algorithm, such as the kernel technique of support vector machine (SVM). By defining a suitable kernel function, such as the Gaussian kernel function ,

[0039] x i and x j are two sample vectors in the input data. In this embodiment of the present invention, x i and x j It can be regarded as feature vectors related to different lamp pole units, which contain information about the lamp pole units such as location, charging capacity, and power status. is a hyperparameter, called bandwidth. It controls the width of the Gaussian function, thus affecting the way the kernel function maps the data. The power dispatching equipment can map the original underlying implicit representation vector to a high-dimensional feature space, in which the relationship is analyzed and reasoned. The advantage of this kernel method is that it can effectively handle complex nonlinear relationships without directly calculating the high-dimensional space vector, thus helping to accurately obtain the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit.

[0040] Accurately obtaining the inference involvement relationship is crucial for the power dispatch of virtual power plants. If the inference involvement relationship is inaccurate, it may lead to wrong judgments in the power dispatch decision. For example, if it is wrongly inferred that there is no power support relationship between the first lamp pole unit and the second lamp pole unit, but in fact they can support each other in some cases, then the opportunity to optimize the load balance of the power grid may be missed during power dispatch, resulting in overcharging or undercharging of some lamp pole units, which in turn affects the operating efficiency and stability of the entire virtual power plant. In addition, if the inference involvement relationship deviates too much from the actual situation, it may also affect the subsequent adjustment of the machine learning network and the final power dispatch decision, making it impossible for the power dispatch of the entire virtual power plant to reach the optimal state, and may even cause power supply safety issues, such as local power grid overload or power supply interruption.

[0041] Step S400: Based on the real involvement relationship and the inferred involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network map, the machine learning network is calibrated to obtain a calibrated machine learning network.

[0042] The real involvement relationship is the actual correlation between the first lamp pole unit and the second lamp pole unit in the distributed charging network map, which is determined based on factors such as physical connection and power transmission logic. For example, in an actual distributed charging network, the first lamp pole unit and the second lamp pole unit may be connected through a specific power line, and the real power transmission capacity and the order of energy interaction between them are determined based on factors such as the capacity of the line and the restrictions on the power transmission direction. This is the real involvement relationship.

[0043] The inferred involvement relationship is an estimated relationship obtained by the power dispatching device after feature mapping the basic implicit representation vectors of the first lamp pole unit and the second lamp pole unit based on the machine learning network in step S300. It is obtained through analysis and processing of data and may have a certain deviation from the real involvement relationship.

[0044] The power dispatching equipment adjusts the machine learning network to enable it to better reflect the real relationship between the lamp pole units, thereby improving the accuracy of subsequent operations. The adjustment process is essentially an optimization process that aims to reduce the difference between the inferred involved relationship and the real involved relationship.

[0045] One calibration method is based on the error back propagation algorithm. Assume that the output of the machine learning network is the predicted value of the inference relationship between the first lamp pole unit and the second lamp pole unit, denoted as , and the value of the real involvement relationship obtained according to the actual situation is recorded as A loss function L can be defined to measure the difference between the predicted value and the true value, such as the mean square error (MSE) loss function, which is , where n is the number of samples (here it can be understood as samples at different observation times or different network states).

[0046] The power dispatching equipment calculates the gradient of the loss function with respect to the parameters of the machine learning network (such as the weights and biases in the neural network), and then updates these parameters according to the gradient descent algorithm. Specifically, if the parameters of the machine learning network are , then the parameter update formula is ,in is the learning rate, which determines the step size of parameter updates, is the loss function L with respect to the parameters By continuously iterating this process, that is, calculating the loss function, calculating the gradient and updating the parameters multiple times, the value of the loss function gradually decreases, so that the inferred involvement relationship is closer and closer to the real involvement relationship.

[0047] Another way to tune the model could be to use regularization techniques. For example, L2 regularization adds a penalty term to the loss function for the size of the machine learning network parameters. The new loss function can be expressed as ,in is the regularization parameter, is a parameter in the machine learning network. The role of this penalty term is to prevent the machine learning network from overfitting, that is, to prevent the network from being too complex and only fitting the noise in the training data, but not generalizing well to new data. The value of can control the strength of regularization, so that the machine learning network can maintain a certain generalization ability while fitting the data, thereby more accurately reflecting the real involvement relationship between the first lamp pole unit and the second lamp pole unit.

[0048] Step S500: Determine the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network.

[0049] The implicit representation vector is a feature representation of the lamp pole unit in the distributed charging network map, which can reflect the various characteristics of the lamp pole unit and the relationship between the lamp pole units and other lamp pole units. The calibrated machine learning network is the network calibrated in step S400, which has been optimized to better reflect the real involvement relationship between the lamp pole units.

[0050] The process of determining the implicit representation vector of each lamp pole unit using the calibrated machine learning network is a process of data processing and conversion based on the network model. For example, suppose there are multiple lamp pole units in the distributed charging network map, and each lamp pole unit has its own attribute data, such as charging power, power level, connected device type, etc. These attribute data have been processed into basic implicit representation vectors in the previous steps and have been feature mapped and calibrated by the machine learning network.

[0051] If the calibrated machine learning network is a neural network, such as a multi-layer perceptron (MLP), for each lamp pole unit, the power dispatching device inputs its relevant input data (which can be the initial data or pre-processed data related to the lamp pole unit) into the neural network. Assume that the input layer has n neurons and the input data is x=[x1, x2,…, x n ], the weight matrix of the neural network is W, the bias vector is b, and the neuron activation function of the hidden layer is f. Then the output h of the hidden layer can be calculated by the formula h=f(Wx + b). The output of this hidden layer can be used as part or all of the implicit representation vector of the lamp pole unit, depending on the design structure of the network.

[0052] For another example, if a machine learning network related to a support vector machine (SVM) is used to determine the implicit representation vector, the power dispatching equipment will process the data of the light pole unit according to the calibrated SVM model parameters (such as support vectors, kernel function parameters, etc.). For the data x of a light pole unit, the kernel function K(x, x i ) (where x i The calculation of the support vector can obtain a value related to the relationship with other data points (support vectors). After certain combination and transformation, this value can be used as the implicit representation vector of the lamp pole unit.

[0053] Taking a simple virtual power plant scenario as an example, assume that there are three light pole units A, B, and C. Light pole unit A has a large charging power, a high power level, and a good connection stability with the power grid; light pole unit B has a moderate charging power, a medium power level, and has a specific connection relationship with some local power devices; light pole unit C has a small charging power, a low power level, and is located at the edge of the network. If the implicit representation vector can accurately reflect these characteristics and relationships, then when performing power dispatch, the power dispatching equipment can make reasonable decisions based on these implicit representation vectors.

[0054] For example, when it is necessary to obtain power supplement from the power grid, based on the implicit representation vectors of each lamp pole unit, the power dispatching device can determine that lamp pole unit A has a greater ability to receive and store more power, while lamp pole unit C may only be suitable for receiving a small amount of power supplement due to its own characteristics (such as low charging power, low power, etc., which are reflected in the implicit representation vector). Similarly, when it is necessary to provide power to external power users, based on the implicit representation vector, the power dispatching device can determine that lamp pole unit B is more suitable for providing power first due to its special connection relationship with local power users (reflected in the implicit representation vector). If the implicit representation vector is inaccurate, it may lead to unreasonable power dispatching. For example, a large amount of power may be mistakenly dispatched to lamp pole unit C, and C cannot effectively receive and utilize the power, resulting in power waste or equipment overload risk; or the special relationship between lamp pole unit B and local power users may be ignored, and power may not be provided to local power users in a timely manner, affecting the normal operation of the power users, and thus affecting the operating efficiency and reliability of the entire virtual power plant.

[0055] Step S600: Performing lamp pole unit power dispatch based on the implicit representation vector of each lamp pole unit in the distributed charging network graph.

[0056] The implicit representation vector of the light pole unit is obtained through a series of previous steps, which comprehensively reflects the various characteristics and relationships of the light pole unit in the distributed charging network map. For example, the implicit representation vector may contain information such as the charging capacity, power status, and degree of association with other light pole units of the light pole unit.

[0057] When dispatching electricity, the power dispatching equipment will make decisions based on the information in these implicit representation vectors. For example, consider the charging capacity of the lamp pole unit. Assume that the implicit representation vector of a lamp pole unit indicates that its upper limit of charging power is P max , the current charge is E, and its charging efficiency is If an external power source charges the virtual power plant, the power dispatching device will determine the charging power allocated to the lamp pole unit based on these parameters. For example, if the total charging power provided by the external power source is P total , and there are multiple lamp pole units that need to be charged, the power dispatching device can allocate the charging power in a certain proportion according to the charging capacity of each lamp pole unit (obtained through the implicit representation vector). A simple allocation method is to allocate according to the proportion of the upper limit of the charging power. Assume that there are n lamp pole units that need to be charged, and the upper limit of the charging power of the i-th lamp pole unit is , then the charging power allocated to the i-th lamp pole unit is The formula can be Calculated.

[0058] Consider the degree of association between lamp pole units. If the implicit representation vector of a lamp pole unit shows that it has a close power transmission association with other lamp pole units, for example, the implicit representation vector of lamp pole unit A shows that it has an efficient connection with lamp pole units B and C in power transmission. When lamp pole unit A has sufficient power and lamp pole unit B or C has insufficient power, the power dispatching equipment can prioritize the power transmission from lamp pole unit A to lamp pole unit B or C based on this degree of association. This power transmission decision also needs to take into account the loss during the power transmission process. Assume that the power transmission loss coefficient from lamp pole unit A to lamp pole unit B is , the current power of lamp pole unit A is , the power dispatching equipment must decide how much power to transfer from A to B These factors will be considered comprehensively when calculating the power consumption of the light pole unit B. , the power dispatching equipment may be based on To determine the amount of power transferred, is the current power of lamp pole unit B.

[0059] For the power status factor, when the virtual power plant needs to supply power to external power devices, the power dispatching device will give priority to the lamp pole units with sufficient power. For example, the implicit representation vector is used to filter out the units with power above a certain threshold. Assume that there are lamp pole units D, E, and F, and their power is ,if , ,and , then the power dispatching equipment will give priority to lamp pole units D and E in the initial power supply selection.

[0060] The power dispatching equipment can establish a power dispatching optimization model to implement the above decision-making process. This model can minimize power transmission losses, maximize power supply efficiency, or meet specific power requirements as the objective function. For example, the objective function can be minimize , where i and j represent different lamp pole units, is the power transmission loss coefficient from pole unit i to pole unit j, is the amount of electricity transmitted from i to j. At the same time, the model also needs to meet some constraints, such as the upper and lower limits of the electricity consumption of each lamp pole unit ( )、Power transmission capacity constraints( ,in is the maximum power transfer from pole unit i to pole unit j), etc.

[0061] Of course, other methods in the prior art may also be used to make decisions, and the present invention does not limit this.

[0062] As an implementation manner, step S200, obtaining a basic implicit representation vector of the first lamp pole unit, includes:

[0063] Step S210: determining adjacent lamp pole units of each lamp pole unit in the first lamp pole unit set based on the distributed charging network map;

[0064] Step S220: connecting the lamp pole units having the same adjacent lamp pole units in the first lamp pole unit set to construct a collaborative relationship graph corresponding to the first lamp pole unit set;

[0065] Step S230: mining graph unit coding features of the lamp pole units in the collaborative relationship graph based on the graph structure of the collaborative relationship graph, and determining a basic implicit representation vector of each lamp pole unit in the first lamp pole unit set.

[0066] In step S210, the power dispatching device determines the adjacent lamp pole units of each lamp pole unit in the first lamp pole unit set based on the distributed charging network map. In the distributed charging network map, there is a complex relationship network between the lamp pole units, and the adjacent lamp pole units refer to other lamp pole units that are directly connected to a specific lamp pole unit or are closely related in terms of power transmission, control logic, etc. For example, in an actual distributed charging network scenario, assume that there are lamp pole units A, B, C, D, etc. in the first lamp pole unit set. Lamp pole unit A is directly connected to lamp pole units B and C through power lines, and is associated with lamp pole unit D in power control logic because they are jointly controlled by a power management module in a certain area. Then for lamp pole unit A, lamp pole units B, C and D are its adjacent lamp pole units.

[0067] The power dispatching equipment can determine the adjacent lamp pole units by traversing the data structure of the distributed charging network graph. If the distributed charging network graph is represented as a graph structure G=(V, E), where V is a vertex set representing all lamp pole units, and E is an edge set representing the connection relationship between lamp pole units. For each lamp pole unit v∈V (here, the lamp pole units in the first lamp pole unit set are taken as an example), the power dispatching equipment checks the edge e∈E connected to v, and finds the corresponding vertices through the connection relationship of the edges. The lamp pole units represented by these vertices are the adjacent lamp pole units of v.

[0068] For step S220, the power dispatching device connects the lamp pole units with the same adjacent lamp pole units in the first lamp pole unit set to construct a collaborative relationship map corresponding to the first lamp pole unit set. This collaborative relationship map is a reconstruction of the internal relationship of the first lamp pole unit set, which highlights the connection between lamp pole units that have similarities in adjacent relationships. For example, continuing with the above lamp pole unit example, assume that although lamp pole unit A and lamp pole unit E are not directly connected, they are both adjacent lamp pole units to lamp pole units B, C and D. In this case, the power dispatching device will connect lamp pole unit A and lamp pole unit E to construct a collaborative relationship map.

[0069] In one implementation, the power dispatching device may establish a data structure to record the adjacent light pole unit information of each light pole unit. For each light pole unit in the first light pole unit set, their adjacent light pole unit sets are compared. If the adjacent light pole unit sets of two light pole units are exactly the same, a connection is established between them. For example, a hash table may be used to store the information of each light pole unit and its adjacent light pole units. For light pole unit i, its adjacent light pole unit information is stored as ,in Represents adjacent lamp pole units. The power dispatching device traverses all lamp pole units in the first lamp pole unit set, compares their hash table values, and establishes a connection if they are equal.

[0070] Then, in step S230, the power dispatching device performs graph unit coding feature mining on the lamp pole units in the collaborative relationship graph based on the graph structure of the collaborative relationship graph, and determines the basic implicit representation vector of each lamp pole unit in the first lamp pole unit set. Graph unit coding feature mining is a process of extracting useful information from the graph structure to generate a representation vector.

[0071] For example, suppose there are lamp pole units F, G, and H in the collaborative relationship graph. Lamp pole unit F is connected to both lamp pole units G and H, and is in a specific position in the graph structure, such as a key node connecting two subgraph structures. When mining the graph unit coding features, the power dispatching equipment will consider this connection relationship and location information of lamp pole unit F. For example, features can be generated based on factors such as the number of lamp pole units it connects to and the distance to other important nodes (such as the central node connecting multiple lamp pole units).

[0072] One method is based on the graph’s Laplacian Eigenmaps. For the collaborative relationship graph G = (V, E), define its Laplacian matrix L = DA, where D is the degree matrix (a diagonal matrix with diagonal elements d iiis equal to the degree of vertex i, that is, the number of edges connected to vertex i), A is the adjacency matrix (if vertex i and vertex j are connected, then a ij =1, otherwise a ij =0). Calculate the eigenvalues ​​and eigenvectors of the Laplace matrix L. Assume that the eigenvector corresponding to the smallest non-zero eigenvalue is , the elements of this feature vector can be used as part of the basic implicit representation vector of the light pole unit. For example, for light pole unit i, an element of its basic implicit representation vector can be .

[0073] Another method is to use the idea of ​​the DeepWalk algorithm. Perform random walks in the collaborative relationship graph to generate a series of node sequences. For example, starting from the lamp pole unit F, randomly select its adjacent lamp pole units for walks according to a certain probability, such as selecting lamp pole unit G with a probability of 0.5 and selecting lamp pole unit H with a probability of 0.5. After multiple walks, a series of node sequences are obtained, such as [F, G, H, F, H, …]. These node sequences are then input into a neural network model (such as the Skip-Gram model) for training. Assuming that the input node is x and the output node is y, the goal of the neural network is to maximize the probability of P(y|x). Through this training process, the hidden layer output of the neural network can be used as the basic implicit representation vector of the lamp pole unit.

[0074] As an implementation manner, step S220, connecting the lamp pole units having the same adjacent lamp pole units in the first lamp pole unit set to construct a collaborative relationship graph corresponding to the first lamp pole unit set, includes:

[0075] Step S221: connecting the lamp pole units having the same adjacent lamp pole units in the first lamp pole unit set to obtain a collaborative relationship graph to be processed;

[0076] Step S222: In the collaborative relationship map to be processed, the connection between two lamp pole units whose connection times are less than the set times is cut off to obtain a collaborative relationship map.

[0077] In step S221, the power dispatching device connects the lamp pole units with the same adjacent lamp pole units in the first lamp pole unit set to obtain a collaborative relationship map to be processed. In the distributed charging network map, each lamp pole unit has its adjacent lamp pole units, and these adjacent relationships constitute the local connection structure of the lamp pole units in the network. For example, suppose there are lamp pole units A, B, C, D, and E in the first lamp pole unit set. The adjacent lamp pole units of lamp pole unit A are F, G, and H; the adjacent lamp pole units of lamp pole unit B are F, G, and I; and the adjacent lamp pole units of lamp pole unit C are F, G, and J. Here, lamp pole units A, B, and C all have the same adjacent lamp pole units F and G. The power dispatching device will connect lamp pole units A, B, and C based on this same adjacent relationship. For other lamp pole units, such as D and E, if they have similar identical adjacent lamp pole unit situations with other lamp pole units, corresponding connection operations will also be performed to obtain a collaborative relationship map to be processed.

[0078] The power dispatching device can implement this operation by establishing a data structure. For example, create a two-dimensional array or matrix to represent the relationship between the lamp pole units in the first lamp pole unit set. Assume that this matrix is ​​R, and its number of rows and columns are equal to the number of lamp pole units in the first lamp pole unit set. For each lamp pole unit i, the power dispatching device traverses the entire first lamp pole unit set and compares the adjacent lamp pole units of lamp pole unit i with other lamp pole units j. If they have the same adjacent lamp pole units, set the values ​​of R[i][j] and R[j][i] to 1, indicating that there is a connection relationship between the two lamp pole units; if there are no identical adjacent lamp pole units, set them to 0. In this way, by filling the entire matrix, a connection relationship matrix representing the collaborative relationship graph to be processed can be obtained.

[0079] Next is step S222, in which the power dispatching equipment cuts off the connection between two lamp pole units whose number of connections is less than the set number in the collaborative relationship map to be processed, and obtains a collaborative relationship map. The set number is a predetermined value, which is used to filter out connection relationships with strong correlation in the collaborative relationship map. For example, in the collaborative relationship map to be processed obtained above, there may be some connection relationships that are established in step S221, but in fact these connection relationships may not be so important or stable due to the complexity of the data or some accidental factors. Assuming that the set number is 3, in the collaborative relationship map to be processed, the lamp pole units K and L are only connected twice, which may mean that the similarity or correlation between them is relatively low, not stable or important enough.

[0080] The power dispatching equipment can once again traverse the connection relationship matrix (assuming it is R) representing the collaborative relationship map to be processed. For each pair of light pole units i and j (i≠j), count the number of times R[i][j] and R[j][i] are 1 (i.e., the number of times they are connected). If this number is less than the set number, the values ​​of R[i][j] and R[j][i] are reset to 0, indicating that the connection between the two light pole units is cut off. Through such operations, after the entire matrix is ​​updated, the final collaborative relationship map is obtained.

[0081] Accurately constructing a collaborative relationship map is of great significance for subsequent operations. The construction of the collaborative relationship map to be processed in step S221 is based on the similarity analysis of the adjacent relationships of the lamp pole units. By connecting the lamp pole units with the same adjacent lamp pole units, the relationship between the lamp pole units with similar local structures can be highlighted to a certain extent. This helps to better mine the common features between these lamp pole units in subsequent operations. For example, if such a connection operation is not performed, when performing the map unit coding feature mining (such as the operation in step S230), the potential similarities between these lamp pole units may be ignored, so that this information cannot be effectively used to generate an accurate basic implicit representation vector.

[0082] The collaborative relationship map to be processed obtained in step S221 may contain some noise or less important connection relationships. In step S222, by setting a reasonable set number of times and cutting off those connection relationships with fewer connections, these less important relationships can be removed, so that the obtained collaborative relationship map can more concisely and accurately reflect the collaborative relationship between the lamp pole units. If there is no step S222, in the subsequent mining process of the map unit coding features, these less important connection relationships may interfere with the mining of the truly important features of the lamp pole units. For example, when generating the basic implicit representation vector, it may cause too much consideration of the impact of those accidental or less stable connection relationships, so that the basic implicit representation vector cannot accurately reflect the true characteristics and relationships of the lamp pole units in the distributed charging network. This will further affect the subsequent operations such as feature mapping of the lamp pole units based on the machine learning network, determining the involved relationships between the lamp pole units, and the final power dispatching decision, which may lead to unreasonable power dispatching and affect the operating efficiency and stability of the virtual power plant.

[0083] As an implementation method, the machine learning network includes a first network component, a second network component and an inference component. Based on this, the method also includes:

[0084] Step S1: determining adjacent lamp pole units of a first lamp pole unit and adjacent lamp pole units of a second lamp pole unit based on a distributed charging network map;

[0085] Step S2: Obtain basic implicit representation vectors of adjacent lamp pole units of the first lamp pole unit and implicit representation vectors of adjacent lamp pole units of the second lamp pole unit.

[0086] At this time, as a first implementation of step S300, that is, based on the machine learning network, feature mapping is performed on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit to obtain the inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, including:

[0087] Step S301: performing feature mapping on a basic implicit representation vector of a first lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the first lamp pole unit based on a first network component to determine a calibrated implicit representation vector of the first lamp pole unit;

[0088] Step S302: performing feature mapping on the basic implicit representation vector of the second lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the second lamp pole unit based on the second network component to determine the adjusted implicit representation vector of the second lamp pole unit;

[0089] Step S303: Reasoning the adjusted implicit representation vector of the first lamp pole unit and the adjusted implicit representation vector of the second lamp pole unit based on the reasoning component to determine the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph.

[0090] In step S1, the power dispatching device determines the adjacent lamp pole units of the first lamp pole unit and the adjacent lamp pole units of the second lamp pole unit based on the distributed charging network map. The distributed charging network map is a complex structure that contains a large number of lamp pole units and the relationship information between them. For the first lamp pole unit and the second lamp pole unit, determining their adjacent lamp pole units is the basis for in-depth analysis of their positional relationship and potential interaction in the network.

[0091] For example, in a distributed charging network diagram of a virtual power plant, assume that there are a first lamp pole unit A and a second lamp pole unit B. Lamp pole unit A may be directly connected to lamp pole units C, D, and E through power lines, and is associated with lamp pole unit F in terms of power control logic because they are jointly controlled by a power management module in a certain area. Then for the first lamp pole unit A, lamp pole units C, D, E, and F are its neighboring lamp pole units. Similarly, for the second lamp pole unit B, assuming that it is directly connected to lamp pole units G and H, and has a collaborative relationship with lamp pole unit I in terms of charging strategy, then lamp pole units G, H, and I are the neighboring lamp pole units of the second lamp pole unit B.

[0092] The power dispatching equipment can determine the adjacent lamp pole units by traversing the distributed charging network graph through the depth-first search (DFS) or breadth-first search (BFS) algorithm. If the distributed charging network graph is represented as a graph structure G=(V,E), where V is a vertex set representing all lamp pole units, and E is an edge set representing the connection relationship between lamp pole units. For the first lamp pole unit (represented by vertex v_1∈V), the power dispatching equipment can start from v_1 and use the DFS or BFS algorithm to search along the edge e∈E to find the lamp pole units represented by other vertices that are directly connected to it or associated with it through a specific logical relationship. These are the adjacent lamp pole units of the first lamp pole unit. The same method is used to process the second lamp pole unit.

[0093] Next is step S2, the power dispatching device obtains the basic implicit representation vector of the adjacent lamp pole unit of the first lamp pole unit and the implicit representation vector of the adjacent lamp pole unit of the second lamp pole unit. The basic implicit representation vector is a feature representation form of the lamp pole unit in the distributed charging network graph, which contains information about the various characteristics of the lamp pole unit and the relationship between it and other lamp pole units.

[0094] Continuing with the above example, assume that the basic implicit representation vector of lamp pole unit C (the adjacent lamp pole unit of the first lamp pole unit A) may contain its charging power range (such as the minimum charging power is , the maximum charging power is ), current power status (such as remaining power is ), the stability of the connection with other light pole units (can be represented by a numerical value express, , the higher the value, the more stable the connection) and other information. Similarly, for other adjacent light pole units such as D, E, F (the adjacent light pole units of the first light pole unit A) and G, H, I (the adjacent light pole units of the second light pole unit B), there are also corresponding basic implicit representation vectors.

[0095] The way for the power dispatching device to obtain these basic implicit representation vectors may be to read them from a pre-stored data structure. For example, there may be a database in which the basic implicit representation vector of each lamp pole unit is stored in a specific format, and the power dispatching device queries and obtains the corresponding basic implicit representation vector from the database according to the identification (such as number or name) of the lamp pole unit.

[0096] Then there are steps S301 - S303 in the first implementation of step S300 based on the above steps S1 - S2 .

[0097] In step S301, the power dispatching device performs feature mapping on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the first lamp pole unit based on the first network component to determine the adjusted implicit representation vector of the first lamp pole unit. The first network component is part of the machine learning network, and its function is to perform feature mapping operations on the input vector to mine deeper feature relationships.

[0098] For example, assuming that the basic implicit representation vector of the first lamp pole unit A is , the basic implicit representation vector of its adjacent light pole unit C is (Here n and m may or may not be equal, depending on the dimensionality of the representation vector.) The first network component can be a neural network, such as a simple multi-layer perceptron (MLP). As input, it undergoes nonlinear transformation through the hidden layer of MLP. Assuming that the weight matrix of MLP is W, the bias vector is b, and the neuron activation function of the hidden layer is f, then for the input vector (here represents concatenating two vectors into a longer vector), the output of the hidden layer The formula can be The output of this hidden layer is calculated. It can be used as part or all of the calibrated implicit representation vector of the first lamp pole unit A, depending on the design structure of the first network component. The purpose of this feature mapping is to integrate and transform the information in the basic implicit representation vector of the first lamp pole unit A and its adjacent lamp pole unit C to obtain a feature representation that better reflects the first lamp pole unit A considering the relationship between adjacent units, that is, the calibrated implicit representation vector.

[0099] In step S302, the power dispatching device performs feature mapping on the basic implicit representation vector of the second lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the second lamp pole unit based on the second network component to determine the adjusted implicit representation vector of the second lamp pole unit. Similar to step S301, the second network component is also part of the machine learning network and performs similar operations on the basic implicit representation vectors of the second lamp pole unit and its adjacent lamp pole units.

[0100] For example, for the second light pole unit B, assuming that its basic implicit representation vector is , the basic implicit representation vector of its adjacent light pole unit G is . Similarly, a neural network (such as MLP) is used as the second network component, and after a calculation process similar to step S301, and Feature mapping is performed to obtain a calibrated implicit representation vector of the second lamp pole unit B. This helps to more accurately represent the features of the second lamp pole unit B based on the relationship between the second lamp pole unit B and adjacent lamp pole units.

[0101] Finally, in step S303, the power dispatching device infers the calibrated implicit representation vector of the first lamp pole unit and the calibrated implicit representation vector of the second lamp pole unit based on the inference component to determine the inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph. The inference component is the part of the machine learning network used to perform relational inference based on the previously obtained calibrated implicit representation vector.

[0102] For example, assuming that the calibration implicit representation vector of the first lamp pole unit A is , the calibration implicit representation vector of the second lamp pole unit B is The reasoning component can use algorithms such as logistic regression model or support vector machine (SVM) for reasoning. Taking the logistic regression model as an example, assuming that the parameters of the model are (where t depends on the complexity of the model), for the input vector (Concatenate the two calibrated implicit representation vectors), calculate ,in is a logical function (such as ), the value of y can represent the probability of a certain specific involvement relationship between the first lamp pole unit A and the second lamp pole unit B. Based on this probability value, the inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network map can be determined. For example, if y > 0.5, it can be considered that there is a strong inference involvement relationship between them; if , the relationship is considered weak or non-existent.

[0103] On the basis of the above, step S1, determining the adjacent lamp pole units of the first lamp pole unit based on the distributed charging network map, may include:

[0104] Step S11: Determine the first-level adjacent lamp pole units of the first lamp pole unit in the distributed charging network map as the adjacent lamp pole units of the first lamp pole unit, wherein the adjacent lamp pole units of the first lamp pole unit are lamp pole units in the second lamp pole unit set; or, obtain the collaborative relationship map corresponding to the first lamp pole unit set, and determine the adjacent lamp pole units of the first lamp pole unit in the collaborative relationship map as the adjacent lamp pole units of the first lamp pole unit, wherein the adjacent lamp pole units of the first lamp pole unit are lamp pole units in the first lamp pole unit set.

[0105] In step S1 of the first implementation method based on the above-mentioned step S300, step S11 requires the power dispatching equipment to determine the first-level adjacent lamp pole unit of the first lamp pole unit in the distributed charging network map as the adjacent lamp pole unit of the first lamp pole unit, wherein the adjacent lamp pole unit of the first lamp pole unit is the lamp pole unit in the second lamp pole unit set; or, obtain the collaborative relationship map corresponding to the first lamp pole unit set, and determine the adjacent lamp pole unit of the first lamp pole unit in the collaborative relationship map as the adjacent lamp pole unit of the first lamp pole unit, wherein the adjacent lamp pole unit of the first lamp pole unit is the lamp pole unit in the first lamp pole unit set.

[0106] A first-level adjacent lamp pole unit refers to a lamp pole unit that is directly connected to the first lamp pole unit in the distributed charging network graph. For example, in a distributed charging network graph, the lamp pole units are regarded as nodes in the graph, and the edges represent the connection relationship between the lamp pole units. Assuming there is a first lamp pole unit A, if lamp pole units B, C, and D are directly connected to lamp pole unit A through edges, then lamp pole units B, C, and D are the first-level adjacent lamp pole units of lamp pole unit A. This direct connection relationship is the most basic case in determining the adjacent relationship of lamp pole units. From the perspective of power transmission, there may be the most direct possibility of power interaction between first-level adjacent lamp pole units. For example, power can be transmitted directly from one lamp pole unit to its first-level adjacent lamp pole unit, and this transmission may have higher efficiency or less loss because the connection between them is more direct.

[0107] When the adjacent lamp pole unit of the first lamp pole unit is a lamp pole unit in the second lamp pole unit set, the power dispatching device determines this first-level neighbor relationship in the distributed charging network graph. The power dispatching device can be implemented by traversing the data structure of the distributed charging network graph. If the distributed charging network graph is represented as a graph G=(V, E), where V is a vertex set representing all lamp pole units, and E is an edge set representing the connection relationship between lamp pole units. For the first lamp pole unit v∈V, the power dispatching device checks the edge e∈E connected to v, and finds the corresponding vertices through the connection relationship of the edges. The lamp pole units represented by these vertices are the first-level adjacent lamp pole units of v. For example, assuming that vertex v represents the first lamp pole unit A, the power dispatching device traverses all edges e related to v. If edge e connects vertex v and another vertex u, and the lamp pole unit represented by vertex u belongs to the second lamp pole unit set, then the lamp pole unit represented by vertex u is the first-level adjacent lamp pole unit of the first lamp pole unit A.

[0108] On the other hand, when the adjacent lamp pole units of the first lamp pole unit are to be determined in the collaborative relationship graph corresponding to the first lamp pole unit set, the collaborative relationship graph is a graph structure that reconstructs the internal relationship of the first lamp pole unit set. For example, suppose that there are lamp pole units A, B, C, D, etc. in the first lamp pole unit set. In the process of constructing the collaborative relationship graph, they may be reconnected according to certain similar attributes or relationships between the lamp pole units. When determining the adjacent lamp pole units of the first lamp pole unit A in this collaborative relationship graph, the power dispatching device can also adopt a traversal method. Assuming that the collaborative relationship graph is represented by G'=(V', E'), for the first lamp pole unit A (the corresponding vertex in V'), the power dispatching device traverses the edge e'∈E' connected to the vertex, finds the corresponding vertices through the edges, and the lamp pole units represented by these vertices are the adjacent lamp pole units of the first lamp pole unit A in the collaborative relationship graph, and these lamp pole units are the lamp pole units in the first lamp pole unit set.

[0109] As an implementation method, the machine learning network includes a first network component, a second network component and an inference component. Based on this, the method also includes:

[0110] Step S10: Acquire unit data of the first lamp pole unit and unit data of the second lamp pole unit;

[0111] Step S20: extracting characterization information from the unit data of the first lamp pole unit to obtain a first lamp pole unit representation vector;

[0112] Step S30: extracting characterization information from the unit data of the second lamp pole unit to obtain a second lamp pole unit representation vector.

[0113] At this time, as a second implementation of step S300, that is, based on the machine learning network, feature mapping is performed on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit to obtain the inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, including:

[0114] Step S310: performing feature mapping on the basic implicit representation vector of the first lamp pole unit and the first lamp pole unit representation vector based on the first network component to obtain the calibration data implicit representation vector of the first lamp pole unit;

[0115] Step S320: performing feature mapping on the basic implicit representation vector of the second lamp pole unit and the second lamp pole unit representation vector based on the second network component to obtain the calibration data implicit representation vector of the second lamp pole unit;

[0116] Step S330: Reasoning the implicit representation vector of the calibration data of the first lamp pole unit and the implicit representation vector of the calibration data of the second lamp pole unit based on the reasoning component to determine the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit.

[0117] In step S10, the power dispatching device obtains the unit data of the first light pole unit and the unit data of the second light pole unit. The unit data refers to the data content corresponding to the light pole unit, which contains various attribute information of the light pole unit. For example, for the first light pole unit A, its unit data may include data on physical attributes, such as the geographical location (latitude and longitude coordinates) of the light pole, assumed to be (x1, y1); hardware specification information of the light pole, such as the rated power P1 of the charging device, the capacity C1 of the battery, etc. For the second light pole unit B, its unit data also contains similar information, such as the geographical location (x2, y2), the rated power P2, the battery capacity C2, etc. These unit data are the basis for a comprehensive description of the light pole unit, and they reflect the basic characteristics of the light pole unit in the virtual power plant. The power dispatching device can obtain these unit data in a variety of ways, such as collecting data from the sensor equipment equipped with the light pole unit itself, these sensors can monitor the various states of the light pole in real time and transmit the data to the central database, and the power dispatching device reads the corresponding unit data from the database; or directly obtains the entered unit data from the pre-established light pole unit information management device.

[0118] Next is step S20, in which the power dispatching equipment extracts characterization information from the unit data of the first lamp pole unit to obtain the first lamp pole unit representation vector. Characterization information extraction is the process of extracting key information from unit data and converting it into a vector form, which can be better used for subsequent calculations and analysis. Taking the first lamp pole unit A as an example, assume that the geographic location (x1, y1) in its unit data can be converted into two elements in a vector through a certain mapping method, such as mapping x1 to the first element of the vector and y1 to the second element of the vector; the rated power P1 can become the third element of the vector after normalization, and the battery capacity C1 can become the fourth element of the vector after a specific conversion (such as based on the proportional relationship between the battery capacity and a certain benchmark capacity), etc. In this way, the representation vector of the first lamp pole unit A is obtained, which is assumed to be . Electric power dispatching equipment can use a variety of technical means to extract representation information. A common method is rule-based conversion, that is, converting different types of unit data into vector elements according to pre-set rules. For example, for geographic location information, rules can be set to convert longitude and latitude into vector elements according to certain mathematical operations; for numerical unit data, such as rated power and battery capacity, they can be normalized or proportionally converted according to their value range and data distribution. Another method is to use feature engineering techniques in machine learning, such as using principal component analysis (PCA) to reduce the dimension and extract features of unit data, and convert the original unit data into a representation vector that better represents its characteristics.

[0119] Then, in step S30, the power dispatching device extracts the characterization information of the unit data of the second light pole unit to obtain the second light pole unit representation vector. This process is similar to step S20. For the second light pole unit B, its unit data (such as geographic location (x2, y2), rated power P2, battery capacity C2, etc.) is converted in a similar way to obtain the second light pole unit representation vector .

[0120] Based on the above steps S10 - S30 , the process proceeds to steps S310 - S330 in the second implementation of step S300 .

[0121] In step S310, the electric power dispatching device performs feature mapping on the basic implicit representation vector of the first lamp pole unit and the first lamp pole unit representation vector based on the first network component to obtain the calibration data implicit representation vector of the first lamp pole unit. As a part of the machine learning network, the first network component has the function of performing feature mapping on the input vector. The basic implicit representation vector of the first lamp pole unit is a pre-characterization result of the first lamp pole unit characteristics that has been obtained before, and the first lamp pole unit representation vector is a new representation vector obtained by extracting the characterization information of the unit data. Combining the two for feature mapping can more comprehensively mine the characteristics of the first lamp pole unit. For example, assuming that the basic implicit representation vector of the first lamp pole unit A is , the first lamp pole unit represents the vector (Here n and m may be different, depending on the dimensionality of the respective vectors.) The first network component can be a neural network, such as a multi-layer perceptron (MLP). and As input, assuming that the weight matrix of the MLP is W, the bias vector is b, and the neuron activation function of the hidden layer is f, for the input vector (concatenate two vectors into a longer vector), the output of the hidden layer The formula can be The output of this hidden layer is calculated. It can be used as part or all of the implicit representation vector of the calibration data of the first lamp pole unit A, depending on the design structure of the first network component. The purpose of this feature mapping is to integrate the information in the basic implicit representation vector and the representation vector directly extracted from the unit data to obtain the implicit representation vector of the calibration data that more accurately and comprehensively reflects the characteristics of the first lamp pole unit A.

[0122] In step S320, the power dispatching device performs feature mapping on the basic implicit representation vector of the second lamp pole unit and the second lamp pole unit representation vector based on the second network component to obtain the calibration data implicit representation vector of the second lamp pole unit. Similar to step S310, for the second lamp pole unit B, assuming that its basic implicit representation vector is , the second lamp pole unit represents the vector . Use a second network component (which can also be a similar neural network) to and For example, after a calculation process similar to step S310, and The input is fed into the neural network, and the output of the hidden layer is obtained through the weight matrix, bias vector and activation function. This output is the implicit representation vector of the calibration data of the second light pole unit B. This helps to comprehensively consider the basic features of the second light pole unit B (basic implicit representation vector) and the direct features extracted from the unit data (second light pole unit representation vector), so as to obtain the calibration data implicit representation vector that can more accurately reflect the features of the second light pole unit B.

[0123] Finally, in step S330, the power dispatching device infers the implicit representation vector of the calibration data of the first lamp pole unit and the implicit representation vector of the calibration data of the second lamp pole unit based on the inference component to determine the inference relationship between the first lamp pole unit and the second lamp pole unit. The inference component is the part of the machine learning network that is used to perform relational inference based on the implicit representation vector of the calibration data obtained previously. For example, assuming that the implicit representation vector of the calibration data of the first lamp pole unit A is , the implicit representation vector of the calibration data of the second lamp pole unit B is The reasoning component can use a variety of algorithms to perform reasoning, such as logistic regression model or support vector machine (SVM). Taking the logistic regression model as an example, assuming that the model parameters are (where t depends on the complexity of the model), for the input vector (Concatenate the two calibration data implicit representation vectors), calculate ,in is a logical function (such as ), the value of y can represent the probability of a certain specific involvement relationship between the first lamp pole unit A and the second lamp pole unit B. According to this probability value, the inference involvement relationship between the first lamp pole unit and the second lamp pole unit can be determined. For example, if y > 0.5, it can be considered that there is a strong inference involvement relationship between them; if y ≤ 0.5, it is considered that the relationship is weak or does not exist.

[0124] As an implementation mode, the distributed charging network map further includes a third lamp pole unit set, the third lamp pole unit set includes a third lamp pole unit, and the machine learning network includes a first network component, a second network component, a third network component, and an inference component. Based on this, the method further includes:

[0125] Step SA: obtaining a basic implicit representation vector of the third lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the third lamp pole unit, wherein the basic implicit representation vector of the third lamp pole unit is obtained according to the involvement relationship of the lamp pole units in the third lamp pole unit set.

[0126] At this time, as a third implementation of step S300, that is, based on the machine learning network, feature mapping is performed on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit to obtain the inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, including:

[0127] Step S300A: performing feature mapping on a basic implicit representation vector of a first lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the first lamp pole unit based on a first network component to obtain a calibrated implicit representation vector of the first lamp pole unit;

[0128] Step S300B: performing feature mapping on the basic implicit representation vector of the second lamp pole unit and the basic implicit representation vectors of adjacent lamp pole units of the second lamp pole unit based on the second network component to obtain a calibrated implicit representation vector of the second lamp pole unit;

[0129] Step S300C: performing feature mapping on the basic implicit representation vector of the third lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the third lamp pole unit based on the third network component to obtain a calibrated implicit representation vector of the third lamp pole unit;

[0130] Step S300D: The reasoning component performs reasoning based on the calibrated implicit representation vector of the first lamp pole unit, the calibrated implicit representation vector of the second lamp pole unit and the calibrated implicit representation vector of the third lamp pole unit to obtain the inferred involvement relationship among the first lamp pole unit, the second lamp pole unit and the third lamp pole unit; the real involvement relationship also includes the real involvement relationship among the first lamp pole unit, the second lamp pole unit and the third lamp pole unit in the distributed charging network map.

[0131] In step SA, the power dispatching device obtains the basic implicit representation vector of the third lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the third lamp pole unit, wherein the basic implicit representation vector of the third lamp pole unit is obtained according to the involvement relationship of the lamp pole units in the third lamp pole unit set.

[0132] The basic implicit representation vector is a characteristic characterization result of the lamp pole unit in the distributed charging network graph. For the third lamp pole unit, it reflects the characteristics of the lamp pole unit in the third lamp pole unit set to which it belongs and the relationship between the lamp pole unit and other lamp pole units in the set. For example, assuming that in a distributed charging network of a virtual power plant, the third lamp pole unit set contains multiple lamp pole units, where the basic implicit representation vector of the third lamp pole unit C may contain information related to its charging efficiency, power interaction capabilities with other lamp pole units in the same set, etc. If the third lamp pole unit C has a higher charging efficiency than other lamp pole units in its set, this characteristic will be reflected in its basic implicit representation vector in a certain numerical form.

[0133] The basic implicit representation vectors of the adjacent lamp pole units of the third lamp pole unit are equally important. These adjacent lamp pole units have direct or indirect relationships with the third lamp pole unit C, and their basic implicit representation vectors contain their respective characteristic information. For example, the basic implicit representation vector of the lamp pole unit D adjacent to the third lamp pole unit C may contain information such as its own power reserve and the stability of its connection with the external power grid. The way in which the power dispatching equipment obtains these basic implicit representation vectors may be through previous analysis and calculation of the distributed charging network map, or reading from a database that specifically stores these data. For example, there may be a data storage device that stores the basic implicit representation vectors of each lamp pole unit and its adjacent lamp pole units according to the identification (such as the number) of the lamp pole unit, and the power dispatching equipment queries and obtains the basic implicit representation vectors of itself and its adjacent lamp pole units based on the identification of the third lamp pole unit.

[0134] Based on step SA, the process proceeds to steps S300A-S300D in the third implementation manner of step S300.

[0135] In step S300A, the power dispatching device performs feature mapping on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the first lamp pole unit based on the first network component to obtain the adjusted implicit representation vector of the first lamp pole unit. As a part of the machine learning network, the first network component performs feature mapping operation on the input vector.

[0136] For example, let the first lamp pole unit be A, and its basic implicit representation vector is , the adjacent lamp pole units of the first lamp pole unit A are assumed to be E, F, and G, and their basic implicit representation vectors are , , . Combine these vectors as the input vector, assuming that the first network component is a multi-layer perceptron (MLP) with a weight matrix W, a bias vector b, and an activation function f. For the input vector (Here, each vector is concatenated into a longer vector), by the formula Calculate the output of the hidden layer ,this It can be used as part or all of the calibrated implicit representation vector of the first lamp pole unit A, depending on the structural design of the first network component. This feature mapping process can comprehensively consider the characteristics of the first lamp pole unit A and its adjacent lamp pole units, and mine deeper feature relationships, thereby obtaining a calibrated implicit representation vector that more accurately reflects the state and relationship of the first lamp pole unit A in the network.

[0137] In step S300B, the power dispatching device performs feature mapping on the basic implicit representation vector of the second lamp pole unit and the basic implicit representation vector of the adjacent lamp pole units of the second lamp pole unit based on the second network component to obtain the adjusted implicit representation vector of the second lamp pole unit. Similar to step S300A, let the second lamp pole unit be B, and its basic implicit representation vector be , the adjacent light pole units are assumed to be H, I, and J, and their basic implicit representation vectors are These vectors are input into the second network component (which can also be a similar MLP), and after a calculation process similar to step S300A, the adjusted implicit representation vector of the second lamp pole unit B is obtained. This helps to more comprehensively adjust the characteristics of the second lamp pole unit B based on the consideration of the second lamp pole unit B and its adjacent lamp pole units, so that the obtained adjusted implicit representation vector can better reflect its characteristics and relationships in the distributed charging network.

[0138] In step S300C, the power dispatching device performs feature mapping on the basic implicit representation vector of the third lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the third lamp pole unit based on the third network component to obtain the adjusted implicit representation vector of the third lamp pole unit. For example, the basic implicit representation vector of the third lamp pole unit C is , the basic implicit representation vector of its adjacent light pole unit D (mentioned earlier) is . Feature mapping is performed through a third network component (such as MLP), and the input vector (composed of the basic implicit representation vectors of the third lamp pole unit C and its adjacent lamp pole units) is processed in a similar calculation method, that is, according to the weight matrix, bias vector and activation function, to obtain the adjusted implicit representation vector of the third lamp pole unit C. This adjusted implicit representation vector can more accurately represent the characteristics of the third lamp pole unit C in the distributed charging network based on the comprehensive consideration of the relationship between the third lamp pole unit C and its adjacent lamp pole units.

[0139] In step S300D, the reasoning component performs reasoning based on the adjusted implicit representation vector of the first lamp pole unit, the adjusted implicit representation vector of the second lamp pole unit, and the adjusted implicit representation vector of the third lamp pole unit to obtain the reasoning involvement relationship of the first lamp pole unit, the second lamp pole unit, and the third lamp pole unit. The reasoning component is a part of the machine learning network used to perform relationship reasoning based on the adjusted implicit representation vector obtained previously.

[0140] Assume that the calibration implicit representation vector of the first lamp pole unit A is , the calibration implicit representation vector of the second lamp pole unit B is , the implicit representation vector of the third lamp pole unit C is The reasoning component can use different algorithms to perform reasoning, such as algorithms based on probabilistic models.

[0141] Taking the Bayesian network as an example, assume that the reasoning involvement relationship between the three lamp pole units can be represented by nodes and edges in the Bayesian network. The nodes represent the first lamp pole unit A, the second lamp pole unit B and the third lamp pole unit C, respectively, and the edges represent the relationship between them. According to the Bayesian theorem, P(A,B,C)=P(A|B,C)P(B|C)P(C), where P(A,B,C) represents the joint probability distribution of the three lamp pole units, P(A|B,C) represents the conditional probability of A when B and C are known, P(B|C) represents the conditional probability of B when C is known, and P(C) represents the prior probability of C. These probability parameters in the Bayesian network can be determined by learning historical data or prior knowledge. The power dispatching equipment inputs the information in the calibrated implicit representation vector as evidence into the Bayesian network, and calculates the probability values ​​of different relationships between the first lamp pole unit, the second lamp pole unit and the third lamp pole unit. Based on these probability values, the reasoning involvement relationship between them can be determined. For example, if it is calculated that under certain conditions, the probability of the first lamp pole unit A transmitting power to the second lamp pole unit B and the third lamp pole unit C is high, then the inference involvement relationship of the power transmission direction can be determined.

[0142] Based on the third implementation of step S300 above, at this time, step S400, based on the real involvement relationship and the inferred involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network map, the machine learning network is adjusted to obtain a calibrated machine learning network; based on the calibrated machine learning network, the implicit representation vector of each lamp pole unit in the distributed charging network map is determined, including:

[0143] Step S410: Based on the real involvement relationship and inferred involvement relationship of the first lamp pole unit, the second lamp pole unit and the third lamp pole unit in the distributed charging network map, the machine learning network is calibrated to determine the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network.

[0144] In step S410 of the implementation of step S400 under the third implementation based on step S300, the power dispatching equipment calibrates the machine learning network based on the real involvement relationship and the inferred involvement relationship of the first lamp pole unit, the second lamp pole unit and the third lamp pole unit in the distributed charging network map, so as to determine the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network.

[0145] The real involvement relationship is the actual relationship between the first light pole unit, the second light pole unit and the third light pole unit in the distributed charging network map, which is based on multiple factors such as physical connection, power transmission logic, and device performance. For example, in the actual distributed charging network, the first light pole unit A, the second light pole unit B and the third light pole unit C may have a direct power transmission line between A and B due to their geographical proximity, and the transmission capacity of the line is a fixed value, which is a real involvement relationship; at the same time, C may have an indirect relationship with A through a power regulation device, which is also part of the real involvement relationship.

[0146] The inference-involved relationship is a relationship inferred by the power dispatching device in the third implementation mode of step S300 based on the adjusted implicit representation vectors of the first lamp pole unit, the second lamp pole unit and the third lamp pole unit through the inference component. It is an estimated relationship based on data and models, which may be different from the real involved relationship.

[0147] The power dispatching equipment adjusts the machine learning network so that it can better reflect the real relationship between the lamp pole units. The adjustment process is essentially an optimization process. For example, the power dispatching equipment can use the gradient descent algorithm to adjust the parameters in the machine learning network (refer to the previous description, which will not be repeated here).

[0148] As an implementation mode, the distributed charging network map further includes a fourth lamp pole unit set, the fourth lamp pole unit set includes a fourth lamp pole unit, and the machine learning network includes a first network component, a second network component, a fourth network component and an inference component. Based on this, the method further includes:

[0149] Step S1000: obtaining a basic implicit representation vector of a fourth lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the fourth lamp pole unit, wherein the basic implicit representation vector of the fourth lamp pole unit is obtained according to the involvement relationship of the lamp pole units in the fourth lamp pole unit set;

[0150] Step S2000: performing feature mapping on a basic implicit representation vector of a second lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the second lamp pole unit based on a second network component to obtain a calibrated implicit representation vector of the second lamp pole unit;

[0151] Step S3000: performing feature mapping on a basic implicit representation vector of a fourth lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the fourth lamp pole unit based on a fourth network component to obtain a calibrated implicit representation vector of the fourth lamp pole unit;

[0152] Step S4000: Reasoning the adjusted implicit representation vector of the second lamp pole unit and the adjusted implicit representation vector of the fourth lamp pole unit based on the reasoning component to obtain the reasoning involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network graph;

[0153] Step S5000: determining the real involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network map based on the distributed charging network map;

[0154] Step S6000: Based on the real involvement relationship and the inferred involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network map, the machine learning network is calibrated to determine the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network.

[0155] In step S1000, the power dispatching device obtains a basic implicit representation vector of a fourth lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the fourth lamp pole unit, wherein the basic implicit representation vector of the fourth lamp pole unit is obtained based on the involvement relationship of lamp pole units in a fourth lamp pole unit set.

[0156] The basic implicit representation vector is a characteristic representation of the light pole unit in the distributed charging network graph. It contains information about the various characteristics of the light pole unit and its relationship with other light pole units. For the fourth light pole unit in the fourth light pole unit set, its basic implicit representation vector reflects the specific state and relationship of the unit in the set to which it belongs. For example, suppose that in a large virtual power plant, the fourth light pole unit set is distributed in a specific area, and the fourth light pole unit D is one of the units. Its basic implicit representation vector may contain information about its charging power range. If its charging power range is from 10 kilowatts to 20 kilowatts, this information will be reflected in the basic implicit representation vector in a certain numerical form; it may also contain information about the power transmission stability between it and other light pole units in the set. If its power transmission stability with a neighboring light pole unit is high, it will also be reflected in the vector.

[0157] The basic implicit representation vectors of the adjacent lamp pole units of the fourth lamp pole unit are equally important. These adjacent lamp pole units have direct or indirect relationships with the fourth lamp pole unit D, and their basic implicit representation vectors contain their respective characteristic information. For example, the basic implicit representation vector of the lamp pole unit E adjacent to the fourth lamp pole unit D may contain its own power reserve. Assuming that its current power reserve is 50% of the rated power, this information is in its basic implicit representation vector; it may also contain information about its connection status with the external power grid. If it is a stable connection, it will also exist in the vector in a specific numerical or coded form. The way in which the power dispatching equipment obtains these basic implicit representation vectors is to read them from the data structure obtained in the previous construction and analysis of the distributed charging network map, or to query and obtain them from a database that specifically stores these vector information.

[0158] Next is step S2000, where the power dispatching device performs feature mapping on the basic implicit representation vector of the second lamp pole unit and the basic implicit representation vectors of adjacent lamp pole units of the second lamp pole unit based on the second network component to obtain a calibrated implicit representation vector of the second lamp pole unit.

[0159] The second network component is part of the machine learning network, and its function is to perform feature mapping operations on the input vector. Assume that the second light pole unit is B, and its basic implicit representation vector is , its adjacent light pole units are assumed to be F, G, and H, and their basic implicit representation vectors are , , . Combine these vectors as the input vector, assuming that the second network component is a multi-layer perceptron (MLP) with a weight matrix W, a bias vector b, and an activation function f. For the input vector (Here, each vector is concatenated into a longer vector), by the formula Calculate the output of the hidden layer ,this It can be used as part or all of the calibrated implicit representation vector of the second light pole unit B, depending on the structural design of the second network component. This feature mapping process can comprehensively consider the characteristics of the second light pole unit B and its adjacent light pole units, and mine deeper feature relationships, thereby obtaining a calibrated implicit representation vector that more accurately reflects the state and relationship of the second light pole unit B in the network.

[0160] Then comes step S3000, where the power dispatching device performs feature mapping on the basic implicit representation vector of the fourth lamp pole unit and the basic implicit representation vectors of adjacent lamp pole units of the fourth lamp pole unit based on the fourth network component to obtain a calibrated implicit representation vector of the fourth lamp pole unit.

[0161] Similar to step S2000, let the basic implicit representation vector of the fourth lamp pole unit D be , the basic implicit representation vector of its adjacent light pole unit E (mentioned above) is . The fourth network component (such as MLP) performs feature mapping and processes the input vector (composed of the basic implicit representation vectors of the fourth light pole unit D and its adjacent light pole units) in a similar calculation manner, i.e., according to the weight matrix, bias vector and activation function, to obtain the adjusted implicit representation vector of the fourth light pole unit D. This adjusted implicit representation vector can more accurately represent the characteristics of the fourth light pole unit D in the distributed charging network based on the comprehensive consideration of the relationship between the fourth light pole unit D and its adjacent light pole units.

[0162] In step S4000, the power dispatching device infers the calibrated implicit representation vector of the second lamp pole unit and the calibrated implicit representation vector of the fourth lamp pole unit based on the inference component to obtain the inference involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network graph.

[0163] The reasoning component is the part of the machine learning network that is used to perform relational reasoning based on the previously obtained calibrated implicit representation vector. Assume that the calibrated implicit representation vector of the second lamp pole unit B is , the implicit representation vector of the fourth lamp pole unit D is The reasoning component can use different algorithms to perform reasoning, such as algorithms based on probabilistic models.

[0164] Taking the naive Bayes classifier as an example, suppose the relationship between the second light pole unit B and the fourth light pole unit D is regarded as a classification problem, for example, the relationship is divided into categories such as "strong association", "weak association" and "no association". According to the Bayesian theorem , where P(C|X) is the feature X (here is and The posterior probability of category C (such as "strong association" and other relationship categories) under the feature vector composed of P(X|C), P(X|C) is the conditional probability of feature X under category C, P(C) is the prior probability of category C, and P(X) is the probability of feature X. These probability parameters can be determined by learning historical data or prior knowledge. The power dispatching equipment inputs the information in the calibrated implicit representation vector as evidence into the naive Bayes classifier, and calculates the probability values ​​of different relationships between the second light pole unit and the fourth light pole unit. Based on these probability values, the inference-involved relationship between them can be determined. For example, if the calculated probability of "strong association" is the highest, then it can be determined that there is a strong inference-involved relationship between the second light pole unit B and the fourth light pole unit D.

[0165] In step S5000, the power dispatching device determines the actual involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network map based on the distributed charging network map.

[0166] In the distributed charging network graph, the real involvement relationship is a relationship determined by various actual factors such as physical connection, power transmission logic, and equipment performance. For example, there may be a power transmission line between the second light pole unit B and the fourth light pole unit D in the graph, but this line may have certain power loss characteristics due to the long distance or passing through some power regulation equipment. This is a real involvement relationship; or they may be controlled by the same power management module due to their position relationship in the power grid topology, which is also part of the real involvement relationship. The power dispatching equipment can determine this real involvement relationship through in-depth analysis of the distributed charging network graph, for example, by checking the connection edges, node attributes, and related power equipment rules in the graph.

[0167] Finally, in step S6000, the power dispatching equipment calibrates the machine learning network based on the real involvement relationship and inferred involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network map, so as to determine the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network.

[0168] The purpose of the power dispatching equipment to adjust the machine learning network is to enable the network to better reflect the real relationship between the lamp pole units. The adjustment process can adopt a variety of methods, such as the aforementioned adjustment of network parameters based on the gradient descent algorithm.

[0169] As an implementation method, the machine learning network includes a first network component and a second network component. Based on this, step S500, based on the calibrated machine learning network, determines the implicit representation vector of each lamp pole unit in the distributed charging network map, including:

[0170] Step S510: performing feature mapping on the basic implicit representation vector of the first lamp pole unit based on the first network component of the calibrated machine learning network to obtain the implicit representation vector of the first lamp pole unit;

[0171] Step S520: Perform feature mapping on the basic implicit representation vector of the second lamp pole unit based on the second network component of the calibrated machine learning network to obtain the implicit representation vector of the second lamp pole unit.

[0172] In step S510, the electric power dispatching device performs feature mapping on the basic implicit representation vector of the first lamp pole unit based on the first network component of the calibrated machine learning network to obtain the implicit representation vector of the first lamp pole unit.

[0173] The calibrated machine learning network is a network that has been optimized through a series of previous steps (such as calibration based on real involvement relationships and inferred involvement relationships, etc.), and can more accurately reflect the relationship between lamp pole units. The first network component is a part of this machine learning network, which has a specific function to process input vectors and perform feature mapping.

[0174] The basic implicit representation vector of the first lamp pole unit is a preliminary characterization result of the first lamp pole unit characteristics obtained previously. It contains some characteristic information of the first lamp pole unit in the distributed charging network map, but it may not be complete or accurate enough. For example, assuming that the first lamp pole unit is A, its basic implicit representation vector It may contain some basic attribute information, such as the initial estimated charge value of A. , basic power parameters of its charging equipment And a simple relationship value with a few adjacent light pole units wait.

[0175] When the electric power dispatching equipment uses the first network component to When performing feature mapping, this process aims to dig out deeper and more comprehensive feature information, so as to obtain an implicit representation vector that can more accurately represent the first lamp pole unit A. Assume that the first network component is a neural network (such as a multi-layer perceptron MLP), which has a weight matrix W, a bias vector b and an activation function f. The power dispatching device will As the input vector, it goes through the MLP calculation process. Specifically, for the input vector ,calculate , here This is the result after feature mapping. This result may be a new vector, such as , which is compared to the original basis implicit representation vector , which may have changed in dimension (n and m may be different), and whose elements Contains feature information mined and converted by the first network component.

[0176] For example, the original basis implicitly represents the initial charge estimate x in the vector A1 After feature mapping, h A1 It may no longer be just a simple power estimation, but a more representative power status indicator after combining other relevant factors in the entire distributed charging network map. This indicator may take into account factors such as the power status of other light pole units connected to the first light pole unit A, the power demand trend of the entire area, and the overall load of the power grid. Similarly, for the basic power parameter x of the charging equipment A2 , h after feature mapping A2 It may reflect the effective power range actually available to the first lamp pole unit A under the current network status, and this range may be dynamically adjusted according to the charging and discharging requirements of other lamp pole units in the network.

[0177] Next comes step S520, where the power dispatching device performs feature mapping on the basic implicit representation vector of the second lamp pole unit based on the second network component of the calibrated machine learning network to obtain the implicit representation vector of the second lamp pole unit.

[0178] Similar to step S510, for the second lamp pole unit (set as B), its basic implicit representation vector Contains some basic characteristic information of the second lamp pole unit B. The second network component is also part of the tuned machine learning network, which Perform similar feature mapping operations.

[0179] Assume that the second network component is also a similar neural network structure and uses the same calculation method, that is, for the input vector , calculated by the weight matrix W' (which may be different from the weight matrix in the first network component, reflecting the different feature mapping logic), the bias vector b' and the activation function f' , here It is the implicit representation vector of the second lamp pole unit B after feature mapping.

[0180] For example, the elements in the basis implicit representation vector of the second light pole unit B are It may represent the geographical location information of B (such as the converted value of longitude and latitude coordinates), after the feature mapping of the second network component, It is possible to combine the power supply and demand characteristics of different regions in the distributed charging network and transform it into an indicator that represents the relative importance of B in the entire power dispatching network. If B is located in the center of a region with high power demand, then The value of may reflect its criticality in meeting the surrounding power demand. If it represents a certain performance parameter of B's ​​charging equipment, after feature mapping, The performance distribution of other similar devices in the network may be taken into account, thus indicating the relative performance advantage or disadvantage of B's ​​charging device in the entire network.

[0181] Figure 2 A hardware entity diagram of a power dispatching device provided by an embodiment of the present invention, such as Figure 2 As shown, the hardware entity of the power dispatching device 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can be run on the processor 1001, and the processor 1001 implements the steps in the method of any of the above embodiments when executing the program.

[0182] The memory 1002 stores a computer program that can be run on the processor. The memory 1002 is configured to store instructions and applications executable by the processor 1001. It can also cache data to be processed or processed by the processor 1001 and each module in the power dispatching device 1000 (for example, image data, audio data, voice communication data, and video communication data). This can be implemented through flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0183] When the processor 1001 executes the program, the steps of any of the above-mentioned methods for dispatching power of a virtual power plant are implemented. The processor 1001 generally controls the overall operation of the power dispatching device 1000 .

[0184] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0185] The above description is only an implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A power dispatching method for a virtual power plant, characterized in that: include: Obtain a distributed charging network map, the distributed charging network map including a first lamp pole unit set and a second lamp pole unit set, the first lamp pole unit set and the second lamp pole unit set do not have an intersection, the first lamp pole unit set includes a first lamp pole unit, the second lamp pole unit set includes a second lamp pole unit, wherein the lamp pole unit is used to indicate data of charging lamp poles in a virtual power plant, the virtual power plant is composed of charging lamp poles arranged on both sides of a road; the distributed charging network map includes a real involvement relationship between the first lamp pole unit and the second lamp pole unit; Obtaining a basic implicit representation vector of the first lamp pole unit and a basic implicit representation vector of the second lamp pole unit, wherein the basic implicit representation vector of the first lamp pole unit is obtained according to the involvement relationship between the lamp pole units in the first lamp pole unit set, and the basic implicit representation vector of the second lamp pole unit is obtained according to the involvement relationship between the lamp pole units in the second lamp pole unit set; Performing feature mapping on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit based on a machine learning network to obtain an inference involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph; Based on the real involvement relationship and the inferred involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, the machine learning network is calibrated to obtain a calibrated machine learning network; Determine an implicit representation vector of each lamp pole unit in the distributed charging network graph based on a calibrated machine learning network; Performing lamp pole unit power dispatching based on the implicit representation vector of each lamp pole unit in the distributed charging network graph; The machine learning network includes a first network component, a second network component and an inference component; and the method further includes: Determine adjacent lamp pole units of the first lamp pole unit and adjacent lamp pole units of the second lamp pole unit based on the distributed charging network map; Obtaining basic implicit representation vectors of adjacent lamp pole units of the first lamp pole unit and implicit representation vectors of adjacent lamp pole units of the second lamp pole unit; Among them, feature mapping is performed on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit based on the machine learning network to obtain the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, including: Performing feature mapping on a basic implicit representation vector of the first lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the first lamp pole unit based on the first network component to determine a calibrated implicit representation vector of the first lamp pole unit; Performing feature mapping on a basic implicit representation vector of the second lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the second lamp pole unit based on the second network component to determine a calibrated implicit representation vector of the second lamp pole unit; Based on the reasoning component, the adjusted implicit representation vector of the first lamp pole unit and the adjusted implicit representation vector of the second lamp pole unit are inferred to determine the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network map.

2. The method according to claim 1, characterized in that: Obtaining a basic implicit representation vector of the first lamp pole unit includes: Determine adjacent lamp pole units of each lamp pole unit in the first lamp pole unit set based on the distributed charging network graph; Connecting lamp pole units having the same adjacent lamp pole units in the first lamp pole unit set to construct a collaborative relationship graph corresponding to the first lamp pole unit set; Based on the graph structure of the collaborative relationship graph, graph unit coding feature mining is performed on the lamp pole units in the collaborative relationship graph to determine the basic implicit representation vector of each lamp pole unit in the first lamp pole unit set.

3. The method according to claim 2, characterized in that: Connecting lamp pole units having the same adjacent lamp pole units in the first lamp pole unit set to construct a collaborative relationship graph corresponding to the first lamp pole unit set includes: Connecting the lamp pole units having the same adjacent lamp pole units in the first lamp pole unit set to obtain a collaborative relationship graph to be processed; In the collaborative relationship map to be processed, the connection between two lamp pole units whose connection times are less than a set number is cut off to obtain the collaborative relationship map.

4. The method according to claim 1, characterized in that: Determining adjacent lamp pole units of the first lamp pole unit based on the distributed charging network map includes: In the distributed charging network graph, the first-level adjacent lamp pole unit of the first lamp pole unit is determined as the adjacent lamp pole unit of the first lamp pole unit, wherein the adjacent lamp pole unit of the first lamp pole unit is a lamp pole unit in the second lamp pole unit set; or, a collaborative relationship graph corresponding to the first lamp pole unit set is obtained, and the adjacent lamp pole unit of the first lamp pole unit is determined in the collaborative relationship graph as the adjacent lamp pole unit of the first lamp pole unit, wherein the adjacent lamp pole unit of the first lamp pole unit is a lamp pole unit in the first lamp pole unit set.

5. The method according to claim 1, characterized in that: The machine learning network includes a first network component, a second network component and an inference component; wherein the method further includes: Acquire unit data of the first lamp pole unit and unit data of the second lamp pole unit; Extracting characterization information from the unit data of the first lamp pole unit to obtain a first lamp pole unit representation vector; Extracting characterization information from the unit data of the second lamp pole unit to obtain a second lamp pole unit representation vector; Among them, feature mapping is performed on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit based on the machine learning network to obtain the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, including: Performing feature mapping on the basic implicit representation vector of the first lamp pole unit and the first lamp pole unit representation vector based on the first network component to obtain an implicit representation vector of the calibration data of the first lamp pole unit; Performing feature mapping on the basic implicit representation vector of the second lamp pole unit and the second lamp pole unit representation vector based on the second network component to obtain an implicit representation vector of the calibration data of the second lamp pole unit; Based on the reasoning component, the adjustment data implicit representation vector of the first lamp pole unit and the adjustment data implicit representation vector of the second lamp pole unit are inferred to determine the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit.

6. The method according to claim 1, characterized in that: The distributed charging network graph further includes a third lamp pole unit set, the third lamp pole unit set includes a third lamp pole unit, the machine learning network includes a first network component, a second network component, a third network component and an inference component; wherein the method further includes: Obtaining a basic implicit representation vector of the third lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the third lamp pole unit, wherein the basic implicit representation vector of the third lamp pole unit is obtained according to the involvement relationship of the lamp pole units in the third lamp pole unit set; Among them, feature mapping is performed on the basic implicit representation vector of the first lamp pole unit and the basic implicit representation vector of the second lamp pole unit based on the machine learning network to obtain the reasoning involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network graph, including: Performing feature mapping on a basic implicit representation vector of the first lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the first lamp pole unit based on the first network component to obtain a calibrated implicit representation vector of the first lamp pole unit; Performing feature mapping on a basic implicit representation vector of the second lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the second lamp pole unit based on the second network component to obtain a calibrated implicit representation vector of the second lamp pole unit; Performing feature mapping on the basic implicit representation vector of the third lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the third lamp pole unit based on the third network component to obtain a calibrated implicit representation vector of the third lamp pole unit; The reasoning component performs reasoning based on the adjusted implicit representation vector of the first lamp pole unit, the adjusted implicit representation vector of the second lamp pole unit, and the adjusted implicit representation vector of the third lamp pole unit to obtain the reasoned involvement relationship of the first lamp pole unit, the second lamp pole unit, and the third lamp pole unit; the real involvement relationship also includes the real involvement relationship of the first lamp pole unit, the second lamp pole unit, and the third lamp pole unit in the distributed charging network graph; The method of calibrating the machine learning network based on the real involvement relationship and the inferred involvement relationship between the first lamp pole unit and the second lamp pole unit in the distributed charging network map to obtain a calibrated machine learning network; and determining the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network includes: Based on the real involvement relationship and inferred involvement relationship of the first lamp pole unit, the second lamp pole unit and the third lamp pole unit in the distributed charging network map, the machine learning network is calibrated to determine the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network.

7. The method according to claim 1, characterized in that: The distributed charging network graph further includes a fourth lamp pole unit set, the fourth lamp pole unit set includes a fourth lamp pole unit, the machine learning network includes a first network component, a second network component, a fourth network component and an inference component; wherein the method further includes: Acquire a basic implicit representation vector of the fourth lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the fourth lamp pole unit, wherein the basic implicit representation vector of the fourth lamp pole unit is obtained according to the involvement relationship of the lamp pole units in the fourth lamp pole unit set; Performing feature mapping on a basic implicit representation vector of the second lamp pole unit and basic implicit representation vectors of adjacent lamp pole units of the second lamp pole unit based on the second network component to obtain a calibrated implicit representation vector of the second lamp pole unit; Performing feature mapping on the basic implicit representation vector of the fourth lamp pole unit and the basic implicit representation vectors of the adjacent lamp pole units of the fourth lamp pole unit based on the fourth network component to obtain a calibrated implicit representation vector of the fourth lamp pole unit; Reasoning the adjusted implicit representation vector of the second lamp pole unit and the adjusted implicit representation vector of the fourth lamp pole unit based on the reasoning component to obtain a reasoning involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network graph; Determine the real involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network map based on the distributed charging network map; Based on the real involvement relationship and the inferred involvement relationship between the second lamp pole unit and the fourth lamp pole unit in the distributed charging network map, the machine learning network is calibrated to determine the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network.

8. The method according to claim 1, characterized in that: The machine learning network includes a first network component and a second network component; wherein, determining the implicit representation vector of each lamp pole unit in the distributed charging network map based on the calibrated machine learning network includes: Performing feature mapping on the basic implicit representation vector of the first lamp pole unit based on the first network component of the calibrated machine learning network to obtain an implicit representation vector of the first lamp pole unit; The second network component of the calibrated machine learning network performs feature mapping on the basic implicit representation vector of the second lamp pole unit to obtain an implicit representation vector of the second lamp pole unit.

9. A power dispatching device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps in the method according to any one of claims 1 to 8 are implemented.

Images