Virtual power grid construction system and method based on source-grid-load-storage integration

Abstract

The invention discloses a virtual power grid construction method based on source-grid-load-storage integration, and belongs to the technical field of intelligent power grid optimization scheduling. The method comprises the following steps: constructing a cloud data center, converging historical data and historical scheduling instruction sets of a power generation device, an energy storage device and load equipment, and constructing a directed weighted graph of a virtual power grid based on the data, node weights and connection edge weights reflecting equipment loss characteristics and line transmission loss characteristics are obtained through learning of a graph neural network model, the weights serve as optimization coefficients, an optimal scheduling instruction is generated through calculation and issued to all equipment to be executed, meanwhile, new data are obtained through the optimal scheduling instruction, and the new data are transmitted to the equipment. And dynamically updating the weight and the directed weighting graph to form adaptive closed-loop control. According to the method, the limitation that all links in traditional power grid dispatching are relatively independent is broken through, and the economical efficiency, reliability and self-adaptive capacity of power grid operation are remarkably improved through a data-driven intelligent method.

Description

Technical Field

[0001] This invention relates to the field of integrated power generation, grid, load and energy storage technology, specifically a virtual power grid construction system and method based on integrated power generation, grid, load and energy storage. Background Technology

[0002] With the transformation of the energy structure and the increasing complexity of the power system, traditional power grids face severe challenges in coping with the large-scale integration of distributed energy resources, load diversification, and improving operational efficiency. As key components of the power system, the independent operation and lack of independent power generation, grid, load, and storage systems can easily lead to problems such as low energy utilization, slow regulatory response, and insufficient system stability. Especially against the backdrop of the continuously increasing proportion of renewable energy, the intermittency and volatility of power generation further exacerbate the difficulty of grid balancing and limit the improvement of overall operating efficiency.

[0003] Existing technologies have attempted to improve grid operation through local monitoring or independent optimization of generation, energy storage, or load sides. However, these methods are often limited to a single link or local area, lacking unified collection, analysis, and coordinated scheduling of data across the entire process of generation, grid, load, and storage, making it difficult to achieve system-level energy efficiency optimization. For example, the lack of a dynamic decision-making mechanism based on real-time efficiency in the coordination between generation and energy storage, and between energy storage and load response, leads to significant losses during energy conversion and transmission, and the overall system efficiency still needs improvement.

[0004] Therefore, there is an urgent need for a virtual power grid construction method that can deeply integrate the power source, grid, load, and storage links to achieve intelligent collaboration and dynamic optimization, so as to comprehensively improve the operating efficiency, flexibility and reliability of the power grid. Summary of the Invention

[0005] The purpose of this invention is to provide a technical solution for constructing a virtual power grid based on the integration of source, grid, load and storage, so as to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a virtual power grid based on the integration of source, grid, load, and storage, comprising the following steps:

[0007] Step S1: Construct a cloud data center for the virtual power grid. Obtain historical data and historical scheduling instruction sets generated by power generation devices, energy storage devices, and load devices through the cloud data center. The historical data includes the power generation change value of the power generation device, the energy storage change value of the energy storage device, and the power consumption change value of the load device, providing a rich information foundation for subsequent analysis.

[0008] Step S2: Based on historical data and historical scheduling instruction sets obtained from the cloud data center, construct a directed weighted graph of the virtual power grid, determine the node weights and connection edge weights between nodes in the directed weighted graph. The node weights reflect the operating efficiency and energy loss of the equipment itself, while the connection edge weights reflect the energy transmission efficiency and loss of the transmission path.

[0009] Step S3: Calculate the weights of power generation devices, energy storage devices, and load devices through the cloud data center, generate the optimal scheduling command, and send the optimal scheduling command to the corresponding power generation devices, energy storage devices, and load devices for execution. This achieves optimized control of the virtual power grid. This real-time control capability can respond promptly to changes in the power grid, such as fluctuations in power generation and changes in load, ensuring that the virtual power grid is always in the optimal operating state.

[0010] Step S4: Obtain new data through optimal scheduling instructions, upload the new data to the cloud data center for storage, dynamically update the directed weighted graph, continue to execute step S3, form closed-loop control, and through continuous cyclic optimization, the virtual power grid can achieve long-term stable and efficient operation.

[0011] Furthermore, the virtual power grid is associated with several power generation devices, energy storage devices, and load devices. A directed weighted graph of the virtual power grid is established, with the location information of each power generation device, energy storage device, and load device as nodes of the directed weighted graph, and the line connection paths between each power generation device, energy storage device, and load device as connection edges.

[0012] Furthermore, the method for determining the node weights and the weights of the connecting edges between nodes in a directed weighted graph is as follows:

[0013] Step S31: The historical scheduling instruction set includes scheduling instructions between various power generation devices, energy storage devices and load devices. Based on the scheduling instructions and historical data, the power loss of each power generation device, energy storage device and load device under different scheduling instructions and the power loss of power transmission between each power generation device, energy storage device and load device are obtained.

[0014] Step S32: Substitute the power loss of each power generation device, energy storage device, and load device under different scheduling instructions into the graph neural network model for training, and obtain the weights of each power generation device node, energy storage device node, and load device node. The calculation formula for the node is as follows:

[0015] ;

[0016] in, This represents the weight of the i-th node. This represents the sigmoid activation function. Represents the weight matrix. Let represent the feature vector of the i-th node, where the feature vector represents the amount of power loss. Let represent the type vector of the i-th node, where the type vectors are respectively the power generation type corresponding to the power generation device, the energy storage type corresponding to the energy storage device, and the power consumption type corresponding to the load device. The bias term, which comprehensively considers the feature vector and type vector of a node, can more accurately evaluate the importance and performance of a node, making the node weights more meaningful. The sigmoid activation function is used. Applying a nonlinear transformation to the calculation results can limit the weights of nodes to a reasonable range, while enhancing the nonlinear expressive power of the model and improving the accuracy of weight calculation.

[0017] Step S33: Substitute the power loss of each power generation device, energy storage device, and load device under different scheduling commands, and the power transmission loss between each power generation device, energy storage device, and load device under different scheduling commands, into the graph neural network model for training to obtain the weights of the connection edges between the nodes of the power generation device, energy storage device, and load device in each transmission path. The calculation formula for the connection edge is as follows:

[0018] ;

[0019] in, Indicates the first The node to the first The weight of the connecting edges of each node Let K represent the softmax normalization formula, where K represents any neighboring node of node i. The LeakyReLU activation function is used. This represents the slope coefficient of the negative input region. This represents the input value of the activation function. This indicates a linear transformation applied to the spliced ​​edge features. The weight matrix representing the node features. This represents the feature vector of the i-th node. This represents the feature vector of the j-th node. The feature vector representing the connection edge between node i and node j is normalized using the softmax formula. Normalizing the weights of the connecting edges ensures that the sum of the weights of all connecting edges is 1, making the weights comparable and facilitating subsequent analysis and decision-making. The LeakyReLU activation function can solve the gradient vanishing problem of the ReLU activation function in the negative input region. At the same time, the linear transformation can further extract the information of the connecting edge features and improve the accuracy of the connecting edge weight calculation.

[0020] Furthermore, the optimal scheduling instruction calculates the minimum comprehensive weight of each transmission path and prioritizes powering transmission paths with lower comprehensive weights. The formula is:

[0021] ;

[0022] in, This represents the set of nodes for all power generation devices. Represents the set of nodes for various energy storage devices. This represents the set of all load device nodes. Represents a node The training weights of the graph neural network model, Indicates a node under instruction set I The amount of electricity lost, where s represents the set of edges in the directed weighted graph. Represents a node The training weights of the graph neural network model, Indicates a node under instruction set I To the node The amount of power loss during transmission. This represents the multi-objective weighting coefficients, which are used to represent the multi-objective weighting coefficients. and The relative importance of the two objectives can be adjusted to achieve multi-objective optimization, so that the operation of the virtual power grid takes into account both the energy loss of the equipment itself and the energy loss during transmission. By calculating the minimum value of the comprehensive weight of each transmission path, power is given priority to the transmission path with the lower comprehensive weight. This strategy can effectively reduce the overall energy loss of the virtual power grid and improve energy utilization efficiency.

[0023] Furthermore, the method for dynamically updating the directed weighted graph comprises the following steps:

[0024] Step S61: Obtain new data through the optimal scheduling instruction. The new data includes the power generation change value of each power generation device, the energy storage change value of the energy storage device, and the power consumption change value of the load device under the optimal scheduling instruction, providing the latest information for the dynamic update of the directed weighted graph.

[0025] Step S62: The newly added data is incorporated into the historical data to generate a new historical dataset. This method of data accumulation and expansion can continuously enrich the scale and diversity of historical data, providing more sufficient data support for the training of graph neural network models and improving the generalization ability and accuracy of the models.

[0026] Step S63: Substitute the new historical dataset into the graph neural network model for calculation to generate new weight sets for power generation devices, energy storage devices, and load devices in each transmission path. Dynamically update the directed weighted graph based on the new weight sets, and generate new optimal scheduling instructions based on the updated directed weighted graph. This dynamic update mechanism enables the directed weighted graph to always remain consistent with the actual operating state of the virtual power grid, thereby generating new optimal scheduling instructions based on the updated directed weighted graph and realizing continuous optimization of the virtual power grid.

[0027] Furthermore, an interactive display platform is provided to digitally display the historical data and historical dispatch command sets generated by various power generation devices, energy storage devices, and load devices in the virtual power grid. Users can view these through the interactive display platform. This visualization method allows users to more intuitively understand the operation status and historical data of the virtual power grid, facilitating analysis and decision-making. It also provides users with a more convenient way to access and analyze data.

[0028] Compared with the prior art, the beneficial effects of the present invention are:

[0029] 1. This invention obtains historical data and scheduling instructions through a cloud data center, constructs a directed weighted graph based on a graph neural network model, accurately calculates the weights of each power generation, energy storage, load node and transmission path, generates optimal scheduling instructions, prioritizes low-loss paths for power transmission, reduces losses during energy conversion and transmission, and improves energy utilization efficiency.

[0030] 2. This invention monitors the power grid operation status in real time through a dynamic update module, integrates the new data generated by the real-time optimal scheduling command with historical data, and uses a graph neural network to dynamically update the weights of nodes and edges, thereby realizing adaptive optimization of the virtual power grid topology. This improves the dynamic response efficiency of the coordinated scheduling capability of source, grid, load and storage, and enhances the overall optimization and robustness of the system under changing operating conditions. Attached Figure Description

[0031] Figure 1 This invention provides a method for constructing a virtual power grid based on the integration of power generation, grid, load, and storage.

[0032] Figure 2 This invention provides a directed weighted graph based on the integration of source, grid, load, and storage.

[0033] Figure 3 This invention relates to a virtual power grid construction system based on the integration of power generation, grid, load, and storage. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Example 1: As Figure 1 As shown, a method for constructing a virtual power grid based on the integration of power generation, grid, load, and storage includes the following steps:

[0036] Step S1: Construct a cloud data center for the virtual power grid. Obtain historical data and historical scheduling instruction sets generated by power generation devices, energy storage devices, and load devices through the cloud data center. The historical data includes the power generation change value of the power generation device, the energy storage change value of the energy storage device, and the power consumption change value of the load device.

[0037] Step S2: Based on historical data and historical scheduling instruction sets obtained from the cloud data center, construct a directed weighted graph of the virtual power grid, and determine the node weights and connection edge weights between nodes in the directed weighted graph.

[0038] Step S3: Calculate the operating status of power generation devices, energy storage devices and load devices through the cloud data center, generate the optimal scheduling command, and send the optimal scheduling command to the corresponding power generation devices, energy storage devices and load devices for execution, so as to realize the optimized control of the virtual power grid.

[0039] Step S4: Obtain new data through optimal scheduling instructions, upload the new data to the cloud data center for storage, dynamically update the directed weighted graph, and continue to execute step S3 to form closed-loop control.

[0040] In this embodiment, there are two power generation devices, including power generation device A and power generation device B, two energy storage devices, including energy storage device C and energy storage device D, and two load devices, including load device E and load device F.

[0041] The cloud data center retrieves historical data generated by these devices over the past month:

[0042] The power generation of power generation device A changes as follows: 100 kWh is generated on day 1, 120 kWh is generated on day 2, and so on.

[0043] Energy storage device C's energy storage changes: The initial energy storage on day 1 is 50kWh, and after one day of use and charging, the initial energy storage on day 2 is 60kWh, etc.

[0044] The power consumption of load device E changes as follows: the power consumption on day 1 is 80 kWh, the power consumption on day 2 is 90 kWh, and so on.

[0045] At the same time, it acquires historical scheduling instruction sets, such as instructions on a certain day requiring power generation device A to charge energy storage device C by 30kWh and supply power to load device E by 50kWh, etc.

[0046] like Figure 2 As shown, the location information of power generation devices A and B, energy storage devices C and D, and load devices E and F are used as nodes in a directed weighted graph, and the line connections between them are used as edges. For example, there is an edge connecting power generation device A to energy storage device C, and an edge connecting energy storage device C to load device E, etc.

[0047] Specifically, based on historical dispatch command sets, the amount of electricity transmitted between various devices and the loss relationships during power transmission are obtained. For example, in the dispatching process over the past month, the average loss when power generation device A transmits electricity to energy storage device C is 5%. Assuming that under a certain dispatch command, power generation device A transmits 30 kWh of electricity to energy storage device C, the actual amount of electricity reaching energy storage device C is 30 × (1 - 5%) = 28.5 kWh, and the loss is 30 - 28.5 = 1.5 kWh.

[0048] Furthermore, the power loss of each power generation device, energy storage device, and load device under different scheduling commands is substituted into the graph neural network model for training, and the weights of each power generation device node, energy storage device node, and load device node are obtained respectively. The calculation formula for the node is as follows:

[0049] ;

[0050] in, Indicates the first The weight of each node, This represents the sigmoid activation function, which maps the result of a linear transformation to the interval (0,1). The weight matrix is ​​a learnable parameter used to perform a linear transformation on node features and types. Let represent the feature vector of the i-th node, where the feature vector represents the energy conversion efficiency. Let represent the type vector of the i-th node, where the type vectors are respectively the power generation type corresponding to the power generation device, the energy storage type corresponding to the energy storage device, and the power consumption type corresponding to the load device. This indicates the bias term.

[0051] Specifically, the power loss of each power generation device, energy storage device, and load device under different scheduling commands is substituted into the graph neural network model for training. Assuming that after training, for power generation device node A, its power loss... The value is 0.95 kWh, indicating that the loss of electricity flowing out of A is relatively small compared to the amount flowing in. (Type vector) The power generation type can be simply encoded as [1,0], and the weight matrix... The training matrix is ​​[0.2, 0.1, 0.3], and the resulting weight matrix is ​​[0.2, 0.1, 0.3]. =[0.1,0.2,0.3] is obtained through dynamic adjustment during the training process of the neural network.

[0052] The specific calculation method is as follows: the weight matrix is ​​randomly initialized to [x1, x2, x3]. The first feature value x1 corresponds to the power input value, the second feature value x2 corresponds to the power output value, and the third feature value x3 corresponds to the device load power. x1, x2, and x3 are normalized respectively. Then, in each iteration, the graph neural network model calculates the weighted sum of the input features and the current weights through forward propagation, and generates the predicted output through the activation function. The difference between the predicted value and the true label is quantified by the loss function. The gradient of the loss with respect to each weight is calculated to clarify the direction and magnitude of the weight adjustment. Finally, the weights are updated along the negative gradient direction through gradient descent to gradually reduce the loss.

[0053] After multiple iterations, the weights converge to the optimal value [0.1, 0.2, 0.3]. At this point, the graph neural network model has the smallest prediction error on the training data, under the normalization function: When x is large, the value of the normalization function is close to 1, indicating that the node may have high importance or strong power transmission capability in the virtual power grid.

[0054] When x is small, the value of the normalization function is close to 0, indicating that the node's importance or power transmission capability is relatively weak. Assume the bias term... If x is 0.1 and x is 0.5, then the weight of node A of the power generation device is... The calculation is as follows:

[0055] First calculate =[0.95,1,0];

[0056] =[0.2,0.1,0.3][0.95,1,0,0]=0.2×0.95+0.1×1+0.3×0=0.19+0.1+0=0.29;

[0057] =0.29 + 0.1 = 0.39;

[0058] ≈0.62;

[0059] =0.62×0.39≈0.24;

[0060] Similarly, the weights of other nodes can be calculated, such as the weight of node C in the energy storage device. ≈0.28, Weight of Load Device E Node ≈0.32.

[0061] Furthermore, the power loss of each power generation device, energy storage device, and load device under different scheduling commands, and the power transmission loss between each power generation device, energy storage device, and load device under different scheduling commands, are substituted into the graph neural network model for training to obtain the weights of the connection edges between the nodes of the power generation device, energy storage device, and load device in each transmission path. The calculation formula for the connection edge is as follows:

[0062] ;

[0063] in, Indicates the first The node to the first The weight of the connecting edges of each node Let K represent the softmax normalization formula, where K represents any neighboring node of node i. ) is the LeakyReLU activation function. The slope coefficient representing the negative input region is a small positive number, representing the weighted input received by the neuron. This represents the input value of the activation function. This indicates a linear transformation applied to the spliced ​​edge features. The weight matrix representing the node features. This represents the feature vector of the i-th node. This represents the feature vector of the j-th node. This represents the feature vector of the connection edge between node i and node j, specifically the transmission loss of the connection edge between node i and node j.

[0064] Specifically, the power losses of each power generation device, energy storage device, and load device under different scheduling commands, as well as the power transmission losses between these devices under different scheduling commands, are substituted into the graph neural network model for training. Assume that after training, the power loss at node A... =0.95kWh, the power loss at node C The power loss during power transmission between AC nodes is 0.9 kWh. =1kWh, the weight matrix W, after training with the graph neural network model, is [0.1, 0.2, 0.3], and the slope coefficient α of the negative input region is 0.01. According to the formula for calculating the edge weights:

[0065] ;

[0066] get =[0.1×0.95,0.2×0.95,0.3×0.95]=[0.095,0.19,0.285];

[0067] =[0.1×0.9,0.2×0.9,0.3×0.9]=[0.09,0.18,0.27];

[0068] =[0.095,0.19,0.285,0.09,0.18,0.27,];

[0069] Assuming the linear transformation part has been calculated =0.5;

[0070] LeakyReLU activation function max(0.01×0.5, =0.5;

[0071] After normalization, the result is 0.4;

[0072] The weight of the connection edge from power generation device A to energy storage device C =0.4 × 0.5 = 0.2;

[0073] Similarly, the weights of other connection edges can be calculated, such as the weight of the connection edge from energy storage device C to load device E. ≈0.25.

[0074] Furthermore, the optimal scheduling instruction prioritizes charging energy storage devices with lower weights or directly powering load devices by calculating the loss on each transmission path. When the load device consumes power, it controls the energy storage devices with lower weights to discharge first. The formula for the optimal scheduling instruction is:

[0075] ;

[0076] in, This represents the set of nodes for all power generation devices. Represents the set of nodes for various energy storage devices. This represents the set of all load device nodes. Represents a node The training weights of the graph neural network model, Indicates a node under instruction set I The state change quantity, s represents the set of connected edges in the directed weighted graph. Represents a node Training weights of a graph neural network model Indicates instruction set I node To the node Transmission loss, This represents the multi-objective weighting coefficient, which is adjusted in real time by the cloud data center based on the relative importance of connection points and connection edges.

[0077] Specifically, the weights of power generation devices, energy storage devices, and load devices are calculated through a cloud data center. Assuming that load device E currently requires 80 kWh of electricity, load device F requires 70 kWh of electricity, power generation device A can currently generate 100 kWh, power generation device B can currently generate 90 kWh, energy storage device C currently stores 60 kWh, and energy storage device D currently stores 50 kWh, according to the optimal scheduling instruction formula:

[0078] ;

[0079] Assuming multi-objective weighting coefficients a = 0.6 and β = 0.4, the optimal scheduling instruction set is generated through calculation. When calculating the optimal scheduling instruction set, the focus is first on meeting the load requirements of the equipment, because unmet load requirements will lead to significant state changes. Based on the first term in the formula, which emphasizes the state changes of the node itself and has a weight of a = 0.6 The power supply is prioritized from the generating units. Generator A has a generating capacity of 100 kWh, and generator B has a generating capacity of 90 kWh. Load devices E require 80 kWh, and F requires 70 kWh. First, generator A supplies 60 kWh to load device E, and generator B supplies 70 kWh to load device F. At this point, load device E still has a remaining demand of 20 kWh, and generator A has a remaining generating capacity of 40 kWh, while generator B has a remaining generating capacity of 20 kWh. Next, energy storage devices are considered. Since energy storage device C has a lower weight, to minimize the overall objective function value, 20 kWh of the remaining power from generator A is allocated to charge energy storage device C, and the remaining 20 kWh from generator B is allocated to charge energy storage device D. This is based on the second term in the formula, which emphasizes transmission loss and has a weight β = 0.4. Calculate the transmission losses for each transmission path: from generator A to load device E, from generator B to load device F, from generator A to energy storage device C, and from generator B to energy storage device D. (I), and combined with each connecting edge Calculate the value of the second term. Finally, add the values ​​of the first and second terms to obtain the objective function value for this instruction set. By performing similar calculations and comparisons with other possible scheduling instruction sets, it is found that this instruction set minimizes the objective function value, thus determining it as the optimal scheduling instruction set.

[0080] Specifically, the system monitors changes in new data of the virtual power grid in real time through sensors and other devices. For example, it can obtain data in real time that the current power generation of power generation device A is 110kWh, which is different from the previous prediction or historical data. This real-time data, along with data such as the actual power consumption of load device E (85kWh), is uploaded to the cloud data center for storage.

[0081] After executing the latest set of scheduling instructions, new data for the current time period is obtained. For example, if the actual charging of energy storage device C by power generation device A is 22kWh, which differs from the optimal scheduling instruction of 20kWh, the new data is merged into the historical data to generate a new historical dataset.

[0082] The new historical dataset is fed into the graph neural network model for calculation, regenerating new weight sets for power generation devices, energy storage devices, and load devices in each transmission path. For example, after recalculation, the weights of power generation device A node are updated to... '≈0.22, connecting edge '≈0.22 etc., dynamically update the directed weighted graph according to the new weight set, generate a new optimal scheduling instruction based on the updated directed weighted graph, and issue the optimal scheduling instruction to the corresponding power generation device, energy storage device and load device for execution.

[0083] Specifically, an interactive display platform is provided to digitally display historical data generated by various power generation devices, energy storage devices, and load devices in the virtual power grid. For example, a bar chart of the daily power generation of power generation device A over the past week and a table of dispatch instructions for a certain day can be displayed. Users can view these through the interactive display platform and click on the icon of power generation device A to view its detailed power generation change curve and related dispatch instruction information.

[0084] Example 2: Figure 3 As shown, a virtual power grid construction system based on the integration of power generation, grid, load, and storage includes:

[0085] The cloud data center module is used to acquire historical data and historical scheduling instruction sets generated by power generation devices, energy storage devices, and load devices. The historical data includes the power generation change value of the power generation device, the energy storage change value of the energy storage device, and the power consumption change value of the load device.

[0086] A directed weighted graph construction module, connected to the cloud data center module, is used to construct a directed weighted graph of the virtual power grid based on historical data and historical scheduling instruction sets, and to determine the node weights and the set of connection edge weights between nodes in the directed weighted graph.

[0087] The calculation module, connected to the directed weighted graph construction module, is used to calculate the weights of the power generation device, energy storage device and load device, generate the optimal scheduling instruction, and send the optimal scheduling instruction to the corresponding power generation device, energy storage device and load device for execution.

[0088] The dynamic update module is connected to the cloud data center module and the directed weighted graph construction module. It obtains new data through optimal scheduling instructions and uploads the new data to the cloud data center module for storage. At the same time, it dynamically updates the directed weighted graph and inputs the updated directed weighted graph into the calculation module.

[0089] Furthermore, the directed weighted graph construction module includes a node weight determination unit and a connection edge weight determination unit. The node weight determination unit is used to calculate the weight set of each power generation device node, energy storage device node, and load device node. The connection edge weight determination unit is used to calculate the weight set of the connection edges between different nodes.

[0090] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for constructing a virtual power grid based on the integration of source, grid, load, and storage, characterized in that: Includes the following steps: Step S1: Construct a cloud data center for the virtual power grid. Obtain historical data and historical scheduling instruction sets generated by power generation devices, energy storage devices, and load devices through the cloud data center. The historical data includes the power generation change value of the power generation device, the energy storage change value of the energy storage device, and the power consumption change value of the load device. Step S2: Based on historical data and historical scheduling instruction sets obtained from the cloud data center, construct a directed weighted graph of the virtual power grid, and determine the node weights and connection edge weights between nodes in the directed weighted graph. Step S3: Calculate the weights of power generation devices, energy storage devices, and load devices through the cloud data center, generate the optimal scheduling command, and send the optimal scheduling command to the corresponding power generation devices, energy storage devices, and load devices for execution, thereby achieving optimized control of the virtual power grid. Step S4: Obtain new data through optimal scheduling instructions, upload the new data to the cloud data center for storage, dynamically update the directed weighted graph, and continue to execute step S3 to form closed-loop control.

2. The method for constructing a virtual power grid based on the integration of source, grid, load, and storage as described in claim 1, characterized in that: The virtual power grid is associated with several power generation devices, energy storage devices, and load devices. A directed weighted graph of the virtual power grid is established, with the location information of each power generation device, energy storage device, and load device as nodes of the directed weighted graph, and the line connection paths between each power generation device, energy storage device, and load device as connection edges.

3. The method for constructing a virtual power grid based on source-grid-load integration according to claim 2, characterized in that: The steps to determine the node weights and edge weights between nodes in a directed weighted graph are as follows: Step S31: The historical scheduling instruction set includes scheduling instructions between various power generation devices, energy storage devices and load devices. Based on the scheduling instructions and historical data, the power loss of each power generation device, energy storage device and load device under different scheduling instructions and the power loss of power transmission between each power generation device, energy storage device and load device are obtained. Step S32: Substitute the power loss of each power generation device, energy storage device, and load device under different scheduling instructions into the graph neural network model for training, and obtain the weights of each power generation device node, energy storage device node, and load device node. The calculation formula for the node is as follows: ； in, This represents the weight of the i-th node. This represents the sigmoid activation function. Represents the weight matrix. Let represent the feature vector of the i-th node, where the feature vector represents the amount of power loss. Let represent the type vector of the i-th node, where the type vectors are respectively the power generation type corresponding to the power generation device, the energy storage type corresponding to the energy storage device, and the power consumption type corresponding to the load device. Indicates the bias term; Step S33: Substitute the power loss of each power generation device, energy storage device, and load device under different scheduling commands, and the power transmission loss between each power generation device, energy storage device, and load device under different scheduling commands, into the graph neural network model for training to obtain the weights of the connection edges between the nodes of the power generation device, energy storage device, and load device in each transmission path. The calculation formula for the connection edge is as follows: ； in, Indicates the first The node to the first The weight of the connecting edges of each node Let K represent the softmax normalization formula, where K represents any neighboring node of node i. The LeakyReLU activation function is used. This represents the slope coefficient of the negative input region. This represents the input value of the activation function. This indicates a linear transformation applied to the spliced ​​edge features. The weight matrix representing the node features. This represents the feature vector of the i-th node. This represents the feature vector of the j-th node. This represents the feature vector of the connecting edge between node i and node j.

4. The method for constructing a virtual power grid based on the integration of source, grid, load, and storage as described in claim 1, characterized in that: The optimal scheduling instruction calculates the minimum comprehensive weight of each transmission path and prioritizes powering transmission paths with lower comprehensive weights. The formula is: ； in, This represents the set of nodes for all power generation devices. Represents the set of nodes for various energy storage devices. This represents the set of all load device nodes. Represents a node The training weights of the graph neural network model, Indicates a node under instruction set I The amount of electricity lost, where s represents the set of edges in the directed weighted graph. Represents a node The training weights of the graph neural network model, Indicates a node under instruction set I To the node The amount of power loss during transmission. This represents the multi-objective weighting coefficient.

5. The method for constructing a virtual power grid based on the integration of source, grid, load, and storage as described in claim 4, characterized in that: The steps of the method for dynamically updating the directed weighted graph are as follows: Step S61: Obtain new data through the optimal scheduling instruction. The new data includes the power generation change value of each power generation device, the energy storage change value of the energy storage device, and the power consumption change value of the load device under the optimal scheduling instruction. Step S62: Incorporate the new data into the historical data to generate a new historical dataset; Step S63: Substitute the new historical dataset into the graph neural network model for calculation to generate new weight sets for power generation devices, energy storage devices and load devices in each transmission path. Dynamically update the directed weighted graph according to the new weight sets, and generate new optimal scheduling instructions again according to the updated directed weighted graph.

6. The method for constructing a virtual power grid based on the integration of source, grid, load, and storage as described in claim 4, characterized in that: An interactive display platform is provided to digitally display the historical data and historical dispatch command sets generated by various power generation devices, energy storage devices and load devices in the virtual power grid, which users can view through the interactive display platform.

7. A virtual power grid construction system based on the integration of power generation, grid, load, and storage, characterized in that: include: The cloud data center module is used to acquire historical data and historical scheduling instruction sets generated by power generation devices, energy storage devices, and load devices. The historical data includes the power generation change value of the power generation device, the energy storage change value of the energy storage device, and the power consumption change value of the load device. A directed weighted graph construction module, connected to the cloud data center module, is used to construct a directed weighted graph of the virtual power grid based on historical data and historical scheduling instruction sets, and to determine the node weights and the set of connection edge weights between nodes in the directed weighted graph. The calculation module, connected to the directed weighted graph construction module, is used to calculate the weights of the power generation device, energy storage device and load device, generate the optimal scheduling instruction, and send the optimal scheduling instruction to the corresponding power generation device, energy storage device and load device for execution. The dynamic update module is connected to the cloud data center module and the directed weighted graph construction module. It obtains new data through optimal scheduling instructions and uploads the new data to the cloud data center module for storage. At the same time, it dynamically updates the directed weighted graph and inputs the updated directed weighted graph into the calculation module.

8. A virtual power grid construction system based on source-grid-load-storage integration according to claim 6, characterized in that: The directed weighted graph construction module includes a node weight determination unit and a connection edge weight determination unit. The node weight determination unit is used to calculate the weight set of each power generation device node, energy storage device node, and load device node. The connection edge weight determination unit is used to calculate the weight set of the connection edges between different nodes.

Images