Virtual power plant intelligent dispatching method and system based on space-time network model

Abstract

The application provides a kind of virtual power plant intelligent scheduling method and system based on space-time network model, belong to intelligent power plant technical field.The method of the application includes collecting real-time data from distributed power station and weather station;The collected data is preprocessed;Using space-time neural network unit, construct space-time network model based on encoder-decoder, and import the data after preprocessing, predict the expected working power of each device of virtual power plant;According to the prediction result of space-time network model, combined with the real-time state of virtual power plant and actual constraint condition, intelligent scheduling decision is made.Through the end-to-end design of space-time network model, the electrical data, weather data, date data, equipment parameter data, market operation data and geographic location information are comprehensively considered, and the detailed scheduling strategy and scheme can be directly obtained, which not only improves the reliability and economy of power supply, but also helps to optimize energy structure, promote energy saving and emission reduction and sustainable development.

Description

Technical Field

[0001] This invention belongs to the field of smart power plant technology, and more specifically, relates to a virtual power plant intelligent scheduling method and system based on a spatiotemporal network model. Background Technology

[0002] One of the key characteristics of virtual power plants is that their power sources, loads, and energy storage devices are often distributed across different geographical locations. This spatial distribution significantly impacts the formulation and implementation of dispatch strategies. While time-series data primarily reflects the patterns of data change over time, spatially distributed data reveals the distribution characteristics of data across different geographical locations. Existing virtual power plant dispatch strategies often focus on the analysis of time-series data in data collection and processing, neglecting the spatial distribution properties of the data.

[0003] The spatial distribution of data allows for forward-looking predictions about power output, load, and energy storage devices. For example, strong cooling typically progresses from north to south at a certain speed. Wind data also shows significant changes over time along the wind direction. If the spatial distribution of distributed power sources, loads, and energy storage devices is not fully considered, even if the system as a whole operates normally, power supply in certain areas may be insufficient or excessive, thus affecting the overall efficiency and stability of the power system. Summary of the Invention

[0004] The main objective of this application is to provide a virtual power plant intelligent scheduling method and system based on a spatiotemporal network model, which can improve the accuracy and rationality of scheduling strategies by combining the spatial distribution properties of data.

[0005] To achieve the above objectives, firstly, this application proposes a virtual power plant intelligent scheduling method based on a spatiotemporal network model, comprising:

[0006] Real-time data is collected from sensors, smart meters, and weather stations in distributed power stations; the collected real-time data includes power equipment data, energy storage equipment data, controllable load data, meteorological data, date data, market and price data, manual control data, and geographic location information.

[0007] The collected real-time data is preprocessed; wherein, the preprocessing includes data cleaning, format conversion, anomaly detection, and data normalization;

[0008] Using spatiotemporal network basic units, a spatiotemporal network model based on encoder-decoder is constructed; based on the preprocessed data, the expected operating power and expected load of each device in the virtual power plant are predicted using the spatiotemporal network model.

[0009] According to the predicted result of the space-time network model, intelligent scheduling decisions are made in combination with real-time states of the virtual power plant and actual constraint conditions, wherein the real-time states of the virtual power plant include real-time information of device operation, real-time information of demand, and business strategy.

[0010] Further, the space-time network model comprises a convolutional neural network and a recurrent neural network, wherein the convolutional neural network is configured to transmit a hidden state value at a previous time point to a next node through spatial convolution processing, and the recurrent neural network is configured to calculate an output value and a new hidden state value according to input data and the hidden state value at the previous time point.

[0011] Further, the preprocessing of the collected real-time data further comprises: drawing a planar graph of power supply devices, energy storage devices, and controllable loads in the virtual power plant according to geographic location information, dividing the planar graph into blocks according to geographic coordinates to form a rectangular graph composed of a plurality of small squares, and assigning each square to a numerical value representing the value of a certain input feature in the geographic area, so as to form a three-dimensional tensor of multiple features at the same time.

[0012] Further, it further comprises: in response to a user operation, displaying the predicted result of the space-time network model, the scheduling decision, and the real-time data on a visual interface to support the user to monitor and operate.

[0013] In a second aspect, the present application further provides a virtual power plant intelligent scheduling system based on a space-time network model, comprising a master control center, a plurality of distributed data acquisition and device control modules, and a network module, wherein the master control center and the distributed data acquisition and device control modules are connected through the network module.

[0014] The distributed data acquisition and device control module is configured to collect real-time data from sensors, smart meters, and weather stations of distributed power plants, wherein the collected real-time data comprises power supply device data, energy storage device data, controllable load data, weather data, date data, market and price data, manual control data, and geographic location information.

[0015] The collected real-time data is preprocessed and sent to the master control center through the network module, wherein the preprocessing comprises data cleaning, format conversion, anomaly detection, and data normalization.

[0016] The master control center uses a space-time network basic unit to construct an encoder-decoder-based space-time network model, and uses the space-time network model to predict expected working power and expected load of each device of the virtual power plant based on the preprocessed data.

[0017] Based on the prediction results of the spatiotemporal network model, combined with the real-time status of the virtual power plant and actual constraints, intelligent scheduling decisions are made; wherein, the real-time status of the virtual power plant includes real-time equipment operation information, real-time demand information, and operating strategies.

[0018] The distributed data acquisition and equipment control module is also used to convert the instructions of the main control center for intelligent scheduling decisions into actual actions to operate the equipment.

[0019] Furthermore, the spatiotemporal network model includes a convolutional neural network and a recurrent neural network; wherein, the convolutional neural network is used to transmit the hidden state value of the previous time point to the next node through spatial convolution processing, and the recurrent neural network is used to calculate the output value and the new hidden state value based on the input data and the hidden state value of the previous time point.

[0020] Furthermore, the preprocessing of the collected real-time data also includes: drawing a planar graphic of the power equipment, energy storage equipment, and controllable load in the virtual power plant based on the geographical location information; dividing the planar graphic into blocks according to the geographical coordinates to form a rectangular map composed of several small squares; and assigning a value to each square to represent the value of a certain input feature in the geographical area, so as to form a three-dimensional tensor from multiple features at the same time.

[0021] Furthermore, it also includes an input module and an output module. The output module is used to respond to user operations and display the prediction results, scheduling decisions and real-time data of the spatiotemporal network model on a visual interface. The input module is used to support user monitoring and operation.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) By using the end-to-end design of the spatiotemporal network model, and taking into account power equipment data, energy storage equipment data, controllable load data, meteorological data (such as rainfall, humidity, temperature and wind speed), date data, market and price data, manual control data and geographical location information, detailed scheduling strategies and plans can be obtained directly, which reduces the operating and maintenance costs in the power dispatching process.

[0024] (2) The working efficiency of the output equipment and energy storage equipment of the distributed power station has been optimized, the degree of user demand has been improved, the differentiated management of various equipment in different regions has been realized, the reliability of important equipment has been improved, and the peak-valley regulation and safe and stable operation of the power grid have been facilitated. Attached Figure Description

[0025] Figure 1 A flowchart illustrating the intelligent scheduling method for virtual power plants based on a spatiotemporal network model provided in an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of the spatiotemporal network model provided in an embodiment of the present invention;

[0027] Figure 3 A logical schematic diagram of a CRNN unit provided in an embodiment of the present invention;

[0028] Figure 4 A schematic diagram of a three-dimensional tensor provided in an embodiment of the present invention;

[0029] Figure 5 A schematic diagram illustrating the workflow of the encoder and decoder provided in an embodiment of the present invention;

[0030] Figure 6 This is a schematic diagram of the structure of a virtual power plant intelligent dispatching system based on a spatiotemporal network model, provided in an embodiment of the present invention.

[0031] The attached diagram is labeled as follows: 1. Main control center; 2. Distributed data acquisition and equipment control module; 3. Network module; 4. Storage and computing module; 5. Input module; 6. Output module; 7. Data acquisition device; 8. Equipment control actuator. Detailed Implementation

[0032] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0033] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0034] Example 1

[0035] Existing literature on virtual power plant dispatch strategies primarily focuses on how to achieve centralized management, optimized dispatch, and resource allocation of distributed power sources through advanced algorithms and technologies. These technologies cover the entire process from data collection, processing, and analysis to dispatch decisions, aiming to improve the reliability, stability, and economy of power supply. Some literature proposes machine learning-based predictive models to predict the output of distributed power sources and user demand, thereby enabling better dispatch decisions.

[0036] One of the key characteristics of virtual power plants is that their power sources, loads, and energy storage devices are often distributed across different geographical locations. This spatial distribution significantly impacts the formulation and implementation of dispatch strategies. Existing virtual power plant dispatch strategies often focus on analyzing time-series data in terms of data collection and processing, neglecting the spatial distribution properties of the data.

[0037] The spatial distribution of data allows for forward-looking predictions about power output, load, and energy storage. For example, strong cooling typically progresses from north to south at a certain speed. Wind data also shows significant changes over time along the wind direction. If the spatial distribution of distributed power sources, loads, and energy storage devices is not fully considered, even if the system as a whole operates normally, some areas may experience power shortages or surpluses, thus affecting the overall efficiency and stability of the power system.

[0038] Based on this, please refer to Figure 1 Firstly, this embodiment provides a virtual power plant intelligent scheduling method based on a spatiotemporal network model, comprising five parts: data acquisition, data preprocessing, spatiotemporal network model, scheduling decision, and user interaction. Specific steps are as follows: S100 to S500.

[0039] S100 collects real-time data from sensors, smart meters, and weather stations at the distributed power station. This real-time data includes power equipment data, energy storage equipment data, controllable load data, meteorological data, date data, market and price data, manual control data, and geographic location information. Meteorological data includes rainfall, humidity, temperature, and wind speed.

[0040] S200: The collected real-time data is preprocessed and sent to the main control center via a network module; wherein, the preprocessing includes data cleaning, format conversion, anomaly detection, and data normalization.

[0041] S300. Using spatiotemporal network basic units, construct a spatiotemporal network model based on encoder-decoder; based on the preprocessed data, use the spatiotemporal network model to predict the expected operating power and expected load of each device in the virtual power plant.

[0042] S400. Based on the prediction results of the spatiotemporal network model, and combined with the real-time status of the virtual power plant and actual constraints, intelligent scheduling decisions are made; wherein, the real-time status of the virtual power plant includes real-time equipment operation information, real-time demand information, and operating strategies.

[0043] S500, in response to user operation, displays the prediction results, scheduling decisions and real-time data of the spatiotemporal network model on a visual interface to support user monitoring and operation.

[0044] The spatiotemporal network model of this embodiment is used to predict the expected operating power and expected load of each device in a virtual power plant, thereby obtaining the device's operation plan (or operation map). This spatiotemporal network model considers historical electrical data, historical meteorological data (such as rainfall, humidity, temperature, and wind speed), date data, device parameter data, market operation data, and geographical location information, directly yielding the device's operation map. (The last sentence appears to be incomplete and possibly refers to a different method or approach.) Data from each moment is input into the network model to predict the future. The operation chart at each point in time is used; the values ​​of and are determined by comprehensively considering computational costs and control effectiveness, and are continuously iterated over time. Manual intervention is also possible in case of urgent or special needs.

[0045] The method in this embodiment, through end-to-end design of a spatiotemporal network model, comprehensively considers power equipment data, energy storage equipment data, controllable load data, meteorological data (such as rainfall, humidity, temperature, and wind speed), date data, market and price data, manual control data, and geographical location information. This allows for the direct generation of detailed scheduling strategies and plans, reducing operating and maintenance costs during power dispatching. It optimizes the working efficiency of distributed power station output equipment and energy storage equipment, improves the satisfaction of user needs, enables differentiated management of various equipment in different regions, enhances the reliability of critical equipment, and contributes to peak-valley regulation and safe and stable operation of the power grid.

[0046] For further details, please refer to Figure 2 , Figure 2 The structure of the spatiotemporal network model is shown. The spatiotemporal network model includes convolutional neural networks and recurrent neural networks, which comprehensively consider the spatial and temporal distribution characteristics of features. Figure 2 Each directed line segment in the diagram represents a signal that has undergone spatial convolution processing and is transmitted to the next node.

[0047] Circular symbols represent input and output. This represents the network input data. This represents the network's output data. These results are multi-dimensional vectors, also known as tensors.

[0048] A rectangular pattern represents a basic unit of a spatiotemporal network (CRNN unit), which is the core part of the spatiotemporal network model. It is used to calculate the output value and the new hidden state value based on the input data and the hidden state value of the previous time point.

[0049] and , ... This represents the hidden state value of a CRNN unit. 0, 1, 2...t. indicates the time sequence, and the subscript init indicates the initial value.

[0050] , ... This represents the input value of the spatiotemporal network.

[0051] Hidden state value at the previous time point , and the input value at the current time point Together they determine the output of the current state. New hidden state and .

[0052] Please see Figure 3 , Figure 3 This is a logical diagram of a CRNN unit. Figure 3 The circular boxes in the diagram represent positional multiplication or addition operations between tensors. The rectangular boxes represent activation operations performed on each element of the tensor individually. Merging line segments indicate that two tensors are concatenated. Output forks indicate that multiple branches have equal values, meaning the tensor has been copied. (Except for...) , In addition, the map also shows... , , The calculation methods for each variable are as follows:

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] The asterisk (*) indicates a convolution operation. The sign indicates that the components are multiplied together, σ represents the sigmoid activation function, and tanh represents the hyperbolic tangent activation function. For a convolution kernel, the first letter of its subscript represents the source data, and the second letter represents the result data. For example... Indicates from source data To the result data The convolution kernel. The bias represents the input data. Both the convolution kernel and the bias are parameters of the spatiotemporal network, which need to be obtained through training with the input data.

[0059] Furthermore, sufficient high-quality data is crucial for training neural networks. In spatiotemporal networks, usable data includes power equipment data, energy storage equipment data, controllable load data, meteorological data, date data, market and price data, manual control data, and geographic location information, among others.

[0060] Meteorological data includes sampled values ​​of data such as temperature, humidity, wind speed, wind direction, and rainfall in both time and space dimensions;

[0061] Calendar data includes the current date's month of the year, day of the month, day of the week, and whether it is a holiday, etc.

[0062] Power equipment data includes the output ratio of various new energy power sources, the highest processing power, the optimal output power, and the output ratio of conventional units.

[0063] Energy storage device data includes the charging and discharging power of the energy storage device, etc.

[0064] Controllable load data includes controllable load power, demand response strategies, etc.

[0065] Market and price data include electricity market prices, carbon emission rights prices, etc.

[0066] Manually controlled data includes data that managers deem necessary to add, such as system constraints;

[0067] Geographic location information data includes the address location information of various power supply equipment, energy storage equipment, controllable loads, etc.

[0068] The collected data is preprocessed, including removing outliers, filling in missing values, and normalizing the data, to ensure the accuracy and consistency of the data.

[0069] Based on geographic location information, a planar diagram of the power supply equipment, energy storage equipment, and controllable loads within a virtual power plant is drawn. This planar diagram is then divided into blocks, forming a rectangular diagram composed of several small squares. Each square is assigned a numerical value representing the value of a certain input feature within that geographic area, thus forming a three-dimensional tensor from multiple features at the same time. Figure 4 As shown.

[0070] The output of a neural network is similar to its input; it is also a multidimensional tensor, and the features of each square correspond one-to-one with the squares of the corresponding input tensor.

[0071] Furthermore, in addition to the input module, output module, and CRNN unit, the spatiotemporal network model also includes some essential basic modules, such as the loss function and optimizer. Depending on the needs of the specific case, Batch Normalization (BN) layers (a regularization method that can accelerate network convergence and improve generalization ability) and dropout layers (to further improve generalization ability) can be added. To address the issue that both input and output involve multiple time steps, this invention also employs an encoder-decoder structure.

[0072] The spatiotemporal network model uses MSE (mean squared error) as the loss function, taking the data from the next time step as the target value for predicting the current data. The closer the two are, the smaller the MSE, indicating better network performance.

[0073] The output layer also includes an optimizer, which updates the network parameters through the error backpropagation algorithm. Various optimizers, including Adam, can be used to optimize the network.

[0074] When training the spatiotemporal network model, the current time step data is used as input and the next time step data is used as output. The system optimizes the network parameters through the backpropagation algorithm of the error.

[0075] Furthermore, in practical applications, spatiotemporal network models typically involve continuous input. , ... common Data at each moment, and predictions for the future. Output at time 1 , ... This is called an encoder-decoder structure, which means that the encoder-decoder structure is a combination of encoder and decoder. The information from the data at each time point is encoded and stored in a hidden state. , In the middle, and use it to predict the future. The output at each moment, such as Figure 6 As shown, the intelligent scheduling system performs scheduling based on the output results of the spatiotemporal network.

[0076] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating the workflow of the encoder and decoder. Figure 5 The encoder on the left encodes the input data, while the decoder on the right produces the actual output. Both the encoder and decoder are built on the same neural network, based on CRNN units. The encoder receives actual data as input, while the decoder makes predictions about the future and doesn't receive actual data; instead, it uses the network's output from the previous time step as the input for the next time step. The encoder also produces an output. , However, only the output value at the last moment is available. It is used as the first input to the decoder, and all other output values ​​are discarded.

[0077] Figure 5 The indices 0, 1, and 2 in the code are merely examples of time series data and do not represent a limitation to processing only three time points. In reality, the number of input and prediction sequences are independent and can be freely chosen. To achieve the desired effect, the number of input time points can be greater, and the number of output time points less. These are determined based on the actual data, prediction performance, and requirements during the training and evaluation of the spatiotemporal network.

[0078] Through the above implementation methods, the virtual power plant intelligent dispatching method based on a spatiotemporal network model in this embodiment can achieve precise control of the operating power of each device in a distributed power station and formulate the optimal intelligent dispatching scheme accordingly. This not only improves the reliability and economy of power supply, but also helps to optimize the energy structure, promote energy conservation and emission reduction, and achieve sustainable development.

[0079] Example 2

[0080] Based on the above implementation methods, please refer to Figure 6 Secondly, this embodiment also provides a virtual power plant intelligent dispatching system based on a spatiotemporal network model, including a main control center 1, a distributed data acquisition and equipment control module 2, and a network module 3. The main control center 1 and the distributed data acquisition and equipment control module 2 are connected through the network module 3. The main control center 1 includes a storage and processing module 4, an input module 5, and an output module 6.

[0081] Storage and computation module 4 is the brain of the system, used to run various necessary software, among which the intelligent scheduling method based on the spatiotemporal network model is the most important. Other software includes handling user input / output and network communication. Input module 5 receives user operations, such as system settings and manual operations. Output module 6 displays the system status and results. Network module 3 communicates with the various distributed data acquisition and device control modules 2.

[0082] The distributed data acquisition and equipment control module 2 is typically used in conjunction with the power supply, energy storage, and controllable load devices of a virtual power plant to collect data and receive scheduling commands from the main control center 1. One main control center 1 can connect to multiple distributed data acquisition and equipment control modules 2. The data acquisition and equipment control module 2 also has a storage and processing module 4, an input module 5, and an output module 6, with functions similar to the corresponding modules in the main control center 1. In addition, it has a data acquisition unit 7 and an equipment control actuator 8. The data acquisition unit 7 can connect to various data acquisition devices and transmit data to the main control center 1 via the network module 3. The function of the equipment control actuator 8 is to translate the commands from the main control center 1 into actual actions that control the equipment.

[0083] In this embodiment, the virtual power plant intelligent dispatching system based on the spatiotemporal network model runs the virtual power plant intelligent dispatching method based on the spatiotemporal network model described in any of the above embodiments. Specifically,

[0084] The distributed data acquisition and equipment control module 2 is used to collect real-time data from the sensors, smart meters and weather stations of the distributed power station. The collected real-time data includes power equipment data, energy storage equipment data, controllable load data, meteorological data (such as rainfall, humidity, temperature and wind speed), date data, market and price data, manual control data and geographical location information, etc.

[0085] The collected data is preprocessed and then sent to the main control center 1 via network module 3; wherein, the preprocessing includes data cleaning, format conversion, anomaly detection, data normalization, etc.

[0086] The main control center 1 uses spatiotemporal network basic units to construct a spatiotemporal network model based on encoder-decoder; based on the preprocessed data, it uses the spatiotemporal network model to predict the expected operating power and expected load of each device in the virtual power plant.

[0087] Based on the prediction results of the spatiotemporal network model, combined with the real-time status of the virtual power plant and actual constraints, intelligent scheduling decisions are made; wherein, the real-time status of the virtual power plant includes real-time equipment operation information, real-time demand information, and operating strategies, etc.

[0088] The distributed data acquisition and equipment control module 2 is also used to convert the instructions of the main control center 1 for intelligent scheduling decisions into actual actions to operate the equipment.

[0089] Furthermore, the spatiotemporal network model includes a convolutional neural network and a recurrent neural network; wherein, the convolutional neural network is used to transmit the hidden state value of the previous time point to the next node through spatial convolution processing, and the recurrent neural network module is used to calculate the output value and the new hidden state value based on the input data and the hidden state value of the previous time point.

[0090] Furthermore, the preprocessing of the collected data also includes: drawing a planar graphic of the power equipment, energy storage equipment, and controllable load in the virtual power plant based on the geographical location information; dividing the planar graphic into blocks according to the geographical coordinates to form a rectangular graphic composed of several small squares; and assigning a value to each square to represent the value of a certain input feature in the geographical area, so as to form a three-dimensional tensor from multiple features at the same time.

[0091] Furthermore, it also includes an input module 5 and an output module 6. The output module 6 is used to respond to user operations and display the prediction results, scheduling decisions and real-time data of the spatiotemporal network model on a visual interface. The input module 5 is used to support user monitoring and operation.

[0092] The virtual power plant intelligent dispatching system based on the spatiotemporal network model in this embodiment reduces the operating and maintenance costs in the power dispatching process through intelligent dispatching and real-time monitoring.

[0093] The above description is merely illustrative of the embodiments of the present invention and is not intended to limit the present invention. For those skilled in the art, any modifications, equivalent substitutions, improvements, etc., made without creative effort within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A virtual power plant intelligent scheduling method based on a spatiotemporal network model, characterized in that, include: Real-time data is collected from sensors, smart meters, and weather stations in distributed power stations; the collected real-time data includes power equipment data, energy storage equipment data, controllable load data, meteorological data, date data, market and price data, manual control data, and geographic location information. The collected real-time data is preprocessed; the preprocessing includes data cleaning, format conversion, anomaly detection, and data normalization; the preprocessing of the collected real-time data also includes: drawing a planar graphic of the power equipment, energy storage equipment, and controllable load in the virtual power plant based on geographical location information, dividing the planar graphic into blocks according to geographical coordinates to form a rectangular map composed of several small squares, and assigning a value to each square to represent the value of a certain input feature in the geographical area, so as to form a three-dimensional tensor from multiple features at the same time. A spatiotemporal network model based on an encoder-decoder is constructed using spatiotemporal network basic units. Based on preprocessed data, the spatiotemporal network model is used to predict the expected operating power and expected load of each device in a virtual power plant. The spatiotemporal network model includes a convolutional neural network and a recurrent neural network. The convolutional neural network is used to transmit the hidden state value of the previous time point to the next node through spatial convolution processing, and the recurrent neural network is used to calculate the output value and the new hidden state value based on the input data and the hidden state value of the previous time point. Based on the prediction results of the spatiotemporal network model, and combined with the real-time status of the virtual power plant and actual constraints, intelligent scheduling decisions are made; wherein, the real-time status of the virtual power plant includes real-time equipment operation information, real-time demand information, and operating strategies.

2. The intelligent scheduling method for virtual power plants based on a spatiotemporal network model according to claim 1, characterized in that, Also includes: In response to user actions, the results of the spatiotemporal network model predictions, scheduling decisions, and real-time data are displayed on a visual interface to support user monitoring and operation.

3. A virtual power plant intelligent dispatching system based on a spatiotemporal network model, characterized in that, It includes a main control center, multiple distributed data acquisition and equipment control modules, and a network module. The main control center and the distributed data acquisition and equipment control modules are connected through the network module. The distributed data acquisition and equipment control module is used to collect real-time data from sensors, smart meters, and weather stations in the distributed power station; the collected real-time data includes power equipment data, energy storage equipment data, controllable load data, meteorological data, date data, market and price data, manual control data, and geographical location information. The collected real-time data is preprocessed and sent to the main control center via a network module. The preprocessing includes data cleaning, format conversion, anomaly detection, and data normalization. The preprocessing of the collected real-time data also includes: drawing a planar graphic of the power equipment, energy storage equipment, and controllable load in the virtual power plant based on geographical location information; dividing the planar graphic into blocks according to geographical coordinates to form a rectangular map composed of several small squares; and assigning a value to each square to represent the value of a certain input feature in the geographical area, so as to form a three-dimensional tensor from multiple features at the same time. The main control center utilizes spatiotemporal network basic units to construct an encoder-decoder-based spatiotemporal network model. Based on preprocessed data, the spatiotemporal network model is used to predict the expected operating power and expected load of each device in the virtual power plant. The spatiotemporal network model includes convolutional neural networks and recurrent neural networks. The convolutional neural network is used to transmit the hidden state value from the previous time point to the next node through spatial convolution processing, and the recurrent neural network is used to calculate the output value and the new hidden state value based on the input data and the hidden state value from the previous time point. Based on the prediction results of the spatiotemporal network model, combined with the real-time status of the virtual power plant and actual constraints, intelligent scheduling decisions are made; wherein, the real-time status of the virtual power plant includes real-time equipment operation information, real-time demand information, and operating strategies. The distributed data acquisition and equipment control module is also used to convert the instructions of the main control center for intelligent scheduling decisions into actual actions to operate the equipment.

4. The virtual power plant intelligent dispatching system based on a spatiotemporal network model according to claim 3, characterized in that, It also includes an input module and an output module. The output module is used to respond to user operations and display the prediction results, scheduling decisions and real-time data of the spatiotemporal network model on a visual interface. The input module is used to support user monitoring and operation.

Images