A source-network-load-storage collaborative virtual power plant multi-time scale power scheduling method and system

Abstract

The present application relates to a kind of source network load storage collaborative virtual power plant multi-time scale power scheduling method and system, belong to data acquisition control system field, more belong to the circuit device field of alternating current trunk line or alternating current distribution network.The method includes: for the source network load storage collaborative of the set power consumption area of target virtual power plant jurisdiction selects the time window of power scheduling, i.e.time scale;Form the long short-term memory network model corresponding to the time scale and target virtual power plant;Collect multi-dimensional source network load storage data under the time scale;Using the long short-term memory network model, the power supply power value, the energy storage power value and the load power value of current time interval are intelligently predicted;Based on the intelligent prediction result, source network load storage collaborative power scheduling is executed.The present application can adapt to the time scale demand of different scheduling levels, so that network structure follows the configuration scale change of virtual power plant, improves the accuracy and efficiency of multi-time scale power prediction and scheduling.

Description

Technical Field

[0001] This invention belongs to the field of data acquisition and control systems, and more specifically to the field of circuit devices for AC trunk lines or AC distribution networks. It particularly relates to a method and system for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination. Background Technology

[0002] Against the backdrop of the ongoing construction of new power systems and the continuous advancement of "dual-carbon" goals, virtual power plants, as an important carrier for integrating distributed photovoltaic (PV), energy storage devices, and controllable loads to participate in grid optimization and dispatch, directly impact the economic efficiency and reliability of grid operation through their power prediction and coordinated dispatch capabilities. By integrating four types of resources—source (distributed PV), grid (distribution network), load (electricity load), and storage (energy storage devices)—virtual power plants achieve multi-energy complementarity and flexible regulation, becoming a key technological means to alleviate the pressure of new energy consumption and improve the operating efficiency of distribution networks.

[0003] For example, Chinese invention patent publication CN116207793A proposes a cloud-edge collaborative virtual power plant scheduling method, device, and computer equipment. The method includes: aggregating operating data of distributed power equipment through a virtual power plant control unit to obtain virtual power plant control unit operating data; obtaining power adjustment limit parameters corresponding to the virtual power plant control unit from the virtual power plant edge cluster; determining power adjustment information for the virtual power plant control unit based on the virtual power plant control unit operating data and power adjustment limit parameters; and forwarding the power adjustment information to the virtual power plant control unit from the virtual power plant edge cluster, enabling the virtual power plant control unit to adjust the operating power of its corresponding distributed power equipment based on the power adjustment information. Compared to traditional methods that cannot perform power scheduling on virtual power plants with massive amounts of data, this solution utilizes a system composed of a cloud server, a virtual power plant edge cluster, and a virtual power plant control unit to schedule the power of the virtual power plant, meeting the increasingly demanding scheduling requirements of virtual power plants based on massive amounts of data.

[0004] For example, Chinese invention patent publication CN118982160A proposes a method and system for optimizing the scheduling of virtual power plant source-load-storage power. The method includes: constructing a virtual power plant system model; establishing an optimization scheduling model; solving the model using an improved cuckoo algorithm; designing and applying a dynamically changing step-size CS algorithm; and running the CS algorithm to obtain an economically optimized scheduling scheme for the virtual power plant by finding the best-located "nest". This invention solves the technical problems of limitations in optimization algorithms, lack of adaptability to the volatility of renewable energy, insufficient demand-side management, the singularity of energy storage system scheduling strategies, and the lack of a scheduling model that comprehensively considers multiple objectives.

[0005] Therefore, it can be seen that the existing virtual power plant dispatching technology has the following main shortcomings:

[0006] First, using a single fixed time scale for power prediction and scheduling cannot adapt to the differences in time characteristics among the source, grid, load and storage units. In fact, photovoltaic output fluctuates at the second level, energy storage regulation has the ability to respond at the minute level, and load changes show an hourly trend. The time dimensions of the three are fixed and not aligned, making it difficult to coordinate multiple time scales.

[0007] Second, the fixed structure of the prediction model makes it unable to adapt to the configuration differences of virtual power plants of different sizes, nor can it adapt to changes in the selected time scale. This results in the use of the same prediction model for multiple different application scales and / or multiple time scales, leading to unsatisfactory accuracy and efficiency of the prediction results.

[0008] Third, the scattered collection of multi-dimensional data and the lack of a unified integration and synchronization mechanism lead to large deviations in power prediction results and insufficient basis for scheduling decisions.

[0009] Therefore, a technical solution is needed that can adaptively select the power dispatch time window for the designated power consumption area under the jurisdiction of the target virtual power plant, dynamically construct a long short-term memory network model that matches the time scale and virtual power plant configuration parameters, integrate multi-dimensional source-grid-load-storage historical data according to a unified time scale, and intelligently predict the source-grid-load-storage power status of the current time interval based on the dynamic model, thereby performing precise coordinated power dispatch. This solution aims to solve the technical problems of low prediction accuracy, dispatch lag, and resource mismatch caused by rigid prediction time scale, fixed model structure, and insufficient integration of multi-source data in existing technologies. Summary of the Invention

[0010] The purpose of this invention is to provide a method and system for multi-timescale power dispatching of a virtual power plant with coordinated generation, grid, load, and storage. This method adaptively determines the dispatching time window through scale selection devices, dynamically constructs a Long Short-Term Memory (LSTM) network model that matches the time scale and virtual power plant configuration parameters, uniformly collects and integrates multi-dimensional historical power data from generation, grid, load, and storage at the selected time scale, and intelligently predicts the power status of generation, grid, load, and storage in the current time interval based on the dynamically constructed network model. Coordinated power dispatching is then executed accordingly. This addresses the technical problems in existing technologies, such as rigid virtual power plant dispatching time scales, fixed prediction model structures, and insufficient integration of multi-source data leading to low prediction accuracy, dispatching lag, and resource mismatch.

[0011] According to a first aspect of the present invention, a method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination is provided, the method comprising:

[0012] The time window, or time scale, for power dispatching in the coordinated power supply, grid, load, and storage of the designated power consumption area under the jurisdiction of the target virtual power plant is selected. The target virtual power plant is used for AC power distribution.

[0013] A long short-term memory network model is constructed that corresponds to the selected time scale and the target virtual power plant. The time step of the long short-term memory network model follows the numerical change trend of the selected time scale. The number of hidden layer units of the long short-term memory network model is proportional to the total number of reference units corresponding to the target virtual power plant. The total number of reference units corresponding to the target virtual power plant is equal to the sum of the number of photovoltaic power generation units of the distributed photovoltaic power supply power source that supplies power to the set power consumption area and the number of energy storage units of the energy storage device configured for the set power consumption area.

[0014] The system collects the power supply values ​​of distributed photovoltaic power sources supplying power to designated power consumption areas in each historical time interval, the energy storage power values ​​of energy storage devices configured for designated power consumption areas in each historical time interval, and the load power consumption values ​​of designated power consumption areas in each historical time interval. These data are used as multi-dimensional source-grid-load-storage data outputs at the selected time scale, where the duration of each time interval is equal to the selected time scale.

[0015] Using a long short-term memory network model corresponding to the selected time scale and the target virtual power plant, based on the selected time scale, the number of photovoltaic power generation units of the distributed photovoltaic power supply power source supplying power to the designated power consumption area, the number of energy storage units of the energy storage device configured for the designated power consumption area, the maximum power supply value of the distributed photovoltaic power supply power source supplying power to the designated power consumption area, the maximum energy storage value of the energy storage device configured for the designated power consumption area, the maximum power consumption value of each power load in the designated power consumption area, and multi-dimensional source-grid-load-storage data under the selected time scale, the intelligent prediction of the power supply value of the distributed photovoltaic power supply power source supplying power to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval. Here, the current time interval is the subsequent time interval of each historical time interval.

[0016] Based on intelligent prediction results, the source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant is carried out for the current time interval.

[0017] Furthermore, the time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale, and the number of hidden layer units of the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant.

[0018] The total number of reference units corresponding to the target virtual power plant is equal to the sum of the number of photovoltaic power generation units of the distributed photovoltaic power supply that provides power to the designated power consumption area and the number of energy storage units configured for the energy storage devices in the designated power consumption area.

[0019] Furthermore, the collection of multi-dimensional source-grid-load-storage data at the selected time scale includes: collecting the power supply values ​​of distributed photovoltaic power sources supplying power to the designated power consumption area in each historical time interval, the energy storage power values ​​of energy storage devices configured for the designated power consumption area in each historical time interval, and the load power values ​​of the designated power consumption area in each historical time interval, as multi-dimensional source-grid-load-storage data at the selected time scale, with each time interval occupying a duration equal to the selected time scale;

[0020] Among them, the target virtual power plant completes the source-grid-load-storage coordination of AC power distribution in the designated power consumption area through distributed photovoltaic power supply to the designated power consumption area, AC distribution network from the distributed photovoltaic power supply to the designated power consumption area, energy storage device configured for the designated power consumption area, and load switching device for the designated power consumption area.

[0021] Among them, the source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant based on intelligent prediction results for the current time interval includes: when the maximum power supply value of the distributed photovoltaic power supply power supply to the designated power consumption area is less than the load power consumption value of the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area is equal to the difference between the two, so as to complete the pure green power supply;

[0022] Specifically, when the maximum power supply of the distributed photovoltaic power source supplying power to the designated power consumption area is less than the load power consumption of the designated power consumption area in the current time interval, the energy storage power of the energy storage device configured for the designated power consumption area in the current time interval is equal to the difference between the two, so as to complete the pure green power supply, includes: when the energy storage power of the energy storage device configured for the designated power consumption area in the current time interval is less than the difference between the two, the energy storage device configured for the designated power consumption area and other power sources work together to complete the power supply of the difference between the two.

[0023] Furthermore, the system collects power supply values ​​for distributed photovoltaic power sources supplying power to designated electricity consumption areas in each historical time interval, energy storage power values ​​for energy storage devices configured for designated electricity consumption areas in each historical time interval, and load power values ​​for designated electricity consumption areas in each historical time interval. These are used as multi-dimensional source-grid-load-storage data outputs at the selected time scale. The duration of each time interval is equal to the selected time scale. All collected power values ​​are represented using binary values ​​with a preset number of bits.

[0024] Among them, the energy storage devices configured for the designated electricity consumption areas include various energy storage units distributed at various nodes of the power distribution network, and the maximum energy storage power of the energy storage devices configured for the designated electricity consumption areas is between 1MWh and 20MWh.

[0025] The time step size of the Long Short-Term Memory (LSTM) network model follows the numerical change trend of the selected time scale as follows: if the selected time scale is 24 hours, the time step size of the LSM network model is 96 steps; if the selected time scale is 12 hours, the time step size of the LSM network model is 48 steps; and if the selected time scale is 1 hour, the time step size of the LSM network model is 24 steps.

[0026] Furthermore, before performing source-grid-load-storage coordination of AC power distribution in the designated electricity-consuming area under the jurisdiction of the target virtual power plant for the current time interval based on intelligent prediction results, the method further includes:

[0027] Receive the power supply value of the distributed photovoltaic power supply power supply to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval, and store the power supply value of the distributed photovoltaic power supply power supply power supply to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval to the big data storage node corresponding to the target virtual power plant;

[0028] This involves allocating different big data storage nodes to different virtual power plants, and transmitting the power supply values ​​of distributed photovoltaic power sources supplying power to designated power consumption areas in the current time interval, the energy storage power values ​​of energy storage devices configured for designated power consumption areas in the current time interval, and the load power consumption values ​​of designated power consumption areas in the current time interval, along with the power plant code values ​​of the target virtual power plant, over the network. This enables the targeted storage of the power supply values ​​of distributed photovoltaic power sources supplying power to designated power consumption areas in the current time interval, the energy storage power values ​​of energy storage devices configured for designated power consumption areas in the current time interval, and the load power consumption values ​​of designated power consumption areas in the current time interval to their matching big data storage nodes.

[0029] Furthermore, before performing source-grid-load-storage coordination of AC power distribution in the designated electricity-consuming area under the jurisdiction of the target virtual power plant for the current time interval based on intelligent prediction results, the method further includes:

[0030] Receives the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval, and performs local synchronous display on the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval;

[0031] The local synchronous display of the power supply value of the distributed photovoltaic power supply power source supplying power to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to supply power to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval includes: using an LCD screen to complete the local synchronous display.

[0032] Furthermore, before the target virtual power plant is used for AC power distribution, the method further includes: (1) Selecting a time window, or time scale, for power dispatching in a coordinated manner for the source-grid-load-storage region under the jurisdiction of the target virtual power plant, the method also includes:

[0033] Set up a user access interface for the target virtual power plant. The user access interface is used to provide a user selection interface for the selection of the power dispatch time window.

[0034] Specifically, a user access interface is set up for the target virtual power plant. The user access interface is used to provide a user selection interface for the selection of the power scheduling time window. This includes: the operator of the target virtual power plant selects the power scheduling time window through the API function of the user access interface.

[0035] Furthermore, after constructing a long short-term memory network model corresponding to the selected time scale and the target virtual power plant, the method further includes:

[0036] By storing the various model data of the long short-term memory network model that corresponds to the selected time scale and the target virtual power plant, the model storage of the long short-term memory network model that corresponds to the selected time scale and the target virtual power plant is completed.

[0037] Specifically, the model storage of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant is completed by storing the various model data of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant. This includes: synchronously storing the selected time scale, the power plant code value of the target virtual power plant, and the various model data of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant into the same physical storage unit.

[0038] Furthermore, the time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale by using a scale change curve to represent the numerical change trend of the selected time scale and a step size change curve to represent the numerical change trend of the time step of the Long Short-Term Memory Network Model.

[0039] The time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale, which further includes the fact that the step size change curve and the scale change curve completely overlap after being normalized to the same length.

[0040] The fact that the number of hidden layer units in the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant includes: using a quantitative analytical function to represent the numerical mapping relationship between the number of hidden layer units in the Long Short-Term Memory Network Model and the total number of reference units corresponding to the target virtual power plant;

[0041] The fact that the number of hidden layer units in the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant also includes: in the quantity analysis function, the number of hidden layer units in the Long Short-Term Memory Network Model is the output value, and the total number of reference units corresponding to the target virtual power plant is the input value.

[0042] According to a second aspect of the present invention, a virtual power plant multi-timescale power dispatching system for source-grid-load-storage coordination is provided, the system comprising:

[0043] The scale selection device is used to select the time window, i.e., the time scale, for the source-grid-load-storage coordination of the power dispatching of the designated power consumption area under the jurisdiction of the target virtual power plant. The target virtual power plant is used for AC power distribution.

[0044] The model building device is used to build a long short-term memory network model that corresponds to the selected time scale and the target virtual power plant. The time step of the long short-term memory network model follows the numerical change trend of the selected time scale, and the number of hidden layer units of the long short-term memory network model is proportional to the total number of reference units corresponding to the target virtual power plant.

[0045] Multidimensional acquisition devices are used to collect the power supply values ​​of distributed photovoltaic power sources supplying power to designated power consumption areas in each historical time interval, the energy storage power values ​​of energy storage devices configured to designated power consumption areas in each historical time interval, and the load power values ​​of designated power consumption areas in each historical time interval, as multidimensional source-grid-load-storage data output at the selected time scale.

[0046] The state prediction device, connected to the scale selection device, model building device, and multi-dimensional acquisition device, is used to intelligently predict the power supply of the distributed photovoltaic power supply to the designated power consumption area, the energy storage power of the energy storage device configured for the designated power consumption area, and the load power consumption of the designated power consumption area in the current time interval, based on the selected time scale and the long short-term memory network model corresponding to the target virtual power plant. This prediction is made using the selected time scale, the number of photovoltaic power generation units of the distributed photovoltaic power supply to the designated power consumption area, the number of energy storage units of the energy storage device configured for the designated power consumption area, the maximum power supply of the distributed photovoltaic power supply to the designated power consumption area, the maximum energy storage value of the energy storage device configured for the designated power consumption area, the maximum power consumption value of each load in the designated power consumption area, and the multi-dimensional source-grid-load-storage data under the selected time scale.

[0047] Power dispatching devices, connected to state prediction devices, are used to perform source-grid-load-storage coordination of AC power distribution in designated power consumption areas under the jurisdiction of a target virtual power plant for the current time interval based on intelligent prediction results.

[0048] Compared with the prior art, the present invention achieves the corresponding technical effects through the following main inventive points:

[0049] By adaptively selecting the time window for power scheduling, the time step of the LSTM model is dynamically adjusted according to the trend of numerical changes in the time scale. The day-ahead scheduling layer uses a long step to reduce the amount of computation, while the real-time scheduling layer uses a short step to improve the response speed, thereby achieving accurate adaptation and efficient coordination of source, grid, load and storage at multiple time scales.

[0050] By using a dual-parameter dynamic model building mechanism, the number of hidden layer units in the LSTM is proportional to the sum of the number of photovoltaic power generation units and the number of energy storage units. The network capacity expands linearly with the actual configuration parameters of the virtual power plant. This avoids overfitting in small-scale scenarios and ensures representation capability in large-scale scenarios, thereby improving the model's adaptability and prediction stability for virtual power plants of different sizes.

[0051] By using a multi-source data acquisition mode, the power supply, energy storage and load power consumption values ​​of each historical time interval are uniformly extracted according to the selected time scale to form a time-aligned multi-dimensional source-grid-load-storage data output. Based on this standardized data and the maximum capacity parameters of photovoltaic, energy storage and load, joint prediction is carried out to improve the spatiotemporal consistency and prediction accuracy of multi-dimensional data, and provide a reliable basis for collaborative scheduling of power dispatch.

[0052] By performing coordinated scheduling of power generation, grid, load and storage based on intelligent prediction results, when photovoltaic output is insufficient to cover load demand, energy storage devices are prioritized to supplement the difference to complete the supply of pure green electricity. When energy storage capacity is insufficient, energy storage devices and other power sources work together to supplement the power difference, thus realizing the complementarity and optimal configuration of power generation, grid, load and storage. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the working scenario of a virtual power plant multi-timescale power dispatching method and system for source-grid-load-storage coordination according to the present invention.

[0054] Figure 2 This is a flowchart illustrating the steps of a multi-timescale power dispatching method for a virtual power plant with source-grid-load-storage coordination according to a first embodiment of the present invention.

[0055] Figure 3 This is a schematic diagram of the internal structure of the core model used in the virtual power plant multi-timescale power dispatching method and system of the present invention, which is based on the coordination of source, grid, load and storage.

[0056] Figure 4 This is a flowchart illustrating the steps of a virtual power plant multi-timescale power dispatching method with source-grid-load-storage coordination according to a second embodiment of the present invention.

[0057] Figure 5 This is a flowchart illustrating the steps of a multi-timescale power dispatching method for a virtual power plant with source-grid-load-storage coordination according to a third embodiment of the present invention.

[0058] Figure 6 This is a flowchart illustrating the steps of a multi-timescale power dispatching method for a virtual power plant with source-grid-load-storage coordination according to a fourth embodiment of the present invention.

[0059] Figure 7 This is a flowchart illustrating the steps of a multi-timescale power dispatching method for a virtual power plant with source-grid-load-storage coordination according to a fifth embodiment of the present invention.

[0060] Figure 8 This is a schematic diagram of a multi-timescale power dispatching system for a virtual power plant with source-grid-load-storage coordination according to the sixth embodiment of the present invention. Detailed Implementation

[0061] The specific embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0062] Figure 1This is a schematic diagram illustrating the working scenario of a virtual power plant multi-timescale power dispatching method and system for source-grid-load-storage coordination according to the present invention. The data acquisition and control system proposed in this invention belongs to the field of circuit devices for AC trunk lines or AC distribution networks.

[0063] exist Figure 1 In the virtual power plant, the target power plant manages the distributed photovoltaic power supply for the designated power consumption area, the AC distribution network from the distributed photovoltaic power supply to the designated power consumption area, the energy storage device configured for the designated power consumption area, and the load switching device for the designated power consumption area.

[0064] exist Figure 1 In the process, the power load is the power load of the designated power consumption area and there is more than one. Based on the intelligent prediction results of the long short-term memory network model (LSTM model) corresponding to the target virtual power plant, the corresponding source-grid-load-storage coordination strategy is determined. Based on the determined source-grid-load-storage coordination strategy, the source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant is executed for the current time interval.

[0065] Specifically, the long short-term memory network model corresponding to the target virtual power plant is actually a long short-term memory network model that corresponds to both the selected time scale and the target virtual power plant. Through the dynamic structural design of the core modules, it can simultaneously match multiple time scales and different virtual power plant sizes.

[0066] This invention addresses a source-grid-load-storage coordinated scheduling scenario for a target virtual power plant. The target virtual power plant is used for AC power distribution and achieves source-grid-load-storage coordination for a designated power consumption area by aggregating distributed photovoltaic power sources, distribution networks, energy storage devices, and load switching devices that supply power to that area. The designated power consumption area can be an industrial park, commercial building, residential community, or public institution's power consumption area.

[0067] The main inventive points of this invention include: adaptive selection of time windows for power dispatch, dynamic construction of different LSTM models corresponding to virtual power plants of different time scales and sizes, targeted screening of multi-source basic data for intelligent prediction, and a source-grid-load-storage coordinated dispatch strategy based on predicted power values.

[0068] Specifically, the main differences between this invention and existing power grid dispatching technologies are as follows:

[0069] First, the size of the prediction model follows the scale of the power plant: Existing technologies use a fixed structure for prediction models regardless of how many devices are installed in the virtual power plant. This is wasteful for small plants and insufficient for large plants. This invention links the number of hidden layer units in the model to the sum of the number of photovoltaic units and energy storage units. For example, if a power plant has 200 photovoltaic units and 100 energy storage units, totaling 300 devices, the model will have 3 hidden layer units; if another power plant has 400 units and 200 energy storage units, totaling 600 devices, the model will have 6 units. In this way, as the power plant expands, the model also expands, preventing overfitting or inability to perform calculations.

[0070] Second, the computational step size follows the scheduling hierarchy: Existing technologies use the same step size for both day-ahead and real-time forecasts. Day-ahead forecasts suffer from slow computation due to the large amount of data, while real-time forecasts lack fine enough step size and are slow to respond. This invention links the time step size with the selected time scale: for 24-hour day-ahead forecasts, the step size is set to 96 steps, one step every 15 minutes, resulting in a moderate computational load; for 1-hour real-time forecasts, the step size is set to 24 steps, one step every 2.5 minutes, resulting in a faster response. Different levels can utilize different methods, avoiding a one-size-fits-all approach.

[0071] Third, multiple data sets are aligned within the same time window: In existing technologies, data from photovoltaics, energy storage, and load are often collected according to their own cycles, resulting in mismatched timestamps and potential errors in forecasting. This invention requires all data to be uniformly extracted according to a selected time scale. For example, if a 1-hour time scale is selected, then a value is taken for each of the past hour's power supply, energy storage power, and load power to form a set of matching data. In this way, the three sets of data are strictly aligned on the time axis, preventing misalignment during forecasting.

[0072] Fourth, the system allows for direct scheduling after forecasting, eliminating the need for an intermediate step: Existing technologies typically forecast power first, then pass the forecast results to the scheduling module for independent decision-making, which can easily lead to a disconnect between the two stages. In this invention, the forecast results are immediately used to guide scheduling operations. For example, if the forecast predicts that the photovoltaic system can only generate 800 kilowatts in the next period, but the load requires 1000 kilowatts, the system immediately allows energy storage to fill the 200 kilowatt gap. If the energy storage system can only generate 150 kilowatts at that time, the system automatically coordinates other power sources to make up the remaining 50 kilowatts. In this way, forecasting and scheduling are integrated, resulting in faster response and fewer errors.

[0073] The following will describe in detail, by way of embodiments, a method and system for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to the present invention.

[0074] <First Embodiment>

[0075] Figure 2The present invention provides a flowchart illustrating a multi-timescale power dispatching method for a virtual power plant with source-grid-load-storage coordination according to a first embodiment of the present invention. The method is implemented at the energy management end of the virtual power plant and includes steps S201 to S205, the specific steps of which are shown below:

[0076] Step S201: Select a time window, i.e. a time scale, for power dispatching in the power generation, grid, load and storage coordination of the designated power consumption area under the jurisdiction of the target virtual power plant. The target virtual power plant is used for AC power distribution.

[0077] Specifically, the time scale is used to characterize the length of the time window covered by power scheduling and state prediction. Different time scales correspond to different scheduling level requirements. For example, the day-ahead scheduling layer selects 24 hours as the time scale, the intraday scheduling layer selects 12 hours as the time scale, and the real-time scheduling layer selects 1 hour as the time scale.

[0078] In this embodiment, the time scale is selected by the scale selection device according to the scheduling task attributes or operator instructions; the target virtual power plant completes the source-grid-load-storage coordination of the AC power distribution in the designated power consumption area through the distributed photovoltaic power supply to the designated power consumption area, the AC distribution network from the distributed photovoltaic power supply to the designated power consumption area, the energy storage device configured for the designated power consumption area, and the load switching device for the designated power consumption area.

[0079] Step S202: Construct a long short-term memory network model that corresponds to the selected time scale and the target virtual power plant;

[0080] Specifically, the model building device dynamically builds the corresponding Long Short-Term Memory (LSTM) network model based on the selected time scale and the configuration parameters of the target virtual power plant. The time step of the LSTM model follows the numerical change trend of the selected time scale, and the number of hidden layer units of the LSTM model is correlated in the same direction with the total number of reference units corresponding to the target virtual power plant.

[0081] For example, the total number of reference units corresponding to the target virtual power plant is 300, and the number of hidden layer units of the LSTM model is 3; the total number of reference units corresponding to the target virtual power plant is 400, and the number of hidden layer units of the LSTM model is 4; the total number of reference units corresponding to the target virtual power plant is 500, and the number of hidden layer units of the LSTM model is 5, and so on.

[0082] The total number of reference units corresponding to the target virtual power plant is equal to the sum of the number of photovoltaic power generation units of the distributed photovoltaic power supply that supplies power to the designated power consumption area and the number of energy storage units configured for the energy storage devices in the designated power consumption area.

[0083] Significantly, this embodies a key inventive point of the present invention: through a dual-parameter dynamic model building mechanism, the number of LSTM hidden layer units is proportional to the sum of the number of photovoltaic power generation units and the number of energy storage units, and the network capacity expands linearly with the actual configuration parameters of the virtual power plant. This avoids overfitting in small-scale scenarios and ensures representation capability in large-scale scenarios, thereby improving the model's adaptability and prediction stability to virtual power plants of different sizes.

[0084] Specifically, the dual-parameter dynamic model building mechanism refers to the fact that when building a long short-term memory network model, the model building device simultaneously uses the time scale and the target virtual power plant configuration scale as two independent input parameters. Through a preset joint mapping rule, it synchronously determines two core structural parameters of the model at one time: the time step and the number of hidden layer units. The first parameter is the selected time scale, which directly determines the number of time span steps by which the model processes historical data. The second parameter is the total number of reference units corresponding to the target virtual power plant, which is the sum of the number of photovoltaic power generation units and the number of energy storage units, and directly determines the number of units in the model's hidden layer. The two parameters are input and parsed in parallel during the model building process, without interfering with each other, but together forming a complete network architecture.

[0085] Specific example: Assume the target virtual power plant is configured with 250 photovoltaic power generation units and 150 energy storage units, the total number of reference units is 400, and the time scale of the scheduling task is selected as 12 hours; after receiving the above two parameters, the model building device synchronously outputs two structural parameters according to the joint mapping rule: the time step is 48 steps and the number of hidden layer units is 4, thus building a long short-term memory network model that matches the time scale and the size of the virtual power plant in one go;

[0086] For example, a virtual power plant is equipped with 150 photovoltaic power generation units and 150 energy storage units. The total number of reference units is 300. The selected time scale is 24 hours. Then the time step for synchronous determination of the model components is 96 steps and the number of hidden layer units is 3.

[0087] Because the two parameters are mapped independently yet take effect simultaneously, when either parameter changes, the model components output a new structure that adapts to it. This allows the network capacity to both expand linearly with the scale of the virtual power plant and dynamically adjust over time, avoiding prediction mismatch in multiple scenarios caused by a single fixed model.

[0088] Furthermore, the time step of the LSTM model follows the numerical change trend of the selected time scale, i.e., the first parameter, so that the day-ahead scheduling layer uses a longer time step to reduce the amount of computation and improve the economic optimization space, while the real-time scheduling layer uses a shorter time step to improve the response speed and enhance agility.

[0089] Significantly, this demonstrates another key inventive point of the present invention: by adaptively selecting the time window through the scale selection mode, the time step of the LSTM model is dynamically adjusted according to the trend of the time scale numerical change. The day-ahead scheduling layer uses a long step to reduce the amount of computation, and the real-time scheduling layer uses a short step to improve the response speed, thereby achieving accurate adaptation and efficient coordination of source, grid, load and storage under multiple time scales.

[0090] Specifically, dynamically adjusting the time step of the Long Short-Term Memory (LSTM) network model according to the numerical trend of the time scale means that the model building device takes the selected time scale as the input variable and outputs the corresponding time step value according to the preset step size mapping rule. Here, the scale change curve represents the numerical trend of the time scale, and the step size change curve represents the numerical trend of the time step. The two curves completely overlap after being normalized to the same length, that is, the time step and the time scale maintain a fixed proportional relationship.

[0091] For example, when the selected time scale is 24 hours, the time step is configured as 96 steps, meaning one time step corresponds to every 15 minutes; when the selected time scale is 12 hours, the time step is configured as 48 steps, again maintaining one time step per 15 minutes; when the selected time scale is 1 hour, the time step is configured as 24 steps, meaning one time step corresponds to every 2.5 minutes. The day-ahead scheduling layer uses a 24-hour time scale with a 96-step size, resulting in a smaller computational load per forward propagation and reducing the overall computational burden; the real-time scheduling layer uses a 1-hour time scale with a 24-step size, shortening the single-step duration and improving response speed.

[0092] Thus, different scheduling levels automatically obtain matching time steps based on their time scales, achieving precise adaptation of source, grid, load and storage at multiple time scales.

[0093] Step S203: Collect multi-dimensional source-grid-load-storage data at the selected time scale;

[0094] Specifically, the multi-dimensional acquisition device collects the power supply values ​​of distributed photovoltaic power sources supplying power to the designated power consumption area in each historical time interval, the energy storage power values ​​of energy storage devices configured to supply power to the designated power consumption area in each historical time interval, and the load power consumption values ​​of the designated power consumption area in each historical time interval, as multi-dimensional source-grid-load-storage data output under the selected time scale; wherein, the duration of each time interval is equal to the selected time scale.

[0095] In this embodiment, each power value collected is represented by a binary value with a preset number of bits to facilitate efficient storage and transmission at the edge computing node.

[0096] The energy storage devices configured for designated power consumption areas include various energy storage units distributed at various nodes of the power distribution network, and the maximum energy storage power of the energy storage devices configured for designated power consumption areas is between 1MWh and 20MWh.

[0097] Significantly, this demonstrates another key inventive point of the present invention: by using a multi-source data acquisition mode, the power supply, energy storage, and load power consumption values ​​of each historical time interval are uniformly extracted according to a selected time scale, forming a time-aligned multi-dimensional source-grid-load-storage data output, thereby improving the spatiotemporal consistency and prediction accuracy of multi-dimensional data;

[0098] Step S204: Using the long short-term memory network model corresponding to the selected time scale and the target virtual power plant, based on the selected time scale, the number of photovoltaic power generation units of the distributed photovoltaic power supply power source supplying power to the set power consumption area, the number of energy storage units of the energy storage device configured for the set power consumption area, the maximum power supply value of the distributed photovoltaic power supply power source supplying power to the set power consumption area, the maximum energy storage value of the energy storage device configured for the set power consumption area, the maximum power consumption value of each power load in the set power consumption area, and the multi-dimensional source-grid-load-storage data under the selected time scale, the power supply value of the distributed photovoltaic power supply power source supplying power to the set power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the set power consumption area in the current time interval, and the load power consumption value of the set power consumption area in the current time interval;

[0099] The current time interval is the subsequent time interval that immediately follows each historical time interval;

[0100] like Figure 3 As shown, the core model used in the source-grid-load-storage collaborative virtual power plant multi-timescale power dispatching method and system of the present invention is a long short-term memory network model corresponding to the selected time scale and the target virtual power plant. It includes an input layer, a long short-term memory hidden layer (LSTM hidden layer), and an output layer connected in sequence. The model is used to receive multi-dimensional source-grid-load-storage time-series data and simultaneously perform three predictions: photovoltaic power supply prediction, energy storage power prediction, and load power prediction. That is, the output of the model is the power supply value of the distributed photovoltaic power supply to the set power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the set power consumption area in the current time interval, and the load power value of the set power consumption area in the current time interval.

[0101] Specifically, the input layer is used to receive multi-dimensional source-grid-load-storage data at a selected time scale. The input dimension of the input layer is equal to the number of dimensions of the multi-dimensional source-grid-load-storage data. The number of dimensions of the multi-dimensional source-grid-load-storage data is equal to the sum of the number of photovoltaic power generation units of distributed photovoltaic power sources supplying power to the designated power consumption area, the number of energy storage units of energy storage devices configured for the designated power consumption area, and the number of power loads in the designated power consumption area. The input layer divides the received multi-dimensional source-grid-load-storage data into continuous time series data according to the time step and then sends it to the long short-term memory hidden layer.

[0102] Specifically, in the long short-term memory hidden layer, the number of hidden layer units contained in each long short-term memory unit layer is proportional to the total number of reference units corresponding to the target virtual power plant; each long short-term memory unit in each long short-term memory unit layer contains an input gate, a forget gate, an output gate, and a cell state;

[0103] The input gate controls the proportion of multi-dimensional source-network-load storage data input to the cell state at the current time step, and its gating signal is calculated by the current input and the previous hidden state using a Sigmoid activation function. The forget gate controls the proportion of cell state information retained from the previous time step to the current time step, and its gating signal is also calculated using a Sigmoid activation function. The output gate controls the proportion of cell state information output to the next layer at the current time step. The cell state is used to record historical information across time steps, and selective updating and long-term retention of information are achieved through the synergistic effect of the input gate and the forget gate.

[0104] Specifically, the output layer is used to output intelligent prediction results. The output dimension of the output layer is three, corresponding to the power supply value of the distributed photovoltaic power supply to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval. The output layer uses a fully connected method to map the hidden state vector of the final time step of the long short-term memory hidden layer to the three predicted power values. The hidden state vector of each time step is obtained by filtering the cell state of that time step through the output gate.

[0105] Specifically, the learning rate of the Long Short-Term Memory network model is between 0.001 and 0.01, used to control the magnitude of network parameter updates; the number of iterations is between 100 and 500, used to control the number of training rounds of the network.

[0106] The Long Short-Term Memory (LSTM) network model uses mean squared error as the loss function, where the mean squared error is equal to the average of the sum of squares of the differences between the predicted power value and the actual collected power value in each historical time interval. The LTM network model uses backpropagation algorithm to update network parameters in order to minimize the mean squared error.

[0107] During training, the Long Short-Term Memory Network model uses multi-dimensional source-network-load data corresponding to each historical time interval as training sample input, and the actual power values ​​corresponding to the time interval immediately following each historical time interval as training labels. The model calculates the predicted power values ​​through forward propagation, and adjusts the weight matrices and bias vectors of the input gate, forget gate, and output gate through backpropagation until the mean square error is less than a preset threshold or the number of iterations is reached.

[0108] In a directly reproducible example, a 12-hour timescale can be selected. The target virtual power plant includes 250 photovoltaic power generation units, 150 energy storage units, and 50 electrical loads. The total number of reference units corresponding to the target virtual power plant is 400. The maximum power output of the photovoltaic power generation units is a total of 10MW. The maximum energy storage capacity of the energy storage devices is 10MWh, and the maximum charge / discharge power is 5MW. The maximum power consumption of each electrical load is a total of 12MW. Correspondingly, the input layer dimension of the LSTM model is 450, the time step is 48 steps, the number of hidden layer units is 4, and the output layer dimension is 3.

[0109] Specifically, the training data can be taken from the operation records of the same designated power consumption area from 00:00 on January 1, 2025 to 24:00 on December 31, 2025, with a sampling period of 15 minutes; each 12-hour historical time interval contains 48 consecutive sampling points; the 450-dimensional power data in a historical time interval are arranged in chronological order as a training sample, and the average value of photovoltaic power supply, energy storage power and load power consumption in the next adjacent 12-hour time interval is used as the training label; the energy storage power is positive in the direction of discharge and negative in the direction of charging;

[0110] After data processing, a total of 1459 training samples were obtained. The first 1021 samples were used as the training set, the next 219 as the validation set, and the last 219 as the test set, in chronological order, without random shuffling. Data with no more than three consecutive missing sampling points were imputed using linear interpolation; data with more than three consecutive missing sampling points were removed entirely from the corresponding historical time intervals. Abnormal power values ​​exceeding the upper limit or falling below the lower limit of the corresponding device were replaced with boundary values.

[0111] The power supply and load power are normalized to 0 to 1 according to the maximum values ​​in the training set, and the energy storage power is normalized to -1 to 1 according to the maximum absolute values ​​of charging and discharging in the training set. The maximum values ​​used for normalization are calculated only from the training set, and the same set of maximum values ​​is used for the validation set and the test set. After training, the model output is restored to the actual power value according to the same set of maximum values.

[0112] The LSTM model uses the Sigmoid activation function for the input gate, forget gate, and output gate, and the Tanh activation function for cell state updates. The input weight matrix is ​​initialized with a Glorot uniform distribution, the cyclic weight matrix is ​​initialized with orthogonal distribution, the bias vectors of the input gate, output gate, and cell state are initialized to 0, and the bias vector of the forget gate is initialized to 1. No Dropout layer is set.

[0113] The Adam optimizer was used during training, with an initial learning rate of 0.005, β1 of 0.9, β2 of 0.999, and ε of 0.00000001. Mean squared error was used as the loss function, with a batch size of 32 and a maximum of 300 iterations. Training samples were randomly shuffled within each iteration, while the validation and test sets maintained their chronological order. Gradients were clipped to a global norm of 1.0. Training stopped when the mean squared error of the validation set no longer decreased after 30 consecutive iterations, and the network parameters were restored to the state they were at their lowest. The preset mean squared error threshold was 0.0005.

[0114] When trained with the above parameters, the model stopped at the 236th iteration. The mean squared error of the validation set was the lowest at the 206th iteration, which was 0.00046. The mean squared error of the training set was 0.00038, and the mean squared error of the test set was 0.00049. The random seed was set to 42. The software development environment was Python 3.10.12 and TensorFlow 2.15.0. On the same computing device with an Intel Core i7-12700 processor and 32GB of memory, a single complete training session took 18 minutes and 36 seconds, and a single prediction session took 42 milliseconds.

[0115] Step S205: Based on the intelligent prediction results, perform source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant for the current time interval;

[0116] Specifically, the power dispatching device performs source-grid-load-storage coordinated dispatching for the current time interval based on the intelligent prediction results output by the state prediction device.

[0117] When the maximum power supply of the distributed photovoltaic power source supplying power to the designated power consumption area is less than the load power consumption of the designated power consumption area in the current time interval, the energy storage power of the energy storage device configured for the designated power consumption area is equal to the difference between the two, so as to complete the pure green power supply.

[0118] When the energy storage power value of the energy storage device configured for the designated power consumption area is less than the difference between the two in the current time interval, the energy storage device configured for the designated power consumption area and other power sources work together to supply the power to make up the difference between the two. The other power sources may include external power grid supply or other distributed energy storage devices.

[0119] Significantly, this demonstrates another key inventive point of the present invention: by using power dispatching devices to perform source-grid-load-storage coordinated dispatch based on intelligent prediction results, when photovoltaic output is insufficient to cover load demand, energy storage devices are prioritized to supplement the difference to complete the pure green power supply; when energy storage capacity is insufficient, other power sources are coordinated to complete the power supply, thereby achieving complementary and mutually supportive and optimized configuration of source-grid-load-storage.

[0120] To facilitate verification of the technical effects of this embodiment, a control model was set up under the same training set, validation set, test set, and the same data preprocessing method. The control model adopted a one-layer fixed-structure LSTM model with a fixed time step of 24 steps, a fixed number of hidden layer units of 3, and the same input layer dimension, output layer dimension, activation function, initialization method, optimizer, loss function, batch size, maximum number of iterations, and early stopping rule as this embodiment.

[0121] Specifically, the prediction error can be represented by the normalized average absolute error, which is equal to the average of the absolute errors between the predicted power and the actual power divided by the corresponding maximum power value, and then multiplied by 100%. Among them, the energy storage power uses the maximum charge and discharge power value of 5MW as the denominator; the average error of the three prediction results is the arithmetic mean of the normalized average absolute errors of photovoltaic, energy storage and load; the dispatch response time refers to the time from the data in the current time interval to the issuance of the dispatch command, and the arithmetic mean is taken after 100 consecutive tests.

[0122] On 219 test samples, the photovoltaic power supply error of the control model was 8.94%, the energy storage power error was 9.76%, the load power consumption error was 5.83%, the average error of the three items was 8.18%, and the average dispatch response time was 1.27 seconds; the photovoltaic power supply error of this embodiment was 6.41%, the energy storage power error was 7.35%, the load power consumption error was 4.18%, the average error of the three items was 5.98%, and the average dispatch response time was 0.86 seconds.

[0123] Based on the above data, the average error of the three items in this embodiment is reduced by 2.20 percentage points compared with the control model, a relative reduction of 26.9%, and the average scheduling response time is shortened by 0.41 seconds, a relative reduction of 32.3%. This comparison result shows that, under the same data and the same training conditions, making the time step follow the time scale and making the number of hidden layer units follow the total number of reference units can simultaneously reduce prediction error and scheduling time.

[0124] <Second Embodiment>

[0125] Figure 4 The following is a flowchart illustrating the steps of a virtual power plant multi-timescale power dispatching method based on source-grid-load-storage coordination according to a second embodiment of the present invention.

[0126] like Figure 4 As shown, with Figure 2 Unlike the previous embodiment, in the virtual power plant multi-timescale power dispatching method of source-grid-load-storage coordination, before performing source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant for the current time interval based on intelligent prediction results, that is, before step S205, the method further includes:

[0127] Step S206: Receive the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval, and store the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval to the big data storage node corresponding to the target virtual power plant;

[0128] Specifically, different big data storage nodes are allocated to different virtual power plants, and the power supply power of the distributed photovoltaic power supply supplying the designated power consumption area in the current time interval, the energy storage power of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption of the designated power consumption area in the current time interval are transmitted together with the power plant code value of the target virtual power plant to achieve targeted storage of the power supply power of the distributed photovoltaic power supply supplying the designated power consumption area in the current time interval, the energy storage power of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption of the designated power consumption area in the current time interval to their matching big data storage nodes.

[0129] Among them, the three power values ​​of the first time interval are stored in the first big data storage node, the three power values ​​of the second time interval are stored in the second big data storage node, and the three power values ​​of the third time interval are stored in the third big data storage node.

[0130] Therefore, by mapping the power plant coded values ​​to big data storage nodes, partitioned and isolated storage of prediction data from multiple virtual power plants can be achieved, avoiding data aliasing and improving data query efficiency and system scalability.

[0131] <Third Embodiment>

[0132] Figure 5 The flowchart illustrates the steps of a multi-timescale power dispatching method for a virtual power plant with source-grid-load-storage coordination according to a third embodiment of the present invention.

[0133] like Figure 5 As shown, with Figure 2 Unlike the previous embodiment, in the virtual power plant multi-timescale power dispatching method of source-grid-load-storage coordination, before performing source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant for the current time interval based on intelligent prediction results, that is, before step S205, the method further includes:

[0134] Step S207: Receive the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval, and perform local synchronous display on the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval;

[0135] Specifically, a liquid crystal display screen is used to complete the local synchronous display;

[0136] The local synchronous display using an LCD screen includes: the LCD screen can display power supply power, energy storage power, load power consumption and corresponding time information in columns according to power type, so that on-site dispatchers can directly view the prediction results and dispatch status.

[0137] <Fourth Embodiment>

[0138] Figure 6 The following is a flowchart illustrating the steps of a multi-timescale power dispatching method for a virtual power plant with source-grid-load-storage coordination according to a fourth embodiment of the present invention.

[0139] like Figure 6As shown, with Figure 2 Unlike the previous implementation, in the source-grid-load-storage coordinated virtual power plant multi-timescale power dispatch method, before the target virtual power plant is used for AC power distribution, i.e., before step S201, the method further includes:

[0140] Step S208: Set the user access interface corresponding to the target virtual power plant. The user access interface is used to provide a user selection interface for the selection of the power dispatch time window.

[0141] Specifically, the operators of the target virtual power plant select the time window for power scheduling through the API functions of the user access interface;

[0142] The selection of the time window for power scheduling by the operator of the target virtual power plant through the API function of the user access interface includes: the operator can access the user access interface through a web browser or mobile terminal.

[0143] <Fifth Embodiment>

[0144] Figure 7 The present invention provides a flowchart illustrating the steps of a multi-timescale power dispatching method for a virtual power plant based on source-grid-load-storage coordination according to a fifth embodiment of the present invention.

[0145] like Figure 7 As shown, with Figure 2 Unlike the previous implementation, in the source-grid-load-storage coordinated virtual power plant multi-timescale power dispatch method, after constructing the long short-term memory network model corresponding to the selected time scale and the target virtual power plant, i.e. after step S202, the method further includes:

[0146] Step S209: Complete the model storage of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant by storing the various model data of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant.

[0147] Specifically, the selected time scale, the power plant code value of the target virtual power plant, and the various model data of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant are synchronously stored in the same physical storage unit.

[0148] The model data includes the input layer weight matrix, the weight matrix and bias vector of each gate in the hidden layer, the output layer weight matrix and bias vector, network hyperparameters, and training state information.

[0149] Thus, by synchronously storing the time scale, power plant code, and model data, precise management of one scale and one model per plant is achieved.

[0150] Furthermore, in the multi-timescale power dispatching method for virtual power plants with source-grid-load-storage coordination shown in the above embodiments of the present invention:

[0151] The time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale, including: using a scale change curve to represent the numerical change trend of the selected time scale, and using a step size change curve to represent the numerical change trend of the time step of the Long Short-Term Memory Network Model.

[0152] The time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale, which further includes the fact that the step size change curve and the scale change curve completely overlap after being normalized to the same length.

[0153] Specifically, the complete overlap of the step size change curve and the scale change curve after equal-length normalization includes: after equal-length normalization, the step size change curve and the scale change curve are respectively obtained as the step size change curve and the scale change curve. The step size change curve and the scale change curve are actually the same curve and can completely overlap.

[0154] The fact that the number of hidden layer units in the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant includes: using a quantitative analytical function to represent the numerical mapping relationship between the number of hidden layer units in the Long Short-Term Memory Network Model and the total number of reference units corresponding to the target virtual power plant;

[0155] The fact that the number of hidden layer units in the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant also includes: in the quantity analysis function, the number of hidden layer units in the Long Short-Term Memory Network Model is the output value, and the total number of reference units corresponding to the target virtual power plant is the input value.

[0156] <Sixth Embodiment>

[0157] Figure 8 This is a schematic diagram of a multi-timescale power dispatching system for a virtual power plant with source-grid-load-storage coordination according to the sixth embodiment of the present invention. The system is implemented by a virtual power plant energy management platform.

[0158] Since the source-grid-load-storage collaborative virtual power plant multi-timescale power dispatching system shown in the sixth embodiment of the present invention corresponds to the source-grid-load-storage collaborative virtual power plant multi-timescale power dispatching method shown in the first embodiment of the present invention, many specific implementation details will not be repeated in the sixth embodiment.

[0159] like Figure 8 As shown, the virtual power plant multi-timescale power dispatch system for source-grid-load-storage coordination includes the following components:

[0160] The scale selection device is used to select the time window, i.e., the time scale, for the source-grid-load-storage coordination of the power dispatching of the designated power consumption area under the jurisdiction of the target virtual power plant. The target virtual power plant is used for AC power distribution.

[0161] The model building device is used to build a long short-term memory network model that corresponds to the selected time scale and the target virtual power plant. The time step of the long short-term memory network model follows the numerical change trend of the selected time scale, and the number of hidden layer units of the long short-term memory network model is proportional to the total number of reference units corresponding to the target virtual power plant.

[0162] Multidimensional acquisition devices are used to collect the power supply values ​​of distributed photovoltaic power sources supplying power to designated power consumption areas in each historical time interval, the energy storage power values ​​of energy storage devices configured to designated power consumption areas in each historical time interval, and the load power values ​​of designated power consumption areas in each historical time interval, as multidimensional source-grid-load-storage data output at the selected time scale.

[0163] The state prediction device, connected to the scale selection device, model building device, and multi-dimensional acquisition device, is used to intelligently predict the power supply of the distributed photovoltaic power supply to the designated power consumption area, the energy storage power of the energy storage device configured for the designated power consumption area, and the load power consumption of the designated power consumption area in the current time interval, based on the selected time scale and the long short-term memory network model corresponding to the target virtual power plant. This prediction is made using the selected time scale, the number of photovoltaic power generation units of the distributed photovoltaic power supply to the designated power consumption area, the number of energy storage units of the energy storage device configured for the designated power consumption area, the maximum power supply of the distributed photovoltaic power supply to the designated power consumption area, the maximum energy storage value of the energy storage device configured for the designated power consumption area, the maximum power consumption value of each load in the designated power consumption area, and the multi-dimensional source-grid-load-storage data under the selected time scale.

[0164] Power dispatching devices, connected to state prediction devices, are used to perform source-grid-load-storage coordination of AC power distribution in designated power consumption areas under the jurisdiction of a target virtual power plant for the current time interval based on intelligent prediction results.

[0165] In addition, the present invention may also cite the following technical contents to highlight the significant technical advancements of the present invention:

[0166] The total number of reference units corresponding to the target virtual power plant is equal to the sum of the number of photovoltaic power generation units of the distributed photovoltaic power supply power source that supplies power to the designated power consumption area and the number of energy storage units of the energy storage device configured for the designated power consumption area. This includes the numerical mapping relationship between the number of photovoltaic power generation units of the distributed photovoltaic power supply power source that supplies power to the designated power consumption area and the number of energy storage units of the energy storage device configured for the designated power consumption area, expressed by a two-input one-output information conversion formula, and the total number of reference units corresponding to the target virtual power plant.

[0167] Among them, the process of implementing the numerical mapping relationship of the two-input one-output information conversion formula can be tested and simulated using programmable logic devices;

[0168] The information conversion formula using two inputs and one output represents the numerical mapping relationship between the number of photovoltaic power generation units of the distributed photovoltaic power supply power source supplying power to the designated power consumption area and the number of energy storage units of the energy storage device configured for the designated power consumption area, and the total number of reference units corresponding to the target virtual power plant. The number of photovoltaic power generation units of the distributed photovoltaic power supply power source supplying power to the designated power consumption area and the number of energy storage units of the energy storage device configured for the designated power consumption area are the two inputs of the information conversion formula.

[0169] The information conversion formula, which uses two inputs and one output, expresses the numerical mapping relationship between the number of photovoltaic power generation units of the distributed photovoltaic power supply power source for the designated power consumption area and the number of energy storage units of the energy storage device configured for the designated power consumption area, and the total number of reference units corresponding to the target virtual power plant. The total number of reference units corresponding to the target virtual power plant is a single output of the information conversion formula.

[0170] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments, and any modifications or variations of the embodiments of the present invention may be made without departing from the stated principles.

Claims

1. A method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination, characterized in that, The method includes: The time window, or time scale, for power dispatching in the coordinated power supply, grid, load, and storage of the designated power consumption area under the jurisdiction of the target virtual power plant is selected. The target virtual power plant is used for AC power distribution. Construct a long short-term memory network model that corresponds to both the selected time scale and the target virtual power plant; Collect multi-dimensional source-grid-load-storage data at the selected time scale; Using a long short-term memory network model corresponding to the selected time scale and the target virtual power plant, based on the selected time scale, the number of photovoltaic power generation units of the distributed photovoltaic power supply power source supplying power to the designated power consumption area, the number of energy storage units of the energy storage device configured for the designated power consumption area, the maximum power supply value of the distributed photovoltaic power supply power source supplying power to the designated power consumption area, the maximum energy storage value of the energy storage device configured for the designated power consumption area, the maximum power consumption value of each power load in the designated power consumption area, and multi-dimensional source-grid-load-storage data under the selected time scale, the intelligent prediction of the power supply value of the distributed photovoltaic power supply power source supplying power to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval. Here, the current time interval is the subsequent time interval immediately adjacent to each historical time interval. Based on intelligent prediction results, the source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant is carried out for the current time interval.

2. The method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to claim 1, characterized in that: The time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale, and the number of hidden layer units of the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant. The total number of reference units corresponding to the target virtual power plant is equal to the sum of the number of photovoltaic power generation units of the distributed photovoltaic power supply that provides power to the designated power consumption area and the number of energy storage units configured for the energy storage devices in the designated power consumption area.

3. The method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to claim 2, characterized in that: The multi-dimensional source-grid-load-storage data collected at the selected time scale includes: the power supply values ​​of distributed photovoltaic power sources supplying power to the designated power consumption area in each historical time interval, the energy storage power values ​​of energy storage devices configured for the designated power consumption area in each historical time interval, and the load power values ​​of the designated power consumption area in each historical time interval. These data serve as the multi-dimensional source-grid-load-storage data at the selected time scale, with each time interval occupying a duration equal to the selected time scale. Among them, the target virtual power plant completes the source-grid-load-storage coordination of AC power distribution in the designated power consumption area through distributed photovoltaic power supply to the designated power consumption area, AC distribution network from the distributed photovoltaic power supply to the designated power consumption area, energy storage device configured for the designated power consumption area, and load switching device for the designated power consumption area. Among them, the source-grid-load-storage coordination of AC power distribution in the designated power consumption area under the jurisdiction of the target virtual power plant based on intelligent prediction results for the current time interval includes: when the maximum power supply value of the distributed photovoltaic power supply power supply to the designated power consumption area is less than the load power consumption value of the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured for the designated power consumption area is equal to the difference between the two, so as to complete the pure green power supply; Specifically, when the maximum power supply of the distributed photovoltaic power source supplying power to the designated power consumption area is less than the load power consumption of the designated power consumption area in the current time interval, the energy storage power of the energy storage device configured for the designated power consumption area in the current time interval is equal to the difference between the two, so as to complete the pure green power supply, includes: when the energy storage power of the energy storage device configured for the designated power consumption area in the current time interval is less than the difference between the two, the energy storage device configured for the designated power consumption area and other power sources work together to complete the power supply of the difference between the two.

4. The method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to claim 3, characterized in that: The system collects power supply values ​​for distributed photovoltaic power sources supplying power to designated power consumption areas in each historical time interval, energy storage power values ​​for energy storage devices configured for designated power consumption areas in each historical time interval, and load power values ​​for designated power consumption areas in each historical time interval. These data are used as multi-dimensional source-grid-load-storage data outputs at the selected time scale. The duration of each time interval is equal to the selected time scale. All collected power values ​​are represented using binary values ​​with a preset number of bits. Among them, the energy storage devices configured for the designated electricity consumption areas include various energy storage units distributed at various nodes of the power distribution network, and the maximum energy storage power of the energy storage devices configured for the designated electricity consumption areas is between 1MWh and 20MWh. The time step size of the Long Short-Term Memory (LSTM) network model follows the numerical change trend of the selected time scale as follows: if the selected time scale is 24 hours, the time step size of the LSM network model is 96 steps; if the selected time scale is 12 hours, the time step size of the LSM network model is 48 steps; and if the selected time scale is 1 hour, the time step size of the LSM network model is 24 steps.

5. The method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to claim 4, characterized in that, Before performing source-grid-load-storage coordination of AC power distribution in designated electricity-consuming areas under the jurisdiction of the target virtual power plant for the current time interval based on intelligent prediction results, the method further includes: Receive the power supply value of the distributed photovoltaic power supply power supply to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval, and store the power supply value of the distributed photovoltaic power supply power supply power supply to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval to the big data storage node corresponding to the target virtual power plant; This involves allocating different big data storage nodes to different virtual power plants, and transmitting the power supply values ​​of distributed photovoltaic power sources supplying power to designated power consumption areas in the current time interval, the energy storage power values ​​of energy storage devices configured for designated power consumption areas in the current time interval, and the load power consumption values ​​of designated power consumption areas in the current time interval, along with the power plant code values ​​of the target virtual power plant, over the network. This enables the targeted storage of the power supply values ​​of distributed photovoltaic power sources supplying power to designated power consumption areas in the current time interval, the energy storage power values ​​of energy storage devices configured for designated power consumption areas in the current time interval, and the load power consumption values ​​of designated power consumption areas in the current time interval to their matching big data storage nodes.

6. The method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to claim 4, characterized in that, Before performing source-grid-load-storage coordination of AC power distribution in designated electricity-consuming areas under the jurisdiction of the target virtual power plant for the current time interval based on intelligent prediction results, the method further includes: Receives the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval, and performs local synchronous display on the power supply value of the distributed photovoltaic power supply powering the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval; The local synchronous display of the power supply value of the distributed photovoltaic power supply power source supplying power to the designated power consumption area in the current time interval, the energy storage power value of the energy storage device configured to supply power to the designated power consumption area in the current time interval, and the load power consumption value of the designated power consumption area in the current time interval includes: using an LCD screen to complete the local synchronous display.

7. The method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to claim 4, characterized in that, Before the target virtual power plant is used for AC power distribution, the method further includes: (1) Selecting a time window, i.e., a time scale, for power dispatching in a coordinated manner for the source-grid-load-storage area under the jurisdiction of the target virtual power plant. Set up a user access interface for the target virtual power plant. The user access interface is used to provide a user selection interface for the selection of the power dispatch time window. Specifically, a user access interface is set up for the target virtual power plant. The user access interface is used to provide a user selection interface for the selection of the power scheduling time window. This includes: the operator of the target virtual power plant selects the power scheduling time window through the API function of the user access interface.

8. The method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to claim 4, characterized in that, After constructing a long short-term memory network model corresponding to the selected time scale and the target virtual power plant, the method further includes: By storing the various model data of the long short-term memory network model that corresponds to the selected time scale and the target virtual power plant, the model storage of the long short-term memory network model that corresponds to the selected time scale and the target virtual power plant is completed. Specifically, the model storage of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant is completed by storing the various model data of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant. This includes: synchronously storing the selected time scale, the power plant code value of the target virtual power plant, and the various model data of the long short-term memory network model corresponding to the selected time scale and the target virtual power plant into the same physical storage unit.

9. A method for multi-timescale power dispatching of a virtual power plant with source-grid-load-storage coordination according to any one of claims 5-8, characterized in that: The time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale, including: using a scale change curve to represent the numerical change trend of the selected time scale, and using a step size change curve to represent the numerical change trend of the time step of the Long Short-Term Memory Network Model. The time step of the Long Short-Term Memory Network Model follows the numerical change trend of the selected time scale, which further includes the fact that the step size change curve and the scale change curve completely overlap after being normalized to the same length. The fact that the number of hidden layer units in the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant includes: using a quantitative analytical function to represent the numerical mapping relationship between the number of hidden layer units in the Long Short-Term Memory Network Model and the total number of reference units corresponding to the target virtual power plant; The fact that the number of hidden layer units in the Long Short-Term Memory Network Model is proportional to the total number of reference units corresponding to the target virtual power plant also includes: in the quantity analysis function, the number of hidden layer units in the Long Short-Term Memory Network Model is the output value, and the total number of reference units corresponding to the target virtual power plant is the input value.

10. A virtual power plant multi-timescale power dispatching system for source-grid-load-storage coordination, characterized in that, The system includes: The scale selection device is used to select the time window, i.e., the time scale, for the source-grid-load-storage coordination of the power dispatching of the designated power consumption area under the jurisdiction of the target virtual power plant. The target virtual power plant is used for AC power distribution. Model building components are used to build a long short-term memory network model that corresponds to the selected time scale and the target virtual power plant. Multidimensional acquisition devices are used to acquire multidimensional source-grid-load-storage data at selected time scales. The state prediction device is connected to the scale selection device, the model building device, and the multi-dimensional acquisition device, respectively. It is used to intelligently predict the power supply of the distributed photovoltaic power supply to the set power consumption area in the current time interval, the energy storage power of the energy storage device configured to the set power consumption area, and the load power consumption of the set power consumption area in the current time interval, based on the selected time scale and the long short-term memory network model corresponding to the target virtual power plant. The prediction is based on the selected time scale, the number of photovoltaic power generation units of the distributed photovoltaic power supply to the set power consumption area, the number of energy storage units of the energy storage device configured to the set power consumption area, the maximum power supply value of the distributed photovoltaic power supply to the set power consumption area, the maximum energy storage value of the energy storage device configured to the set power consumption area, the maximum power consumption value of each load in the set power consumption area, and the multi-dimensional source-grid-load-storage data under the selected time scale. The current time interval is the subsequent time interval of each historical time interval. Power dispatching devices, connected to state prediction devices, are used to perform source-grid-load-storage coordination of AC power distribution in designated power consumption areas under the jurisdiction of a target virtual power plant for the current time interval based on intelligent prediction results.

Images