A method, system, equipment, and medium for predicting aggregated load trends in virtual power plants.

Abstract

This invention relates to a method, system, equipment, and medium for predicting aggregated load trends in virtual power plants, belonging to the field of power system management technology. The method includes: collecting historical source-grid-load-storage data of the virtual power plant; constructing source-grid-load-storage characteristics based on the historical source-grid-load-storage data, including node operating status, changes in node resource configuration, and inter-node connectivity; constructing a virtual power plant aggregated load prediction model including a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer, and training it using the source-grid-load-storage characteristics; collecting real-time source-grid-load-storage data of the virtual power plant, and predicting future load data of the virtual power plant using the trained model. This invention can simultaneously characterize changes in virtual power plant resource combinations, energy connections between nodes, and energy storage operation constraints, improving the stability and accuracy of aggregated load trend prediction.

Description

Technical Field

[0001] This invention relates to a method, system, equipment, and medium for predicting aggregated load trends in virtual power plants, belonging to the field of power system management technology. Background Technology

[0002] Virtual power plants typically aggregate distributed power sources, controllable loads, and energy storage resources to achieve unified planning, rolling optimization, and real-time dispatch. To ensure the feasibility and economy of day-ahead, intraday, and real-time control strategies, it is often necessary to forecast the aggregated load (or equivalent net load) for future periods.

[0003] Existing load forecasting technologies can be broadly categorized into: statistical models based on historical curves (such as regression and ARIMA), machine learning models based on feature engineering (such as support vector regression, random forest, and gradient boosting), and deep time series models represented by RNN / LSTM / Transformer. These methods are generally effective in scenarios with "single load points and relatively stable electricity consumption behavior," but in virtual power plant scenarios, the forecasting objects exhibit significant complexity. The aggregated object is composed of multiple heterogeneous resources and is affected by factors such as weather, price, industrial and commercial conditions, demand response strategies, distributed power fluctuations, and energy storage charging and discharging strategies, resulting in a stronger non-stationarity and abrupt change in the load sequence. The number and combination of available resources can change over time (temporary node access / exit, changes in available capacity, adjustments to the allocation ratio, etc.), making the input space and statistical distribution of the "same virtual power plant" prone to drift. There are electrical / energy couplings and policy linkages between resources (such as peak shaving and valley filling, tracking commands, energy storage boundary constraints, etc.). If only simple aggregate quantities or single sequence models are used, it is often difficult to stably characterize the aggregation mechanism. Engineering data is often accompanied by problems such as missing data, delays, outliers, and inconsistent measurement methods, which further amplify the risk of prediction errors affecting scheduling decisions.

[0004] Therefore, how to achieve stable trend prediction of virtual power plant aggregated load under the condition of multi-source heterogeneous data, taking into account the time series pattern, resource combination changes and constraint feasibility, is still a direction that needs to be improved. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a method, system, equipment, and medium for predicting aggregated load trends in virtual power plants.

[0006] The technical solution of the present invention is as follows: On the one hand, the present invention provides a method for predicting the aggregated load trend of a virtual power plant, comprising the following steps: Collect historical power plant source-grid-load-storage data; Source-grid-load-storage characteristics are constructed based on historical source-grid-load-storage data to characterize node operating status, changes in node resource configuration, and energy connections between nodes. These characteristics include time-series features and structural features. A virtual power plant aggregated load prediction model is constructed, which includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer. The node aggregation layer is used to extract the node aggregation features of each source-grid-load-storage node included in the virtual power plant at each time step in the time series features. The graph coding layer is used to encode the structural features. The energy storage coding layer is used to encode the energy storage node operation data. The output layer generates the virtual power plant future load data prediction results based on the node aggregation features, the coding results of the graph structural data, and the coding results of the energy storage node operation data. The virtual power plant aggregated load prediction model is trained using source-grid-load-storage characteristics; Collect real-time source-grid-load-storage data of the virtual power plant, and predict future load data of the virtual power plant through the trained virtual power plant aggregated load prediction model.

[0007] Preferably, the source-grid-load-storage characteristics include temporal characteristics and structural characteristics; The time-series characteristics specifically include the operation data of each source-grid-load-storage node included in the virtual power plant at different times, as well as the node resource configuration data of the virtual power plant at different times. The node resource configuration data of the virtual power plant is specifically a feature vector constructed from the energy allocation of each source-grid-load-storage node of the virtual power plant at that moment. The structural features specifically include graph structure data that constructs the connections between the various source-grid-load-storage nodes included in the virtual power plant at different times. In the graph structure data, each node represents a source-grid-load-storage node, the node attributes are the configuration data of that node, and each edge represents the connection relationship between nodes.

[0008] Preferably, the virtual power plant aggregated load prediction model includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer.

[0009] Preferably, the node aggregation layer is used to extract node aggregation features at each time step in the feature vector time series; The node aggregation layer includes a pre-encoding unit, a gating unit, a node data correction unit, and an aggregation unit; The pre-encoding unit is used to encode the operating data of each source-grid-load-storage node included in the virtual power plant at different times, as well as the node resource configuration data of the virtual power plant at different times. The gating unit includes amplitude gating, bias gating, and phase gating, as shown in the following formula: ; ; ; in: express Time Node Amplitude threshold value; express Time Node The bias gate value; express Time Node The phase gate value; Indicates the activation function; express Time Node The encoded value of the running data; express Time node resource configuration data encoding value; express Time node resource configuration data encoding value and The difference in the encoded values ​​of the resource configuration data at each time node; , , Represents the learnable weight matrix; , , Represents a learnable bias vector; The node data correction unit is specifically shown in the following formula: ; in: Indicates the revised version Time Node The encoded value of the running data; Indicates will Project to Same-dimensional space; This indicates element-wise multiplication; The specific aggregation unit is shown in the following formula: ; ; ; in: express The node aggregation features output by the time-mapping aggregation unit; express The set of nodes that are constantly included in the virtual power plant; express Time Node Normalized aggregate weights; express Time Node The aggregate fraction; Indicates the temperature coefficient; express Time Node The aggregate fraction; Represents a learnable projection vector; Indicates the transpose operation; This represents the weight matrix of the scoring network; express Time Node Energy supply allocation; express Time Node Energy supply allocation and Time Node The difference in the amount of energy supplied.

[0010] Preferably, the graph coding layer is used to encode graph structure data, specifically including: For any given moment in the graph structure data, the weight of each edge in the graph structure data is encoded as follows: ; ; ; ; in: Represents a node With nodes The basic encoding of the edge weights connecting the edges; This represents the basic encoded learnable vector; Represents a node Attributes; Represents a node Attributes; Represents a node With nodes The difference in output power between them; Represents a node With nodes The residual suppression weights between; This represents the residual suppression strength coefficient; Represents a node The energy balance residual; Indicates the inflow node Energy; Indicates outflow node Energy; Represents a node The converter loss; Represents a node With nodes Encoding of the weights of the connecting edges; Encoding node attributes based on edge weight encoding involves the following steps: The residual signature vector of each node is constructed by concatenating the energy balance residual of each node, the mean of the historical energy balance residual over a preset time span, and the variance of the historical energy balance residual over a preset time span. A dual-branch weight calculation unit is set up to calculate the basic encoding weight and resource encoding weight of node attributes respectively, as shown in the following formula: ; ; in: Represents a node The basic coding weights; Represents a node Resource coding weights; , This represents a learnable transformation mapping function; This represents the encoded value of the node resource configuration data at that moment; The basic encoding of each node's attributes is calculated based on the residual signature vector of each node, as shown in the following formula: ; in: Represents a node The basic encoded value of the attribute; express function; The basic encoding of node attributes is represented by a learnable weight matrix; The basic encoding of node attributes is a learnable bias vector; Represents a node The residual signature vector; The node attribute code is calculated based on the basic attribute code, basic code weight, and resource code weight of each node, as shown in the following formula: ; in: Represents a node Attribute encoding.

[0011] Preferably, the energy storage coding layer is used to encode the operating data of the energy storage nodes, and the specific steps are as follows: The upper limit of the power projection is applied to the difference between the charging and discharging power of the energy storage node, as shown in the following formula: ; in: express Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates energy storage node The upper limit of charging and discharging power; express Real-time energy storage nodes The difference in charging and discharging power; The upper limit projection of the power difference between the charging and discharging power of the energy storage node is corrected by a ramping process, as shown in the following formula: ; in: Indicates the result after hill climbing correction Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates the result after hill climbing correction Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates energy storage node The charging efficiency; Indicates energy storage node The discharge efficiency; express Time to The time span of a moment; The energy storage node operation data and the upper limit of the power difference between charging and discharging power after ramp correction are spliced ​​together and then input into the encoder to obtain the energy storage node operation data code.

[0012] Preferably, the output layer is constructed based on a multilayer perceptron. The outputs of the node aggregation layer, graph coding layer, and energy storage coding layer at different historical moments are spliced ​​together and then input into the multilayer perceptron. The multilayer perceptron outputs virtual power plant load forecast data for a preset future time step. The virtual power plant load forecast data for all future time steps are combined to obtain the virtual power plant predicted load trend.

[0013] On the other hand, the present invention also provides a virtual power plant aggregated load trend prediction system, including a data acquisition module, a feature construction module, a load prediction model construction module, a model training module, and a load prediction module; The data acquisition module is used to collect historical source-grid-load-storage data of the virtual power plant, and to collect real-time source-grid-load-storage data of the virtual power plant. The feature construction module is used to construct source-grid-load-storage features based on historical source-grid-load-storage data. The source-grid-load-storage features include time-series features and structural features. The load forecasting model construction module is used to construct a virtual power plant aggregated load forecasting model, which includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer. The node aggregation layer is used to extract the node aggregation features of each source-grid-load-storage node included in the virtual power plant at each time step in the time series features. The graph coding layer is used to encode the structural features. The energy storage coding layer is used to encode the energy storage node operation data. The output layer is used to generate the virtual power plant future load data prediction results based on the node aggregation features, the coding results of the graph structural data, and the coding results of the energy storage node operation data. The model training module is used to train the virtual power plant aggregated load prediction model using source-grid-load-storage characteristics; The load forecasting module is used to predict the future load data of the virtual power plant based on the real-time source-grid-load-storage data of the virtual power plant using the trained virtual power plant aggregated load forecasting model.

[0014] In another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the present invention.

[0015] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described in the present invention.

[0016] The present invention has the following beneficial effects: 1. This invention constructs a source-grid-load-storage feature that simultaneously includes temporal and structural features, and uses graph structure data formed by node operation data, node resource configuration data, and inter-node connection relationships of the virtual power plant at different times for training and prediction. This enables the model to more fully characterize the aggregation mechanism in scenarios where the node set and resource allocation change dynamically, thereby improving the stability and accuracy of the virtual power plant's future load trend prediction. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] 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.

[0019] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0020] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0022] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0023] See Figure 1 In some embodiments, a method for predicting aggregated load trends in virtual power plants is proposed, comprising the following steps: Collect historical power plant source-grid-load-storage data; Source-grid-load-storage characteristics are constructed based on historical source-grid-load-storage data to characterize node operating status, changes in node resource configuration, and energy connections between nodes. These characteristics include time-series features and structural features. A virtual power plant aggregated load prediction model is constructed, which includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer. The node aggregation layer is used to extract the node aggregation features of each source-grid-load-storage node included in the virtual power plant at each time step in the time series features. The graph coding layer is used to encode the structural features. The energy storage coding layer is used to encode the energy storage node operation data. The output layer generates the virtual power plant future load data prediction results based on the node aggregation features, the coding results of the graph structural data, and the coding results of the energy storage node operation data. The virtual power plant aggregated load prediction model is trained using source-grid-load-storage characteristics; Collect real-time source-grid-load-storage data of the virtual power plant, and predict future load data of the virtual power plant through the trained virtual power plant aggregated load prediction model.

[0024] In one specific embodiment, the source-grid-load-storage data are respectively source-side node operation data, grid-side node operation data, load node operation data, and energy storage node operation data; The source-side node operating data includes output active power, adjustable unit power, and meteorological data; The grid-side node operation data includes grid connection point voltage, frequency, power flow data, and power purchase and sale plan data. The load node operation data includes load demand. The energy storage node's operating data includes charging and discharging power, charging and discharging efficiency, and the upper limit of charging and discharging power.

[0025] In some embodiments, the source-grid-load-storage characteristics include time-series characteristics and structural characteristics; The time-series characteristics specifically include the operation data of each source-grid-load-storage node included in the virtual power plant at different times, as well as the node resource configuration data of the virtual power plant at different times. The node resource configuration data of the virtual power plant is specifically a feature vector constructed from the energy allocation of each source-grid-load-storage node of the virtual power plant at that moment. The structural features specifically include graph structure data that constructs the connections between the various source-grid-load-storage nodes included in the virtual power plant at different times. In the graph structure data, each node represents a source-grid-load-storage node, the node attributes are the configuration data of that node, and each edge represents the connection relationship between nodes.

[0026] In one specific embodiment, for the energy allocation of each source-grid-load-storage node, the energy supplied by the energy supply node is positive, and the energy supplied by the energy dissipation node is negative.

[0027] In a specific embodiment, the configuration data of the source-grid-load-storage node is as follows: Source-side node configuration data: resource type (photovoltaic / wind turbine / gas turbine / diesel engine / waste heat, etc.), rated capacity, rated voltage level, minimum output and maximum output; Network-side node configuration data: feeder number, topology connection relationship (adjacency information), capacity limit, voltage level, and transformer capacity; Load node configuration data: load type (industrial and commercial / buildings / charging piles / cooling plants / production lines, etc.) and maximum demand; Energy storage node configuration data: energy storage type (battery / supercapacitor / hybrid energy storage), rated energy, rated power, and upper and lower limits of SOC.

[0028] In some embodiments, the virtual power plant aggregated load prediction model includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer.

[0029] In some embodiments, the node aggregation layer is used to extract node aggregation features at each time step in the feature vector time series; The node aggregation layer includes a pre-encoding unit, a gating unit, a node data correction unit, and an aggregation unit; The pre-encoding unit is used to encode the operating data of each source-grid-load-storage node included in the virtual power plant at different times, as well as the node resource configuration data of the virtual power plant at different times. The gating unit includes amplitude gating, bias gating, and phase gating, as shown in the following formula: ; ; ; in: express Time Node Amplitude threshold value; express Time Node The bias gate value; express Time Node The phase gate value; Indicates the activation function; express Time Node The encoded value of the running data; express Time node resource configuration data encoding value; express Time node resource configuration data encoding value and The difference in the encoded values ​​of the resource configuration data at each time node; , , Represents the learnable weight matrix; , , Represents a learnable bias vector; The node data correction unit is specifically shown in the following formula: ; in: Indicates the revised version Time Node The encoded value of the running data; Indicates will Project to Same-dimensional space; This indicates element-wise multiplication; The specific aggregation unit is shown in the following formula: ; ; ; in: express The node aggregation features output by the time-mapping aggregation unit; express The set of nodes that are constantly included in the virtual power plant; express Time Node Normalized aggregate weights; express Time Node The aggregate fraction; Indicates the temperature coefficient; express Time Node The aggregate fraction; Represents a learnable projection vector; Indicates the transpose operation; This represents the weight matrix of the scoring network; express Time Node Energy supply allocation; express Time Node Energy supply allocation and Time Node The difference in the amount of energy supplied.

[0030] In some embodiments, the graph coding layer is used to encode graph structure data, specifically including: For any given moment in the graph structure data, the weight of each edge in the graph structure data is encoded as follows: ; ; ; ; in: Represents a node With nodes The basic encoding of the edge weights connecting the edges; This represents the basic encoded learnable vector; Represents a node Attributes; Represents a node Attributes; Represents a node With nodes The difference in output power between them; Represents a node With nodes The residual suppression weights between; This represents the residual suppression strength coefficient; Represents a node The energy balance residual; Indicates the inflow node Energy; Indicates outflow node Energy; Represents a node The converter loss; Represents a node With nodes Encoding of the weights of the connecting edges; Encoding node attributes based on edge weight encoding involves the following steps: The residual signature vector of each node is constructed by concatenating the energy balance residual of each node, the mean of the historical energy balance residual over a preset time span, and the variance of the historical energy balance residual over a preset time span. A dual-branch weight calculation unit is set up to calculate the basic encoding weight and resource encoding weight of node attributes respectively, as shown in the following formula: ; ; in: Represents a node The basic coding weights; Represents a node Resource coding weights; , This represents a learnable transformation mapping function; This represents the encoded value of the node resource configuration data at that moment; The basic encoding of each node's attributes is calculated based on the residual signature vector of each node, as shown in the following formula: ; in: Represents a node The basic encoded value of the attribute; express function; The basic encoding of node attributes is represented by a learnable weight matrix; The basic encoding of node attributes is a learnable bias vector; Represents a node The residual signature vector; The node attribute code is calculated based on the basic attribute code, basic code weight, and resource code weight of each node, as shown in the following formula: ; in: Represents a node Attribute encoding.

[0031] In some embodiments, the energy storage coding layer is used to encode the operating data of the energy storage node, and the specific steps are as follows: The upper limit of the power projection is applied to the difference between the charging and discharging power of the energy storage node, as shown in the following formula: ; in: express Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates energy storage node The upper limit of charging and discharging power; express Real-time energy storage nodes The difference in charging and discharging power; The upper limit projection of the power difference between the charging and discharging power of the energy storage node is corrected by a ramping process, as shown in the following formula: ; in: Indicates the result after hill climbing correction Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates the result after hill climbing correction Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates energy storage node The charging efficiency; Indicates energy storage node The discharge efficiency; express Time to The time span of a moment; The energy storage node operation data and the upper limit of the power difference between charging and discharging power after ramp correction are spliced ​​together and then input into the encoder to obtain the energy storage node operation data code.

[0032] In some embodiments, the virtual power plant aggregated load prediction model further includes a causal event coding layer for encoding the collected electricity consumption event data. The specific steps are as follows: The collected electricity consumption event data includes demand response events, power rationing events, electricity price change events, and maintenance events. The electricity consumption events at each time point are combined into an electricity consumption event data set. The node aggregation features and attribute codes of each node at each time step are concatenated to construct the main feature data sequence. The electricity consumption event data set and the main feature sequence are input into the event intensity encoding unit for event intensity encoding, as shown in the following formula: ; in: express Event intensity coding of real-time electricity consumption event data; This indicates that the event intensity encoding unit can learn the weight matrix; This indicates that the event intensity coding unit can learn a bias vector; express Real-time electricity consumption event data; express Key feature data at any given time; The electricity consumption data set is input into the event discrete time delay distribution layer to obtain the event discrete time delay distribution, as shown in the following formula:

[0033] in: This represents the maximum allowed delay steps, which is a preset hyperparameter; express The discrete time delay probability distribution from triggering to taking effect of a power consumption event. , express The probability of an event taking effect at any given time. Indicates by From the starting point to The probability of an electricity consumption event taking effect within a time delay step; The learnable weight matrix represents the discrete time delay distribution of events. The learnable bias vector represents the discrete time delay distribution of events. The intensity encoding of electricity consumption events at each time point and the discrete time delay distribution are input into the intervention input layer to obtain the intervention input quantity, as shown in the following formula: ; in: express The amount of intervention input at any given moment; express Intensity coding of electricity consumption events at any given time; express Time by From the starting point to The probability of an electricity consumption event taking effect within a time delay step; By fusing the intervention input at each time point with the main feature sequence, the causal coding of the electricity consumption event is obtained, as shown in the following formula: ; in: express Causal coding of electricity consumption events at any given moment; This means mapping the intervention input to a learnable matrix of the same dimension as the main feature sequence; Represents the normalization operator; The causal coding of electricity consumption events is used as the output of the causal event coding layer. It is then concatenated with the outputs of the node aggregation layer, graph coding layer, and energy storage coding layer and input into the multilayer sensor. The multilayer sensor outputs virtual power plant load prediction data for preset future time steps. The virtual power plant load prediction data for all future time steps are combined to obtain the virtual power plant predicted load trend.

[0034] In some embodiments, the output layer is constructed based on a multilayer perceptron. The outputs of the node aggregation layer, graph coding layer, and energy storage coding layer at different historical moments are spliced ​​together and input into the multilayer perceptron. The multilayer perceptron outputs virtual power plant load forecast data for a preset future time step. The virtual power plant load forecast data for all future time steps are combined to obtain the virtual power plant predicted load trend.

[0035] In one specific embodiment, the node aggregation features, the basic attribute codes of each node, and the energy storage node operation data codes are concatenated and then input into the multilayer perceptron.

[0036] In some embodiments, a virtual power plant aggregated load trend prediction system is proposed, including a data acquisition module, a feature construction module, a load prediction model construction module, a model training module, and a load prediction module; The data acquisition module is used to collect historical source-grid-load-storage data of the virtual power plant, and to collect real-time source-grid-load-storage data of the virtual power plant. The feature construction module is used to construct source-grid-load-storage features based on historical source-grid-load-storage data. The source-grid-load-storage features include time-series features and structural features. The load forecasting model construction module is used to construct a virtual power plant aggregated load forecasting model, which includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer. The node aggregation layer is used to extract the node aggregation features of each source-grid-load-storage node included in the virtual power plant at each time step in the time series features. The graph coding layer is used to encode the structural features. The energy storage coding layer is used to encode the energy storage node operation data. The output layer is used to generate the virtual power plant future load data prediction results based on the node aggregation features, the coding results of the graph structural data, and the coding results of the energy storage node operation data. The model training module is used to train the virtual power plant aggregated load prediction model using source-grid-load-storage characteristics; The load forecasting module is used to predict the future load data of the virtual power plant based on the real-time source-grid-load-storage data of the virtual power plant using the trained virtual power plant aggregated load forecasting model.

[0037] In some embodiments, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method as described in any embodiment of the present invention.

[0038] In some embodiments, a computer-readable storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the method as described in any embodiment of the present invention.

[0039] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0040] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0041] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0042] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0043] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting the aggregated load trend of a virtual power plant, characterized in that, Includes the following steps: Collect historical power plant source-grid-load-storage data; Source-grid-load-storage characteristics are constructed based on historical source-grid-load-storage data to characterize node operating status, changes in node resource configuration, and energy connections between nodes. These characteristics include time-series features and structural features. A virtual power plant aggregated load prediction model is constructed, which includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer. The node aggregation layer is used to extract the node aggregation features of each source-grid-load-storage node included in the virtual power plant at each time step in the time series features. The graph coding layer is used to encode the structural features. The energy storage coding layer is used to encode the energy storage node operation data. The output layer generates the virtual power plant future load data prediction results based on the node aggregation features, the coding results of the graph structural data, and the coding results of the energy storage node operation data. The virtual power plant aggregated load prediction model is trained using source-grid-load-storage characteristics; Collect real-time source-grid-load-storage data of the virtual power plant, and predict future load data of the virtual power plant through the trained virtual power plant aggregated load prediction model.

2. The virtual power plant aggregated load trend prediction method according to claim 1, characterized in that, The time-series characteristics specifically include the operation data of each source-grid-load-storage node included in the virtual power plant at different times, as well as the node resource configuration data of the virtual power plant at different times. The node resource configuration data of the virtual power plant is specifically a feature vector constructed from the energy allocation of each source-grid-load-storage node of the virtual power plant at that moment. The structural features specifically include graph structure data that constructs the connections between the various source-grid-load-storage nodes included in the virtual power plant at different times. In the graph structure data, each node represents a source-grid-load-storage node, the node attributes are the configuration data of that node, and each edge represents the connection relationship between nodes.

3. The virtual power plant aggregated load trend prediction method according to claim 2, characterized in that, The node aggregation layer is used to extract node aggregation features at each time step in the feature vector time series; The node aggregation layer includes a pre-encoding unit, a gating unit, a node data correction unit, and an aggregation unit; The pre-encoding unit is used to encode the operating data of each source-grid-load-storage node included in the virtual power plant at different times, as well as the node resource configuration data of the virtual power plant at different times. The gating unit includes amplitude gating, bias gating, and phase gating, as shown in the following formula: ； ； ； in: express Time Node Amplitude threshold value; express Time Node The bias gate value; express Time Node The phase gate value; Indicates the activation function; express Time Node The encoded value of the running data; express Time node resource configuration data encoding value; express Time node resource configuration data encoding value and The difference in the encoded values ​​of the resource configuration data at each time node; , , Represents the learnable weight matrix; , , Represents a learnable bias vector; The node data correction unit is specifically shown in the following formula: ； in: Indicates the revised version Time Node The encoded value of the running data; Indicates will Project to Same-dimensional space; This indicates element-wise multiplication; The specific aggregation unit is shown in the following formula: ； ； ； in: express The node aggregation features output by the time-mapping aggregation unit; express The set of nodes that are constantly included in the virtual power plant; express Time Node Normalized aggregate weights; express Time Node The aggregate fraction; Indicates the temperature coefficient; express Time Node The aggregate fraction; Represents a learnable projection vector; Indicates the transpose operation; Represents the weight matrix of the scoring network; express Time Node Energy supply allocation; express Time Node Energy supply allocation and Time Node The difference in the amount of energy supplied.

4. The virtual power plant aggregated load trend prediction method according to claim 3, characterized in that, The graph coding layer is used to encode graph-structured data, specifically including: For any given moment in the graph structure data, the weight of each edge in the graph structure data is encoded as follows: ； ； ； ； in: Represents a node With nodes The basic encoding of the edge weights connecting the edges; This represents the basic encoded learnable vector; Represents a node Attributes; Represents a node Attributes; Represents a node With nodes The difference in output power between them; Represents a node With nodes The residual suppression weights between; This represents the residual suppression strength coefficient; Represents a node The energy balance residual; Indicates the inflow node Energy; Indicates outflow node Energy; Represents a node The converter loss; Represents a node With nodes Encoding of the weights of the connecting edges; Encoding node attributes based on edge weight encoding involves the following steps: The residual signature vector of each node is constructed by concatenating the energy balance residual of each node, the mean of the historical energy balance residual over a preset time span, and the variance of the historical energy balance residual over a preset time span. A dual-branch weight calculation unit is set up to calculate the basic encoding weight and resource encoding weight of node attributes respectively, as shown in the following formula: ； ； in: Represents a node The basic coding weights; Represents a node Resource coding weights; , This represents a learnable transformation mapping function; This represents the encoded value of the node resource configuration data at that moment; The basic encoding of each node's attributes is calculated based on the residual signature vector of each node, as shown in the following formula: ； in: Represents a node The basic encoded value of the attribute; express function; The basic encoding of node attributes is represented by a learnable weight matrix; The basic encoding of node attributes is a learnable bias vector; Represents a node The residual signature vector; The node attribute code is calculated based on the basic attribute code, basic code weight, and resource code weight of each node, as shown in the following formula: ； in: Represents a node Attribute encoding.

5. The virtual power plant aggregated load trend prediction method according to claim 4, characterized in that, The energy storage coding layer is used to encode the operational data of energy storage nodes. The specific steps are as follows: The upper limit of the power projection is applied to the difference between the charging and discharging power of the energy storage node, as shown in the following formula: ； in: express Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates energy storage node The upper limit of charging and discharging power; express Real-time energy storage nodes The difference in charging and discharging power; A ramp correction is applied to the upper limit projection of the power difference between the charging and discharging power of the energy storage node, as shown in the following formula: ； in: Indicates the result after hill climbing correction Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates the result after hill climbing correction Real-time energy storage nodes The upper limit projection of the power difference between charging and discharging power; Indicates energy storage node The charging efficiency; Indicates energy storage node The discharge efficiency; express Time to The time span of a moment; The energy storage node operation data and the upper limit of the power difference between charging and discharging power after ramp correction are spliced ​​together and then input into the encoder to obtain the energy storage node operation data encoding.

6. The virtual power plant aggregated load trend prediction method according to claim 5, characterized in that, The output layer is built on a multilayer perceptron. The outputs of the node aggregation layer, graph coding layer and energy storage coding layer at different historical moments are spliced ​​together and then input into the multilayer perceptron. The multilayer perceptron outputs virtual power plant load forecast data for preset future time steps. The virtual power plant load forecast data for all future time steps are combined to obtain the virtual power plant predicted load trend.

7. A virtual power plant aggregated load trend prediction system, characterized in that, It includes a data acquisition module, a feature construction module, a load forecasting model construction module, a model training module, and a load forecasting module; The data acquisition module is used to collect historical source-grid-load-storage data of the virtual power plant, and to collect real-time source-grid-load-storage data of the virtual power plant. The feature construction module is used to construct source-grid-load-storage features based on historical source-grid-load-storage data. The source-grid-load-storage features include time-series features and structural features. The load forecasting model construction module is used to construct a virtual power plant aggregated load forecasting model, which includes a node aggregation layer, a graph coding layer, an energy storage coding layer, and an output layer. The node aggregation layer is used to extract the node aggregation features of each source-grid-load-storage node included in the virtual power plant at each time step in the time series features. The graph coding layer is used to encode the structural features. The energy storage coding layer is used to encode the energy storage node operation data. The output layer is used to generate the virtual power plant future load data prediction results based on the node aggregation features, the coding results of the graph structural data, and the coding results of the energy storage node operation data. The model training module is used to train the virtual power plant aggregated load prediction model using source-grid-load-storage characteristics; The load forecasting module is used to predict the future load data of the virtual power plant based on the real-time source-grid-load-storage data of the virtual power plant using the trained virtual power plant aggregated load forecasting model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

Images