CN-122026308-A - Virtual power plant system, load prediction method thereof and readable storage medium

CN122026308ACN 122026308 ACN122026308 ACN 122026308ACN-122026308-A

Abstract

The embodiment of the invention provides a virtual power plant system, a load prediction method thereof and a readable storage medium, and belongs to the technical field of virtual power plant prediction. The load prediction method comprises the steps of obtaining historical load data and historical meteorological data of each distributed power supply in a virtual power plant system, obtaining depth characteristics according to the historical load data and the historical meteorological data, building a load prediction network model, training the load prediction network model by the aid of the training set, obtaining real-time load of each distributed power supply and real-time accumulated load of all distributed power supplies in the virtual power plant system, and comprehensively analyzing to obtain final predicted load by means of obtaining the predicted load of each distributed power supply and total predicted load of all distributed power supplies.

Inventors

SU HONGLI
LV ZITONG
ZHAO QING
WANG SINING
WEI ZHIFENG
LIU HAIYANG
GUO PUPU
ZHAO MENGJIE
NIU NING
ZHAO ZHENDONG
HAO XIUYAN

Assignees

国网思极数字科技(北京)有限公司
北京国网信通埃森哲信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251217

Claims (10)

1. A method of load prediction for a virtual power plant system, comprising: acquiring historical load data and historical meteorological data of each distributed power supply in a virtual power plant system; acquiring depth characteristics according to the historical load data and the historical meteorological data, and constructing a training set; Constructing a load prediction network model; Training the load prediction network model by adopting the training set; acquiring the real-time load of each distributed power supply in the current virtual power plant system; Inputting the real-time load of each distributed power supply into the trained load prediction network model to obtain the current predicted load of each distributed power supply; acquiring real-time accumulated loads of all distributed power supplies in the current virtual power plant system; inputting the real-time accumulated load into the trained load prediction network model to obtain the total predicted load of all the current distributed power supplies; and obtaining a final predicted load according to the current predicted load of each distributed power supply and the overall predicted load.
2. The load prediction method of claim 1, wherein obtaining historical load data and historical meteorological data for each distributed power source in a virtual power plant system comprises: Carrying out data cleaning and data fusion on the historical load data and the historical meteorological data; the historical load data is time aligned with the historical meteorological data.
3. The load prediction method of claim 2, wherein obtaining depth features from the historical load data and the historical meteorological data and constructing a training set comprises: Extracting characteristics of the historical load data; Extracting characteristics of the historical meteorological data; and constructing a training set of each distributed power supply according to the characteristics of the historical load data and the characteristics of the historical meteorological data.
4. The load prediction method of claim 1, wherein the load prediction network model comprises a long-term memory network embedded in a transducer encoder layer.
5. The load prediction method of claim 4, wherein training the load prediction network model using the training set comprises: inputting sequences in the training set into the transducer encoder layer; the transducer encoder layer outputs a new sequence; Inputting the new sequence into the long-term and short-term memory network; Inputting the output of the long-period memory network to a full-connection layer to obtain model output; And optimizing parameters of the load prediction network model according to the model output.
6. The load prediction method according to claim 1, wherein obtaining the real-time cumulative load of all distributed power sources in the virtual power plant system at present comprises: acquiring real-time accumulated loads of all distributed power sources according to a formula (1), ,(1) Wherein, the At the present time for all of the distributed power sources The load is accumulated in real time at the moment, Is the first Each of the distributed power sources is currently The real-time load of the moment in time, For the number of distributed power sources in the virtual power plant system, Numbered as integers.
7. The load prediction method according to claim 6, wherein obtaining a final predicted load from the current predicted load of each of the distributed power sources and the overall predicted load comprises: The cumulative predicted load of all the distributed power sources is obtained according to the formula (2), ,(2) Wherein, the For all the distributed power sources The cumulative predicted load at the moment in time, In order to predict the period of time, Is the first Each of the distributed power sources is at Predicted load at time; Obtaining a load prediction threshold range of the virtual power plant system according to formula (3), ,(3) Wherein, the For a load forecast range of the virtual power plant system, For the ratio of the threshold value, A load is predicted for the population.
8. The load prediction method according to claim 7, wherein obtaining a final predicted load from the current predicted load of each of the distributed power sources and the overall predicted load further comprises: judging whether the accumulated predicted load is within the load prediction threshold range; outputting the accumulated predicted load as a final predicted load if the accumulated predicted load is determined to be within the load prediction threshold range; and outputting the total predicted load as a final predicted load in the case where the accumulated predicted load is judged not to be within the load prediction threshold range.
9. A virtual power plant system, comprising: A plurality of distributed power sources; a virtual power plant connected to a plurality of said distributed power sources for performing the predictive method of any one of claims 1-8; And the terminal load is connected with a plurality of distributed power sources and the virtual power plant.
10. A computer readable storage medium storing instructions for reading by a machine to cause the machine to perform the load prediction method of any one of claims 1-8.

Description

Virtual power plant system, load prediction method thereof and readable storage medium Technical Field The invention relates to the technical field of virtual power plant prediction, in particular to a virtual power plant system, a load prediction method thereof and a readable storage medium. Background The virtual power plant utilizes advanced technologies such as digitalization and intellectualization to aggregate, coordinate and optimize the resources such as adjustable load, distributed click and energy storage in a certain area at the demand side, and combines a corresponding power market mechanism to form a system with the capability of responding to the running adjustment of the power grid. Virtual power plants predict the load for a particular period of time in the future based on the distributed power sources within their aggregate range. At present, for load prediction in an aggregation range, a mode of training a neural network model to perform load prediction is mostly adopted in a virtual power plant. However, this method relies on a single output of the neural network model, which results in poor load prediction accuracy and low reliability. The inventor of the present application has found that the above-mentioned scheme in the prior art has the defects of poor load prediction accuracy and low reliability in the process of implementing the present application. Disclosure of Invention It is an object of an embodiment of the present invention to provide a virtual power plant system and a load prediction method thereof, a readable storage medium, the virtual power plant system, the load prediction method thereof and the readable storage medium have the functions of high load prediction precision and high reliability. In order to achieve the above object, an aspect of an embodiment of the present invention provides a load prediction method for a virtual power plant system, including: acquiring historical load data and historical meteorological data of each distributed power supply in a virtual power plant system; acquiring depth characteristics according to the historical load data and the historical meteorological data, and constructing a training set; Constructing a load prediction network model; Training the load prediction network model by adopting the training set; acquiring the real-time load of each distributed power supply in the current virtual power plant system; Inputting the real-time load of each distributed power supply into the trained load prediction network model to obtain the current predicted load of each distributed power supply; acquiring real-time accumulated loads of all distributed power supplies in the current virtual power plant system; inputting the real-time accumulated load into the trained load prediction network model to obtain the total predicted load of all the current distributed power supplies; and obtaining a final predicted load according to the current predicted load of each distributed power supply and the overall predicted load. Optionally, obtaining historical load data and historical meteorological data for each distributed power source in the virtual power plant system includes: Carrying out data cleaning and data fusion on the historical load data and the historical meteorological data; the historical load data is time aligned with the historical meteorological data. Optionally, obtaining depth features according to the historical load data and the historical meteorological data, and constructing a training set includes: Extracting characteristics of the historical load data; Extracting characteristics of the historical meteorological data; and constructing a training set of each distributed power supply according to the characteristics of the historical load data and the characteristics of the historical meteorological data. Optionally, the load prediction network model includes a long-term memory network embedded in the transducer encoder layer. Optionally, training the load prediction network model using the training set includes: inputting sequences in the training set into the transducer encoder layer; the transducer encoder layer outputs a new sequence; Inputting the new sequence into the long-term and short-term memory network; Inputting the output of the long-period memory network to a full-connection layer to obtain model output; And optimizing parameters of the load prediction network model according to the model output. Optionally, obtaining the real-time accumulated load of all distributed power sources in the virtual power plant system comprises: acquiring real-time accumulated loads of all distributed power sources according to a formula (1), ,(1) Wherein, the At the present time for all of the distributed power sourcesThe load is accumulated in real time at the moment,Is the firstEach of the distributed power sources is currentlyThe real-time load of the moment in time,For the number of distributed power sources in the virtual power plant