A new energy station group energy optimization control and model training method and device

Abstract

The application relates to the technical field of power control, and discloses a new energy station group energy optimization control and an optimization control model training method, which comprises the following steps: a directed weighted graph of a new energy station group is constructed; according to the time scale of power generation of the new energy group, the time scale classification of each transmission path is determined; according to the required time scale type at present, the edges that need to be reserved are determined to form a dynamic adjacency matrix; node features, edge features and the dynamic adjacency matrix are input into an optimization control model to obtain a scheduling instruction; and the scheduling instruction is used for control. The application establishes the transmission path of the new energy station group by constructing a directed weighted graph, divides the long-term and short-term time scales of the path on the basis, judges the long-term and short-term paths required by the current running state, considers the different influences of the energy storage forms of different energy storage stations, finally inputs the generated dynamic adjacency matrix into the optimization control model to obtain the scheduling instruction, and the accuracy of the scheduling control is improved.

Description

Technical Field

[0001] This invention relates to the field of power control technology, specifically to an energy optimization control method for new energy power plant clusters and a training method for the optimization control model. Background Technology

[0002] With the increase in wind and solar power installed capacity, and considering the strong uncertainty in the output of new energy power generation, it is necessary to configure large-scale adjustable energy storage power stations to mitigate the volatility of new energy power generation. Currently, new energy power plants mainly use electrochemical energy storage, with a few using adjustable energy storage power stations such as thermal energy storage and hydrogen energy storage. Existing technologies treat adjustable energy storage power stations as independent units. For new energy power plant clusters with multiple adjustable energy storage power stations, the topological dependence of energy conversion between different energy storage forms is not considered, nor is the impact of new energy uncertainty on the energy storage forms of different adjustable energy storage power stations distinguished. In actual control processes, the schemes of allocating according to capacity ratio or average allocation usually ignore the efficiency issues of energy conversion between different energy forms. Summary of the Invention

[0003] This invention provides a method, device, and product for energy optimization control of new energy power plant clusters, which solves the problems of lack of consideration for the topological dependence of multi-energy form conversion and neglect of the differences in conversion efficiency of heterogeneous energy in traditional new energy power plant cluster scheduling strategies, thereby making scheduling control more accurate.

[0004] In a first aspect, the present invention provides an energy optimization control method for a cluster of new energy power plants, comprising: A directed weighted graph of the new energy power plant cluster is constructed, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. Based on the time scale of the generation uncertainty of each power group in the directed weight graph, the time scale classification of each transmission path is determined. The required time scale type is determined based on the current operating status of the new energy power station cluster; Based on the required time scale type and the time scale classification of each transmission path, determine the edges that need to be retained in the directed weight graph, and form a dynamic adjacency matrix of the directed weight graph. The node features, edge features, and dynamic adjacency matrix of the directed weighted graph are input into a pre-established optimization control model for calculation to obtain the scheduling instructions for the new energy power station group. The energy transmission of the new energy power plant cluster is controlled by scheduling commands.

[0005] Based on the real-time operating status of the new energy power plant cluster, the pre-determined path type is selected, the matching transmission path is activated, and a dynamic adjacency matrix is ​​obtained. Then, the training optimization control model is used to calculate the scheduling instructions of the new energy power plant cluster and output the optimal power scheduling instructions. This method combines the topological dependence of energy form conversion and the time scale difference, so that the scheduling instructions are more in line with the actual physical characteristics and regulation capabilities of each power source.

[0006] In one alternative implementation, the method further includes: The update interval is determined based on the link delay of the control system and the instruction response time of the scheduling command. The time scale type required for updating the new energy power station cluster is determined according to the update interval.

[0007] The update interval is determined by considering link delay and command response time. By taking link delay into account, it is avoided that the command will be overwritten before it takes effect due to the update interval being too short. By taking command response time into account, it is avoided that the control will lag behind the system state change due to the interval being too long. The update interval determined in this way is used to update the time scale type required by the new energy power station group, ensuring that the speed of updating the dynamic adjacency matrix matches the adjustment capability of the equipment.

[0008] In one alternative implementation, the edge features include edge weight data, which is determined based on one or more of the following: the regulating power supply response time, energy conversion efficiency, and real-time grid electricity price of the transmission path corresponding to the edge.

[0009] By comprehensively considering one or more of the factors such as the power supply response time, energy conversion efficiency, and real-time grid electricity price, edge weight data is generated, enabling the optimization control model to comprehensively consider factors such as speed, efficiency, and economy when selecting paths.

[0010] In one optional implementation, the current operating status includes renewable energy output fluctuation data, grid connection point frequency change rate data, load peak-valley difference data, and renewable energy predicted wind and solar curtailment data. The required time scale type is determined based on the current operating status of the renewable energy power station cluster, including: If the power output fluctuation data of new energy sources is higher than the power output fluctuation threshold, or the frequency change rate data of the grid connection point is higher than the frequency change threshold, the current time scale required by the new energy power station group is determined to be short-time path. If the load peak-valley difference data is higher than the peak-valley difference threshold, the current required time scale type of the new energy power station group is determined to be a long-short time mixed path; If the predicted curtailment data for wind and solar power exceeds the curtailment threshold, the current required time scale for the new energy power plant cluster is determined to be a long-term path.

[0011] By comprehensively considering one or more of the factors such as the power supply response time, energy conversion efficiency, and real-time grid electricity price, edge weight data is generated, enabling the optimization control model to comprehensively consider speed, efficiency, and economy when selecting paths.

[0012] This invention provides a method for optimizing control model training, comprising: A directed weighted graph of the new energy power plant cluster is constructed, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. Based on the time scale of the generation uncertainty of each power group in the directed weight graph, the time scale classification of each transmission path is determined. Multiple training samples are constructed based on the historical operation data of the new energy power station cluster at different times. The training samples include node features, edge features, dynamic adjacency matrix, and actual energy transmission values ​​of each energy transmission path determined based on the historical operation data. The node features, edge features, and dynamic adjacency matrix of each sample are input into the model to be trained to obtain the scheduling instructions corresponding to each sample. The scheduling instructions contain the energy instruction transmission values ​​of each energy transmission path. The loss value is calculated by classifying and categorizing the actual energy transmission value, energy command transmission value, and time scale of each energy transmission path for each energy transmission path. If the loss value does not meet the preset conditions, the model to be trained is updated according to the loss value. The process returns to the step of inputting the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained, until the loss value does not meet the preset conditions, and the optimized control model is obtained.

[0013] Training samples were constructed using historical operational data, and time-scale classification was introduced as input to train the model. The path was divided into paths of different time scales, enabling the model to identify the response characteristics of different regulatory resources. This ensures that fast-response energy storage devices can smooth out short-term fluctuations and frequency regulation, while slow-response energy storage devices can handle the absorption of abandoned power and energy time shift, thereby achieving accurate regulation of electricity and avoiding resource misallocation.

[0014] In one optional implementation, the loss value is calculated using the actual energy transmission value, the energy command transmission value, and the time scale classification of each energy transmission path, including: The energy conversion efficiency loss value is calculated based on the real-time conversion efficiency data and transmitted energy data of each edge in each energy transmission path; The response sensitivity loss value is calculated based on the command value and the actual value of each energy transmission path, combined with the physical constraints of the system. Calculate the path switching smoothness loss value based on the current selection state vector and the selection state vector after a preset time step for each energy transmission path. The loss value is obtained by fusing the energy conversion efficiency loss value, response sensitivity loss value, and path switching smoothness loss value with preset hyperparameters.

[0015] By considering the loss values ​​of energy conversion efficiency, response sensitivity, and path switching smoothness, the model can simultaneously achieve efficiency, response speed, and smoothness during training, thus realizing multi-objective collaborative optimization.

[0016] In one optional implementation, the energy conversion efficiency loss value is calculated based on the real-time conversion efficiency data and transmitted energy data of each edge in each energy transmission path, including: Based on the sum of the real-time conversion efficiency data of each edge in each energy transmission path, the energy loss data of each energy transmission path is determined. The energy conversion efficiency loss of each energy transmission path is determined by multiplying the energy loss data with the transmitted energy data. The energy conversion efficiency loss of each energy transmission path is summed to obtain the energy conversion efficiency loss value.

[0017] In one optional implementation, the command value for each energy transmission path includes a power command value and an energy command value, the actual value for each energy transmission path includes an actual power value and an actual energy value, and the system physical constraints include ramp rate constraints and capacity constraints. Based on the command value and actual value of each energy transmission path, and in conjunction with the system physical constraints, the response sensitivity loss value is calculated, including: The ratio of the absolute difference between the power command value and the actual power value of the energy transmission path at each short time scale to the ramp rate constraint is summed and calculated. Combined with the short time scale weight, the short-term impact error is determined. The ratio of the absolute difference between the energy command value and the actual energy value of the energy transmission path at each long time scale to the capacity constraint is summed and calculated. Combined with the long time scale weights, the long-term impact error is determined. The short-term impact error and the long-term impact error are summed to obtain the response sensitivity loss value.

[0018] In one optional implementation, the path switching smoothness loss value is calculated based on the selection state vector of each energy transmission path at the current moment and the selection state vector after a preset time step, including: The timing smoothness penalty term is determined based on the difference between the current selected state vector of each energy transmission path and the selected state vector after the preset time step. The path consistency penalty term is determined by combining the path consistency penalty weight with the difference between the selection state vector of each energy transmission path and the selection state vector of all other paths except the current path. The timing smoothness penalty term and the path consistency penalty term are summed to obtain the path switching smoothness loss value.

[0019] Secondly, the present invention provides an energy optimization control device for a new energy power station cluster, comprising: The graph generation module is used to construct a directed weighted graph of the new energy power plant cluster, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. The path scale classification module is used to determine the time scale classification of each transmission path based on the time scale of the generation uncertainty of each power group in the directed weight graph. The scale confirmation module is used to determine the required time scale type based on the current operating status of the new energy power station group. The matrix generation module is used to determine the edges that need to be retained in the directed weight graph based on the current required time scale type and the time scale classification of each transmission path, and to form a dynamic adjacency matrix of the directed weight graph. The instruction generation module is used to input the node features, edge features, and dynamic adjacency matrix of the directed weighted graph into a pre-established optimization control model for calculation to obtain the scheduling instructions of the new energy power station group. The command control module is used to control the energy transmission of the new energy power plant cluster using scheduling commands.

[0020] This invention provides an optimized control model training device, comprising: The directed weighted graph generation module is used to construct a directed weighted graph of the new energy power plant cluster, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. The scale classification module is used to determine the time scale classification of each transmission path based on the time scale of the generation uncertainty of each power group in the directed weight graph. The sample construction module is used to construct multiple training samples based on the historical operation data of the new energy power station group at different times. The training samples include node features, edge features, dynamic adjacency matrix and actual energy transmission values ​​of each energy transmission path determined based on the historical operation data. The instruction generation module is used to input the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained, and obtain the scheduling instructions corresponding to each sample. The scheduling instructions contain the energy instruction transmission values ​​of each energy transmission path. The loss calculation module is used to calculate the loss value by classifying and classifying the actual energy transmission value, energy command transmission value and time scale of each energy transmission path for each energy transmission path. The training optimization module is used to update the model to be trained based on the loss value if the loss value does not meet the preset conditions. It returns the steps of inputting the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained until the loss value no longer meets the preset conditions, thus obtaining the optimized control model.

[0021] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform a new energy power station cluster energy optimization control method or an optimization control model training method according to the first aspect or any corresponding embodiment described above.

[0022] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions, which are used to cause a computer to execute a new energy power station cluster energy optimization control method or an optimization control model training method according to the first aspect or any corresponding embodiment described above.

[0023] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute a new energy power station cluster energy optimization control method or an optimization control model training method according to the first aspect or any corresponding embodiment described above. Attached Figure Description

[0024] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an energy optimization control method for a new energy power station cluster according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the optimized control model training method according to an embodiment of the present invention; Figure 4 This is a structural block diagram of an energy optimization control device for a new energy power station cluster according to an embodiment of the present invention; Figure 5This is a structural block diagram of an optimized control model training device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0027] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0028] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0029] As an optional application scenario of this invention, such as Figure 1 As shown, this energy optimization control system for a new energy power station cluster may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0030] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0031] This invention provides an energy optimization control method for a new energy power station cluster. By constructing a directed weighted graph, the transmission path of the new energy power station cluster is established. Based on this, the path is divided into long-term and short-term time scales. By determining the long-term and short-term paths required for the current operating state, the differences in the energy storage forms of different energy storage power stations are taken into account. Finally, the generated dynamic adjacency matrix is ​​input into the optimization control model to obtain scheduling instructions, thereby improving the accuracy of scheduling control.

[0032] According to an embodiment of the present invention, an embodiment of an energy optimization control method for a new energy power station group is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0033] This embodiment provides an energy optimization control method for a new energy power station cluster. Figure 2 This is a flowchart of an energy optimization control method for a new energy power station cluster according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Using each power generation group and grid connection node in the new energy power plant cluster as nodes, and the energy transmission path between each power generation group and grid connection node as edges, construct a directed weighted graph of the new energy power plant cluster.

[0034] Nodes represent entities within a renewable energy power plant cluster and grid-connected nodes, each containing real-time status information. Edges represent the connections between entities, characterizing the energy transmission paths between nodes. A directed weighted graph is a graph structure that uses various power sources within a renewable energy power plant cluster as nodes, represents energy transmission paths with directed edges, and assigns weights to each edge to quantify transmission costs or efficiency.

[0035] For example, nodes may include new energy fleets such as wind power or photovoltaic power, adjustable power sources with electrochemical energy storage, thermal energy storage or hydrogen energy storage, and grid-connected nodes. The feature vector of a node contains information such as power and operating status.

[0036] Step S202: Based on the time scale of the generation uncertainty of each power group in the directed weight graph, determine the time scale classification of each transmission path.

[0037] The time scale of power generation uncertainty for each power generation group refers to the fluctuation characteristics, prediction difficulty, and variation patterns exhibited by different power generation groups, such as wind power or photovoltaic power, within different time ranges due to the influence of natural conditions on their power output. The time scale classification of each transmission path refers to categorizing each energy transmission path into categories applicable to short or long time scales based on factors such as the type of power source involved, response speed, and regulation capability. This allows for the dynamic activation or disabling of corresponding paths under different operating conditions, achieving precise scheduling.

[0038] For example, the time scale classification of transmission paths can be divided into short time scales with a time scale of seconds-minutes-hours and long time scales with a time scale of days before and within days.

[0039] As an example, short-term transmission paths can include: surplus renewable energy - electrochemical energy storage - power grid; and off-peak power from the power grid - electrochemical energy storage - power grid. These are used for short-term wind and solar power fluctuation smoothing, frequency regulation, and participation in peak shaving and peak regulation.

[0040] Long-term transmission paths can include: solar thermal power - thermal energy storage - power grid; surplus renewable energy - thermal energy storage - power grid; surplus renewable energy - hydrogen energy storage - power grid. This is primarily used for long-term optimization of wind and solar power power fluctuations and for addressing day-ahead renewable energy forecasting biases.

[0041] Step S203: Determine the required time scale type based on the current operating status of the new energy power station group.

[0042] By monitoring real-time operational data such as power output fluctuations, peak-to-valley load differences, and curtailment rates of renewable energy power plant clusters, it can be determined whether the current system's adjustment needs are short-term, long-term, or a mixed time scale, thereby deciding which type of energy transmission path to activate.

[0043] Step S204: Determine the edges that need to be retained in the directed weight graph based on the current required time scale type and the time scale classification of each transmission path, and form a dynamic adjacency matrix of the directed weight graph.

[0044] Based on the short-term, long-term, or mixed time scales required by the system, and combined with the time scale category of each energy transmission path, edges that meet the conditions are selected from the complete directed weight graph. The adjacency matrix of the directed weight graph is then updated to generate a real-time updated adjacency matrix for subsequent graph neural network optimization calculations.

[0045] For example, the adjacency matrix of a directed weighted graph can be updated according to the following formula:

[0046] in, These are elements in the adjacency matrix. The corresponding element is set to 1 when a physical connection exists between two nodes and the path is active at the current time scale, indicating that the path is active; otherwise, it is set to 0, indicating that the path is closed.

[0047] Step S205: Input the node features, edge features, and dynamic adjacency matrix of the directed weighted graph into the pre-established optimization control model for calculation to obtain the scheduling instructions for the new energy power station group.

[0048] Node features include the operating state parameters of each power node, edge features include the weight of each valid path, and the dynamic adjacency matrix represents the paths currently active. The optimized control model refers to the core algorithm model, pre-trained, used to calculate the optimal power scheduling commands for each power source in the renewable energy power plant cluster. The scheduling commands are the specific power command values ​​and path commands output by the optimized control model to control the operation of each power source in the renewable energy power plant cluster.

[0049] For example, the power command value can be positive or negative. As an example, for an energy storage node, a positive power value indicates that the current energy storage node needs to discharge, while a negative power value indicates that the current energy storage node needs to charge.

[0050] For example, the optimization control model can be constructed using an improved Dynamic Graph Convolutional Network (DGCN). As an example, the architecture of the optimization control model can include a spatiotemporal encoder, a processor, and a decoder. In the spatiotemporal encoder, a graph convolutional network (GCN) is used to capture the spatial dependencies (i.e., energy transfer relationships) of nodes at the current time step. Simultaneously, the node features from historical windows are input into a Long Short-Term Memory (LSTM) network unit to extract the patterns of node state evolution over time. In the processor, the spatial features output by the GCN and the temporal features output by the LSTM are fused and input into a message-passing neural network. The message-passing mechanism simulates the "flow" and "transformation" of energy between nodes. Through multiple iterations, each node aggregates neighbor information to identify the globally optimal energy transfer path. In the decoder, the high-dimensional features output by the processor are mapped back to physical space, outputting the final decision instruction.

[0051] Step S206: Control the energy transmission of the new energy power station group using scheduling instructions.

[0052] The power dispatch commands, such as the power generation and charging / discharging power of each power source calculated by the optimized control model, are sent to the node equipment in the new energy power plant group for execution. By actually adjusting the operating status of each power source, the control of energy transmission path and power is realized.

[0053] This embodiment provides an energy optimization control method for a new energy power plant cluster. Based on the real-time operating status of the new energy power plant cluster, it selects a pre-determined path type, activates the matching transmission path, and obtains a dynamic adjacency matrix. Then, it uses a trained optimization control model to calculate the scheduling instructions for the new energy power plant cluster and outputs the optimal power scheduling instructions. This method combines the topological dependence of energy form conversion and the time scale difference, making the scheduling instructions more consistent with the actual physical characteristics and regulation capabilities of each power source.

[0054] In an optional embodiment, when performing step S203, which determines the required timescale type based on the current operating status of the new energy power station cluster, the method further includes: Step a1: Determine the update interval based on the link delay of the control system and the instruction response time of the scheduling command.

[0055] Link latency refers to the communication transmission time from when the control system issues a command to when the device receives the command. Command response time refers to the time required for the device to complete the actual execution of the command after receiving it. Update interval refers to the minimum time interval required to obtain the time scale type of the new energy power station cluster.

[0056] Step a2: Update the time scale type required for the new energy power station cluster according to the update interval.

[0057] The time scale type required to update the new energy power station cluster refers to the time scale required to reacquire the current operating status of the new energy power station cluster and re-determine the current status of the new energy power station cluster based on the current operating status.

[0058] This embodiment provides an energy optimization control method for a new energy power station cluster. The update interval is determined by link delay and command response time. By considering link delay, the method avoids the command being overwritten before it takes effect due to an excessively short update interval. By considering command response time, the method avoids the control being delayed after the system state changes due to an excessively long interval. The update interval determined in this way is used to update the time scale type required by the new energy power station cluster, ensuring that the speed of updating the dynamic adjacency matrix matches the adjustment capability of the equipment.

[0059] In an optional embodiment, when performing step S205, the node features, edge features, and dynamic adjacency matrix of the directed weighted graph are input into a pre-established optimization control model for calculation to obtain the scheduling instructions for the new energy power station group, wherein the edge features include edge weight data, and the weight data is determined based on one or more of the following: the adjustment power supply response time, energy conversion efficiency, and real-time grid electricity price of the transmission path corresponding to the edge.

[0060] Edge weights are used to measure and represent the cost, efficiency, or priority of the path at the current moment.

[0061] For example, the edge weights can be calculated using the following formula:

[0062] in, Let be the weight of edge ij. To adjust the power supply response time for the normalized path ij, The energy conversion efficiency of path ij after normalization. For real-time electricity prices on the power grid, , and This represents the dynamic weighting coefficient.

[0063] This embodiment provides an energy optimization control method for a new energy power plant cluster. By comprehensively considering one or more of the following factors—the response time of the regulating power source, energy conversion efficiency, and the real-time electricity price of the power grid—it generates edge weight data, enabling the optimization control model to comprehensively consider factors such as speed, efficiency, and economy when selecting paths.

[0064] In an optional embodiment, the current operating status includes new energy power output fluctuation data and grid connection point frequency change rate data. Step S203 determines the required time scale type based on the current operating status of the new energy power station group, specifically including: Step b1: If the power output fluctuation data of new energy sources is higher than the power output fluctuation threshold, or the frequency change rate data of the grid connection point is higher than the frequency change threshold, the current required time scale type of the new energy power station group is determined to be short-time path.

[0065] New energy power output fluctuation data refers to a quantitative indicator of the magnitude of change in wind or solar power generation over a short period of time, used to trigger short-time path activation judgments. The power output fluctuation threshold is a pre-set critical value for power change, used to determine whether power generation fluctuations require rapid adjustment. Grid connection point frequency change rate data refers to the rate of change of the grid frequency between the power generation group and the grid connection point over time, used to determine whether a rapid power imbalance has occurred in the grid. The frequency change threshold is a pre-set critical value for the frequency change rate, used to determine whether grid frequency fluctuations require rapid adjustment.

[0066] High fluctuations in renewable energy output indicate that renewable energy generation is changing drastically in a short period of time, requiring rapid-response equipment such as electrochemical energy storage to charge and discharge quickly to suppress the impact of fluctuations on the power grid. High frequency change rate at grid connection points indicates that the power supply and demand imbalance in the power grid is rapidly intensifying, requiring frequency regulation resources such as electrochemical energy storage to respond and support frequency stability.

[0067] In an optional embodiment, the current operating status also includes load peak-valley difference data. Step S203 determines the required time scale type based on the current operating status of the new energy power station group, specifically including: Step c1: If the load peak-valley difference data is higher than the peak-valley difference threshold, determine that the current required time scale type of the new energy power station group is a long-short time mixed path.

[0068] Load peak-valley difference data refers to the difference between the maximum and minimum values ​​of the power grid load, used to measure the amplitude of power grid load fluctuations. The peak-valley difference threshold refers to a pre-set critical value for load peak-valley difference, used to determine whether the amplitude of power grid load fluctuations requires simultaneous activation of short-term regulation by electrochemical energy storage and long-term regulation by hydrogen energy storage or thermal energy storage for coordinated peak shaving.

[0069] When the peak-to-valley difference exceeds the threshold, it indicates that the load fluctuation is large and lasts for a long time. Short-term paths are needed to handle the rapidly changing parts, while long-term paths are needed to handle the discharge of electricity during the valley period and the discharge during the peak period, so as to achieve economical and efficient peak regulation in a coordinated manner.

[0070] In an optional embodiment, the current operating status also includes new energy predicted wind and solar curtailment data. Step S203 involves determining the required time scale type based on the current operating status of the new energy power station cluster, specifically including: Step d1: If the predicted wind and solar curtailment data for new energy sources is higher than the curtailment threshold, determine that the current required time scale type for the new energy power station cluster is a long-term path.

[0071] The data on predicted wind and solar curtailment refers to the amount of renewable energy generation that may not be absorbed by the grid and will be forced to be abandoned, as calculated in advance by the renewable energy power prediction system. This data is used to determine whether long-term energy storage needs to be activated. The curtailment threshold is a pre-set critical value for the predicted amount of wind and solar curtailment.

[0072] When the predicted curtailment data for wind and solar power exceeds the curtailment threshold, it indicates that a large amount of renewable energy will not be absorbed by the grid in the future. It is necessary to activate large-capacity, slow-response long-term energy storage resources for energy time shifting. Therefore, the current required time scale type is determined to be long-term path.

[0073] This embodiment provides an energy optimization control method for a new energy power station cluster. Based on a time-scale matching judgment mechanism, it can identify the current adjustment demand type of the system according to real-time or predicted data such as new energy fluctuations, frequency changes, curtailment rate, and load peak-valley difference. This allows for the on-demand deployment of short-term rapid response resources and long-term response resources, avoiding control failures caused by resource mismatch.

[0074] According to an embodiment of the present invention, an optimized control model training method is provided. Figure 3 This is a flowchart illustrating the optimized control model training method according to an embodiment of the present invention, as shown below. Figure 3 As shown, it specifically includes: Step S301: Using each power generation group and grid connection node in the new energy power plant cluster as nodes, and the energy transmission path between each power generation group and grid connection node as edges, construct a directed weighted graph of the new energy power plant cluster.

[0075] This process is the same as step S201. Please refer to step S201 for details, which will not be repeated here.

[0076] Step S302: Based on the time scale of the generation uncertainty of each power group in the directed weight graph, determine the time scale classification of each transmission path.

[0077] This process is the same as step S202. Please refer to step S202 for details, which will not be repeated here.

[0078] Step S303: Construct multiple training samples based on the historical operation data of the new energy power station group at different times. The training samples include node features, edge features, dynamic adjacency matrix, and actual energy transmission values ​​of each energy transmission path determined based on the historical operation data.

[0079] Training samples refer to a complete set of data units extracted from the historical operation data of new energy power plant clusters, which are used to train and optimize the control model.

[0080] For example, each training sample may include node features, edge features, dynamic adjacency matrix, and actual energy transmission values ​​of each energy transmission path. When constructing samples, hardware topology constraints such as input-output power balance of the same node and equipment safe operating range constraints may also be considered.

[0081] Step S304: Input the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained to obtain the scheduling instructions corresponding to each sample. The scheduling instructions contain the energy instruction transmission values ​​of each energy transmission path.

[0082] The scheduling instruction corresponding to each sample refers to the scheduling instruction value calculated and output by the model after the training sample is input into the optimization control model to be trained. It includes the expected power instruction transmission value on each energy transmission path and is used to compare with the actual historical scheduling instructions in the sample.

[0083] Each sample is input into the model to be trained to obtain the corresponding scheduling instruction method, which is the same as step S205 and will not be repeated here.

[0084] Step S305: Calculate the loss value by classifying the actual energy transmission value, energy command transmission value, and time scale of each energy transmission path for each energy transmission path.

[0085] Actual energy transfer value refers to the actual energy transfer amount measured and recorded on each energy transfer path during historical operation. Energy command transfer value refers to the power command value calculated and output by the optimized control model based on input data, which is expected to be executed on each energy transfer path. The time-scale classification of each energy transfer path refers to pre-classifying them into short-time paths, long-time paths, or mixed paths based on the response speed, regulation capability, and applicable scenarios of the power sources connected to each path. This is used for differentiated calculations for different paths during subsequent model training. The loss value measures the degree of difference between the energy command transfer value predicted by the optimized control model and the historical actual transfer value, reflecting the prediction accuracy of the model in the current training state. The model is trained by minimizing the loss value using the backpropagation algorithm.

[0086] For example, the optimizer can be an Adaptive Moment Estimation (Adam) optimizer, which can employ time series cross-validation to prevent future information leakage.

[0087] Step S306: If the loss value does not meet the preset conditions, update the model to be trained according to the loss value, and return to the step of inputting the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained until the loss value meets the preset conditions and the optimized control model is obtained.

[0088] For example, the preset conditions could be that the loss value is less than a preset loss value threshold or that the loss value has stably converged. Updating the model to be trained based on the loss value means calculating the gradient through backpropagation and adjusting the model parameters along the gradient descent direction. Repeating this process allows the model's predictions to gradually approach the true values.

[0089] This embodiment provides an optimized control model training method that constructs training samples using historical operating data and introduces time-scale classification as input to train the model. The path is divided into paths of different time scales, enabling the model to identify the response characteristics of different regulatory resources. This ensures that fast-response energy storage devices can smooth out short-term fluctuations and frequency regulation, while slow-response energy storage devices can handle the absorption of abandoned power and energy time shift, thereby achieving accurate regulation of electricity and avoiding resource misallocation.

[0090] In an optional embodiment, step 305, which calculates the loss value using the actual energy transmission value, energy command transmission value, and time scale classification of each energy transmission path, specifically includes: Step e1: Calculate the energy conversion efficiency loss value based on the real-time conversion efficiency data and transmitted energy data of each edge in each energy transmission path.

[0091] Energy conversion efficiency loss refers to the cumulative efficiency loss during energy conversion in multiple directions. It guides the model to prioritize high-efficiency paths and reduce energy loss. Real-time conversion efficiency data refers to the actual efficiency value of energy conversion stages on each energy transmission path. Transmitted energy data refers to the cumulative amount of electricity actually transmitted through each energy transmission path.

[0092] Step e2: Calculate the response sensitivity loss value based on the command value and actual value of each energy transmission path, combined with the system's physical constraints.

[0093] The response sensitivity loss value is used to measure the degree of closeness between the energy command transmission value output by the model in response to the scheduling command and the actual energy transmission value measured after actual execution, ensuring that the short-time path power deviation is small and the long-time path energy deviation is small.

[0094] Step e3: Calculate the path switching smoothness loss value based on the current selection state vector of each energy transmission path and the selection state vector after the preset time step.

[0095] The path switching smoothness loss value, based on the difference between the current moment and the path selection state vector after a preset time step, penalizes frequent path switching, thereby extending equipment lifespan. The selection state vector is an identifier that records whether each energy transmission path is selected for execution at the current moment. The preset time step is used to compare the time intervals of path selection state changes to measure the frequency of path switching and penalize unnecessary oscillations.

[0096] Step e4: Combine the energy conversion efficiency loss value, response sensitivity loss value, and path switching smoothness loss value with preset hyperparameters to calculate the loss value.

[0097] Preset hyperparameters are configuration parameters that are set in advance before model training and are used to control the balance of the loss function during model training.

[0098] For example, the loss value can be calculated according to the following formula:

[0099] in, This is the total loss value. This represents the energy conversion efficiency loss value. For the response sensitivity loss value, This represents the path switching smoothness loss value. The energy conversion efficiency loss hyperparameter, To respond to the sensitivity loss hyperparameter, This is the hyperparameter for path switching smoothness loss.

[0100] As an example, if energy conversion efficiency loss needs to be considered more, the hyperparameter can be... , , If the loss of path switching smoothness needs to be considered more, the hyperparameter can be... , , This is just an example and is not intended to be restrictive.

[0101] This embodiment provides an optimized control model training method that, by considering energy conversion efficiency loss, response sensitivity loss, and path switching smoothness loss, enables the model to simultaneously achieve efficiency, response speed, and smoothness during training, thus realizing multi-objective collaborative optimization.

[0102] In an optional embodiment, step e1, calculating the energy conversion efficiency loss value based on the real-time conversion efficiency data and transmitted energy data of each edge in each energy transmission path, specifically includes: Step f1: Determine the energy loss data of each energy transmission path based on the sum of the real-time conversion efficiency data of each edge in each energy transmission path.

[0103] For example, it can be based on Calculate the energy loss data for each energy transmission path, where, Let e ​​be the real-time conversion efficiency of edge e at time t. Let be the product of the efficiencies of all edges on path p, representing the total transformation efficiency of the entire path. Let p be the energy loss rate of path p.

[0104] Step f2: Determine the energy conversion efficiency loss of each energy transmission path based on the product of the energy loss data and the transmitted energy data.

[0105] For example, it can be based on Calculate the energy conversion efficiency loss for each energy transport path. Among them, The energy loss rate of path p. The energy value transmitted for path p.

[0106] Step f3 involves summing up the energy conversion efficiency loss of each energy transmission path to obtain the energy conversion efficiency loss value.

[0107] For example, the energy conversion efficiency loss can be calculated using the following formula:

[0108] in, This represents the energy conversion efficiency loss value. Let e ​​be the real-time conversion efficiency of edge e at time t. Let be the product of the efficiencies of all edges on path p, representing the total transformation efficiency of the entire path. The energy loss rate of path p. The energy value transmitted for path p. This is the set of all energy transfer paths selected for execution at the current moment.

[0109] In an optional embodiment, step e2, calculating the response sensitivity loss value based on the command value and the actual value of each energy transmission path, combined with the system's physical constraints, specifically includes: Step g1 involves summing the absolute difference between the power command value and the actual power value of each short-timescale energy transmission path and the ratio of the ramp rate constraint, and then determining the short-term impact error by combining the short-timescale weights.

[0110] For example, it can be based on Determine the short-term impact error, where, For short-term dynamic weights, The path set is divided into short-time scales. Let be the commanded power value on path p at time t. Let be the actual power value on path p at time t. The slope rate is a constraint.

[0111] Step g2 involves summing the ratio of the absolute difference between the energy command value and the actual energy value of the energy transmission path at each long time scale to the capacity constraint, and then determining the long-term impact error by combining the long time scale weights.

[0112] For example, it can be based on Determine the long-term impact error, among which, For long-term dynamic weights, For a long-term scale partitioned set of paths, Let be the planned value of the energy flow on path p at time t. Let be the actual value of the energy flow on path p at time t. This is a capacity constraint.

[0113] Step g3 involves summing the short-term and long-term impact errors to obtain the response sensitivity loss value.

[0114] Because fast response requires accurate instantaneous power tracking, the short-time path formula focuses on power deviation; because energy time shift requires accurate cumulative charge, the long-time path formula focuses on energy deviation.

[0115] For example, the response sensitivity loss value can be calculated according to the following formula:

[0116] in, For the response sensitivity loss value, For short-term impact error terms, This represents the long-term impact on the error term. As an example, the dynamic weights for short-term and long-term timescales can be adjusted according to the different levels of importance the system places on short-term response and long-term energy storage. For instance, the dynamic weight for short-term timescales could be 0.4, and the dynamic weight for long-term timescales could be 0.6; no restrictions are imposed here.

[0117] In an optional embodiment, step e3, based on the selection state vector of each energy transmission path at the current moment and the selection state vector after a preset time step, calculates the path switching smoothness loss value, specifically including: Step h1: Determine the timing smoothness penalty term based on the difference between the selected state vector of each energy transmission path at the current moment and the selected state vector after the preset time step.

[0118] The temporal smoothness penalty term is used to penalize the change in the selection state of the same path at different times, and to suppress the frequent switching of paths over time.

[0119] For example, it can be based on Calculations are performed, in which, The set of all energy transport paths. Let p be the choice state vector at time t. For path p at time The choice of state vector, It represents the square of the L2 norm, which is the sum of the squares of the differences between the components of the vector.

[0120] Step h2: Determine the path consistency penalty term based on the difference between the selection state vector of each energy transmission path and the selection state vectors of all paths except the current path, combined with the path consistency penalty weight.

[0121] The path consistency penalty is used to penalize the differences in the selected states of different paths at the same time, thereby promoting coordination and consistency among paths.

[0122] For example, it can be based on Calculations are performed, in which, As an energy transfer path, This is another energy transfer path besides p. The set of all energy transport paths. Let p be the choice state vector at time t. For path The choice of state vector at the same time t This is the path consistency penalty weight coefficient, used to adjust the importance of this item in the total loss.

[0123] Step h3 involves summing the temporal smoothness penalty term and the path consistency penalty term to obtain the path switching smoothness loss value.

[0124] For example, the path switching smoothness loss value can be calculated according to the following formula:

[0125] in, This represents the path switching smoothness loss value. This is a time-series smoothness penalty term. This is a path consistency penalty.

[0126] This embodiment also provides an energy optimization control device for a new energy power station cluster. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0127] This embodiment provides an energy optimization control device for a new energy power station cluster, such as... Figure 4 As shown, it includes: The graph generation module 401 is used to construct a directed weighted graph of the new energy power plant cluster, using each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges.

[0128] The path scale classification module 402 is used to determine the time scale classification of each transmission path based on the time scale of the generation uncertainty of each power group in the directed weight graph.

[0129] The scale confirmation module 403 is used to determine the required time scale type based on the current operating status of the new energy power station group.

[0130] The matrix generation module 404 is used to determine the edges that need to be retained in the directed weight graph based on the current required time scale type and the time scale classification of each transmission path, and to form a dynamic adjacency matrix of the directed weight graph.

[0131] The instruction generation module 405 is used to input the node features, edge features, and dynamic adjacency matrix of the directed weighted graph into a pre-established optimization control model for calculation to obtain the scheduling instructions of the new energy power station group.

[0132] The instruction control module 406 is used to control the energy transmission of the new energy power plant group using scheduling instructions.

[0133] The energy optimization control device for a new energy power station cluster provided in this embodiment of the invention can execute the energy optimization control method for a new energy power station cluster provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments, and will not be repeated here.

[0134] This embodiment also provides an optimized control model training device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. This embodiment provides an energy optimization control device for a new energy power station cluster, such as... Figure 5 As shown, it includes: The directed weighted graph generation module 501 is used to construct a directed weighted graph of the new energy power plant group, with each power generation group and grid connection node in the new energy power plant group as nodes and the energy transmission path between each power generation group and grid connection node as edges. The scale classification module 502 is used to determine the time scale classification of each transmission path based on the time scale of the power generation uncertainty of each power group in the directed weight graph. The sample construction module 503 is used to construct multiple training samples based on the historical operation data of the new energy power station group at different times. The training samples include node features, edge features, dynamic adjacency matrix and actual energy transmission values ​​of each energy transmission path determined based on the historical operation data. The instruction generation module 504 is used to input the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained to obtain the scheduling instructions corresponding to each sample. The scheduling instructions contain the energy instruction transmission values ​​of each energy transmission path. The loss calculation module 505 is used to calculate the loss value by classifying the actual energy transmission value, energy command transmission value and time scale of each energy transmission path for each energy transmission path. The training optimization module 506 is used to update the model to be trained based on the loss value if the loss value does not meet the preset conditions. It returns to the step of inputting the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained until the loss value does not meet the preset conditions, thus obtaining the optimized control model.

[0135] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0136] The following is a detailed reference. Figure 6This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0137] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0138] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the energy optimization control method for a new energy power station cluster according to embodiments of the present invention.

[0139] Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0140] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, it implements the energy optimization control method for a new energy power station cluster shown in the above embodiments.

[0141] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0142] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for energy optimization control of a new energy power station cluster, characterized in that, The method includes: A directed weighted graph of the new energy power plant cluster is constructed, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. Based on the time scale of the generation uncertainty of each power group in the directed weight graph, the time scale classification of each transmission path is determined. The required time scale type is determined based on the current operating status of the new energy power station cluster; The edges that need to be retained in the directed weight graph are determined based on the current required time scale type and the time scale classification of each transmission path, forming the dynamic adjacency matrix of the directed weight graph. The node features, edge features, and dynamic adjacency matrix of the directed weighted graph are input into a pre-established optimization control model for calculation to obtain the scheduling instructions for the new energy power station group. The energy transmission of the new energy power station group is controlled using the scheduling instructions.

2. The method according to claim 1, characterized in that, The method further includes: The update interval is determined based on the link delay of the control system and the instruction response time of the scheduling command; The time scale type required to update the new energy power station cluster is determined according to the update interval.

3. The method according to claim 1, characterized in that, The edge features include edge weight data, which is determined based on one or more of the following: the power supply response time, energy conversion efficiency, and real-time grid electricity price of the transmission path corresponding to the edge.

4. The method according to claim 3, characterized in that, The current operating status includes new energy power output fluctuation data, grid connection point frequency change rate data, load peak-valley difference data, and predicted wind and solar curtailment data. Determining the required time scale type based on the current operating status of the new energy power station cluster includes: If the power output fluctuation data of the new energy source is higher than the power output fluctuation threshold, or the frequency change rate data of the grid connection point is higher than the frequency change threshold, the current required time scale type of the new energy power station group is determined to be short-time path. If the load peak-valley difference data is higher than the peak-valley difference threshold, it is determined that the current required time scale type of the new energy power station group is a long-short time mixed path; If the predicted wind and solar curtailment data for new energy sources is higher than the curtailment threshold, the current required time scale type for the new energy power station cluster is determined to be a long-term path.

5. A method for training the optimization control model as described in claim 1, characterized in that, The method includes: A directed weighted graph of the new energy power plant cluster is constructed, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. Based on the time scale of the generation uncertainty of each power group in the directed weight graph, the time scale classification of each transmission path is determined. Multiple training samples are constructed based on the historical operation data of the new energy power station cluster at different times. The training samples include node features, edge features, dynamic adjacency matrix, and actual energy transmission values ​​of each energy transmission path determined based on the historical operation data. The node features, edge features, and dynamic adjacency matrix of each sample are input into the model to be trained to obtain the scheduling instructions corresponding to each sample. The scheduling instructions include the energy instruction transmission values ​​of each energy transmission path. The loss value is calculated by classifying and categorizing the actual energy transmission value, energy command transmission value, and time scale of each energy transmission path for each energy transmission path. If the loss value does not meet the preset conditions, the model to be trained is updated according to the loss value, and the step of inputting the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained is returned until the loss value does not meet the preset conditions, and the optimized control model is obtained.

6. The method according to claim 5, characterized in that, The calculation of loss values ​​using the actual energy transmission value, energy command transmission value, and time scale classification of each energy transmission path includes: The energy conversion efficiency loss value is calculated based on the real-time conversion efficiency data and transmitted energy data of each edge in each energy transmission path; The response sensitivity loss value is calculated based on the command value and the actual value of each energy transmission path, combined with the physical constraints of the system. Based on the current selection state vector and the selection state vector after a preset time step for each energy transmission path, calculate the path switching smoothness loss value. The loss value is obtained by fusing the energy conversion efficiency loss value, response sensitivity loss value, and path switching smoothness loss value with preset hyperparameters.

7. The method according to claim 6, characterized in that, The calculation of energy conversion efficiency loss values ​​based on the real-time conversion efficiency data and transmitted energy data of each edge in each energy transmission path includes: Based on the sum of the real-time conversion efficiency data of each edge in each energy transmission path, the energy loss data of each energy transmission path is determined. The energy conversion efficiency loss of each energy transmission path is determined by multiplying the energy loss data with the transmitted energy data. The energy conversion efficiency loss of each energy transmission path is summed to obtain the energy conversion efficiency loss value.

8. The method according to claim 6, characterized in that, The command values ​​for each energy transmission path include power command values ​​and energy command values; the actual values ​​for each energy transmission path include actual power values ​​and actual energy values; the system physical constraints include ramp rate constraints and capacity constraints; and the calculation of the response sensitivity loss value based on the command values ​​and actual values ​​of each energy transmission path, combined with the system physical constraints, includes: The ratio of the absolute difference between the power command value and the actual power value of the energy transmission path at each short time scale to the ramp rate constraint is summed and calculated. Combined with the short time scale weight, the short-term impact error is determined. The ratio of the absolute difference between the energy command value and the actual energy value of the energy transmission path at each long time scale to the capacity constraint is summed and calculated. Combined with the long time scale weights, the long-term impact error is determined. The short-term impact error and the long-term impact error are summed to obtain the response sensitivity loss value.

9. The method according to claim 6, characterized in that, The step of calculating the path switching smoothness loss value based on the current selection state vector of each energy transmission path and the selection state vector after a preset time step includes: The timing smoothness penalty term is determined based on the difference between the current selection state vector of each energy transmission path and the selection state vector after a preset time step. The path consistency penalty term is determined based on the difference between the selection state vector of each energy transmission path and the selection state vector of all paths except the current path, combined with the path consistency penalty weight. The path switching smoothness loss value is obtained by summing the time-series smoothness penalty term and the path consistency penalty term.

10. An energy optimization control device for a new energy power station cluster, characterized in that, The device includes: The graph generation module is used to construct a directed weighted graph of the new energy power plant cluster, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. The path scale classification module is used to determine the time scale classification of each transmission path based on the time scale of the power generation uncertainty of each power group in the directed weight graph. The scale confirmation module is used to determine the required time scale type based on the current operating status of the new energy power station group. The matrix generation module is used to determine the edges that need to be retained in the directed weight graph based on the current required time scale type and the time scale classification of each transmission path, and to form the dynamic adjacency matrix of the directed weight graph. The instruction generation module is used to input the node features, edge features, and dynamic adjacency matrix of the directed weighted graph into a pre-established optimization control model for calculation to obtain the scheduling instructions of the new energy power station group. The instruction control module is used to control the energy transmission of the new energy power station group using the scheduling instructions.

11. An optimized control model training device, characterized in that, The device includes: The directed weighted graph generation module is used to construct a directed weighted graph of the new energy power plant cluster, with each power generation group and grid connection node in the new energy power plant cluster as nodes and the energy transmission path between each power generation group and grid connection node as edges. The scale classification module is used to determine the time scale classification of each transmission path based on the time scale of the power generation uncertainty of each power group in the directed weight graph. The sample construction module is used to construct multiple training samples based on the historical operation data of the new energy power station group at different times. The training samples include node features, edge features, dynamic adjacency matrix and actual energy transmission values ​​of each energy transmission path determined based on the historical operation data. The instruction generation module is used to input the node features, edge features, and dynamic adjacency matrix of each sample into the model to be trained to obtain the scheduling instruction corresponding to each sample. The scheduling instruction contains the energy instruction transmission value of each energy transmission path. The loss calculation module is used to calculate the loss value by classifying and classifying the actual energy transmission value, energy command transmission value and time scale of each energy transmission path for each energy transmission path. The training optimization module is used to update the model to be trained according to the loss value if the loss value does not meet the preset conditions, and return to the step of inputting the node features, edge features and dynamic adjacency matrix of each sample into the model to be trained, until the loss value does not meet the preset conditions, and thus obtain the optimized control model.

12. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform either the energy optimization control method for a new energy power station cluster as described in any one of claims 1 to 6, or the optimization control model training method as described in any one of claims 7 to 11.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which are used to cause the computer to execute the energy optimization control method for a new energy power station group according to any one of claims 1 to 6, or the optimization control model training method according to any one of claims 7 to 11.

14. A computer program product, characterized in that, The method includes computer instructions for causing a computer to execute the energy optimization control method for a new energy power station cluster according to any one of claims 1 to 6, or the optimization control model training method according to any one of claims 7 to 11.

Images