A new energy photovoltaic short-term power prediction method and system

Abstract

The present application relates to a kind of new energy photovoltaic short-term power prediction method and system.The method includes: obtaining the historical multidimensional parameter time series data of distributed photovoltaic power station group;Multidimensional parameter time series data includes power data and meteorological data;According to the historical multidimensional parameter time series data obtained, train a first power prediction model by neural network;First power prediction model is used for the final prediction of photovoltaic short-term power;Train a lightweight second power prediction model;Second power prediction model is used for predicting photovoltaic short-term power;The historical multidimensional parameter time series data is input to first power prediction model and second power prediction model, whether the concept drift of first power prediction model appears by time window judgment.This application uses the prediction error of two different models to determine whether the concept drift of prediction model exists compared with prior art;By the determination of concept drift, the prediction model can be optimized more effectively.

Description

Technical Field

[0001] This invention relates to the field of power generation prediction for new energy sources in power systems, and particularly to a method and system for short-term power prediction of new energy photovoltaic power. Background Technology

[0002] Photovoltaic energy is being connected to the grid on a large scale. However, photovoltaic power generation is affected by the complexity and unpredictability of weather factors, and direct integration of photovoltaic energy into the power grid system would threaten the safe and stable operation of the distribution network. Therefore, it is necessary to accurately and effectively predict photovoltaic power generation.

[0003] For example, the Chinese patent application with publication number CN117035184A describes a method and apparatus for short-term power prediction of distributed photovoltaic power based on K-means++ and BILSTM. This patent application first identifies the photovoltaic output data according to the local climate characteristics of the photovoltaic power station, and then uses K-means++ clustering to divide the distributed photovoltaic power station groups in different seasons in the region, so that the weather in each power station group is consistent; then, BILSTM is used to predict the power of multiple photovoltaic power stations in the group at the same time, so as to improve the accuracy of short-term power prediction of large-scale distributed photovoltaic power station groups.

[0004] While the aforementioned technologies have achieved a certain degree of accuracy and effectiveness in predicting photovoltaic power generation, the conditional distribution in photovoltaic power prediction changes over time. This change in data flow leads to a concept drift problem, causing the mapping relationship of the original prediction model to become less accurate over time.

[0005] This application aims to establish a method and system for short-term power prediction of new energy photovoltaic power to solve the above-mentioned problems. Summary of the Invention

[0006] In order to achieve the above-mentioned objectives and other advantages of the present invention, a first objective of the present invention is to provide a method for short-term power prediction of new energy photovoltaics, comprising the following steps:

[0007] Acquire historical multi-dimensional parameter time-series data of distributed photovoltaic power station clusters; the multi-dimensional parameter time-series data includes power data and meteorological data.

[0008] Based on the acquired historical multidimensional parameter time-series data, a first power prediction model is trained through a neural network; the first power prediction model is used for the final prediction of short-term photovoltaic power.

[0009] Train a lightweight second power prediction model; the second power prediction model is used to predict short-term photovoltaic power.

[0010] Historical multidimensional parameter time series data are input into the first power prediction model and the second power prediction model. A sliding time window is used to determine whether the first power prediction model has concept drift.

[0011] In a preferred embodiment, obtaining historical multi-dimensional parameter time-series data of a distributed photovoltaic power station cluster specifically includes the following steps:

[0012] Acquire historical power data and historical meteorological data of distributed photovoltaic power station clusters;

[0013] Missing data completion and data normalization processes are performed on historical power data and historical meteorological data to obtain multidimensional time series data.

[0014] In a preferred embodiment, the second power prediction model is a lightweight prediction model trained on the first prediction network through knowledge distillation; the second power prediction model does not undergo data optimization during training.

[0015] In a preferred embodiment, determining whether concept drift occurs using a sliding time window specifically includes the following steps:

[0016] Set the length of the time window to T;

[0017] Extract the most recent T time points from historical multidimensional parameter time series data and input them into the time window;

[0018] The time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window.

[0019] In a preferred embodiment, the time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window. This specifically includes the following steps:

[0020] The time series data within the time window is input into the first power prediction model to output the first photovoltaic power prediction value at the next moment of the time window.

[0021] The time series data within the time window is input into the second power prediction model to output the second photovoltaic power prediction value at the next moment of the time window.

[0022] Acquire real-time multidimensional parameter data for the next moment in the time window; where the real-time power data is the actual value of photovoltaic power;

[0023] Based on the first photovoltaic power prediction value, the second photovoltaic power prediction value, and the actual photovoltaic power value, the prediction error ε1 of the first power prediction model and the prediction error ε2 of the second power prediction model are obtained.

[0024] The relationship between prediction errors ε1 and ε2 is used to determine whether concept drift occurs in the current time window. If ε2 < ε1, concept drift is determined to occur in the current time window; if ε2 ≥ ε1, concept drift is determined to occur in the current time window.

[0025] In a preferred embodiment, if no concept drift occurs within the current time window, the following steps are also included:

[0026] The time window is slid to the next time point; the data in the slid-up time window includes the time series data of the most recent T-1 time nodes extracted from the historical multidimensional parameter time series data and one real-time multidimensional parameter data.

[0027] The time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window.

[0028] In a preferred embodiment, if concept drift occurs within the current time window, the following steps are also included:

[0029] Obtain the time series data of the K time nodes following the current time window;

[0030] The time series data of the current time window T time nodes and the time series data of the time nodes after the time window K time nodes are input into the first power prediction model for training and updating the first power prediction model.

[0031] The time window is slid to the next time point; the data in the slid-up time window includes the time series data of the most recent T-1 time nodes extracted from the historical multidimensional parameter time series data and one real-time multidimensional parameter data.

[0032] The time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window.

[0033] A second objective of this invention is to provide a short-term power prediction system for new energy photovoltaic systems, comprising:

[0034] The photovoltaic data acquisition unit is used to acquire historical multi-dimensional parameter time-series data of distributed photovoltaic power station groups; the multi-dimensional parameter time-series data includes power data and meteorological data.

[0035] The first model training unit is used to train a first power prediction model through a neural network; the first power prediction model is used for the final prediction of short-term photovoltaic power.

[0036] The second model training unit is used to train a lightweight second power prediction model; the second power prediction model is used to predict short-term photovoltaic power.

[0037] The concept drift judgment unit is used to input historical multidimensional parameter time series data into the first power prediction model and the second power prediction model, and to judge whether the first power prediction model has concept drift by using a sliding time window.

[0038] A third objective of this invention is to provide a computer-readable storage medium having program instructions stored thereon, which, when executed, implement a method for short-term power prediction of new energy photovoltaic power.

[0039] A fourth objective of this invention is to provide an electronic device comprising: a processor and a memory, wherein the memory is used to store one or more programs; when the one or more programs are executed by the processor, a method for short-term power prediction of new energy photovoltaics is implemented.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] This invention relates to a method for short-term power prediction of new energy photovoltaics. Based on training a large and stable neural network model for photovoltaic power prediction, another lighter comparative model without parameter optimization is trained. The prediction errors of the two different models are used to determine whether there is concept drift in the prediction model. By determining concept drift, the prediction model can be optimized more effectively.

[0042] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail in the following embodiments and their accompanying drawings. Attached Figure Description

[0043] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0044] Figure 1 A flowchart of a short-term power prediction method for new energy photovoltaics provided in an embodiment of the present invention;

[0045] Figure 2 This is a flowchart provided by an embodiment of the present invention for determining whether a first power prediction model has experienced concept drift through a time window;

[0046] Figure 3 This is a schematic diagram of a new energy photovoltaic short-term power prediction system provided in an embodiment of the present invention;

[0047] Figure 4 A schematic diagram of an electronic device for short-term power prediction of new energy photovoltaic power provided in an embodiment of the present invention;

[0048] Figure 5 A schematic diagram of a computer storage medium for short-term power prediction of new energy photovoltaics provided in an embodiment of the present invention. Detailed Implementation

[0049] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0050] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this invention are used to distinguish different objects, rather than to limit a specific order.

[0051] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "module," "part," or "unit" may be used interchangeably.

[0052] For ease of description, some of the nouns or terms appearing in this invention will be explained in detail below.

[0053] Concept drift detector: Under ideal conditions, it is assumed that the distribution of a fixed data block P(x,y) does not exhibit concept drift. In the context of the data stream, at a specific time t, the distribution is defined as P. t (x,y), is represented as P in a specific time interval [t,k]. [t,k] (x,y); when the relationship between the input and the target is P t (x,y)≠P k The relationship between (x,y) or time intervals [t,k] and [i,j] is P. [t,k] (x,y)≠P [i,j] (x,y) indicates the existence of concept drift; the statistical properties of what the model attempts to predict change unpredictably over time. Photovoltaic power generation data are subject to complex environmental influences, and large amounts of data often arrive in the form of data streams. With ongoing evolution, concept drift may occur in the data streams.

[0054] Sliding time window: A sliding window is a concept based on two pointers, where a window is formed between the elements pointed to by the two pointers. There are two types of windows: fixed-size windows and dynamically changing-size windows. The sliding time window is an application of the sliding window to time series data.

[0055] Long Short-Term Memory Neural Network (LSTM): This is a special type of recurrent neural network. Unlike general feedforward neural networks, LSTM can analyze inputs using time series data.

[0056] Bidirectional Long Short-Term Memory (biLSTM): It is composed of a forward LSTM neural network and a backward LSTM neural network. A single-layer biLSTM is composed of two LSTMs, one of which processes the input sequence in the forward direction and the other processes the sequence in the backward direction. After processing, the outputs of the two LSTMs are concatenated.

[0057] Knowledge distillation is a commonly used method for model compression. Unlike pruning and quantization in model compression, knowledge distillation builds a lightweight small model and uses the supervision information of a larger, higher-performing model to train this small model, aiming to achieve better performance and accuracy.

[0058] According to one aspect of the present invention, a method for short-term power prediction of new energy photovoltaic is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0059] Figure 1 A flowchart of a short-term power prediction method for new energy photovoltaics provided in an embodiment of the present invention is shown. The method includes the following steps:

[0060] S102. Obtain historical multi-dimensional parameter time series data of distributed photovoltaic power station groups;

[0061] The aforementioned multidimensional parameter time series data includes power data and meteorological data;

[0062] Optionally, the multidimensional parameter time series data is obtained by acquiring historical power data and historical meteorological data of the distributed photovoltaic power station cluster, and performing missing data completion and data normalization processing on the historical power data and historical meteorological data to obtain multidimensional time series data; wherein, meteorological data includes, but is not limited to, data such as irradiance, cloud cover, humidity, temperature, and wind speed. In the specific implementation process, there is no limitation on the time series length of the historical multidimensional parameter time series data.

[0063] S104. Based on the acquired historical multidimensional parameter time series data, a first power prediction model is trained through a neural network; the first power prediction model is used for the final prediction of short-term photovoltaic power.

[0064] Optionally, the neural network used to train the first power prediction model can be a recurrent neural network, including but not limited to LSTM, biLSTM, etc., capable of analyzing and processing time series inputs. Before training the first power prediction model, it can be initialized to improve convergence speed and accuracy during training.

[0065] Furthermore, due to the strong fluctuations and randomness of photovoltaic (PV) output power, traditional PV power prediction processes do not perform noise reduction on historical data, and the large amount of noise may prevent accurate prediction and analysis of PV output. Therefore, before training with a recurrent neural network, machine learning can be used to filter key features of multi-dimensional parameter time-series data, reducing the PV time-series dimension and thus optimizing the training process. The machine learning models used include, but are not limited to, linear regression, support vector machines (SVM), nearest neighbor (KNN), logistic regression, decision trees, k-means, and random forests.

[0066] S106. Train a lightweight second power prediction model; the second power prediction model is used to predict short-term photovoltaic power.

[0067] In this embodiment, the second power prediction model primarily serves as a comparison model with the first power prediction model. Compared to the first power prediction model, the second power prediction model is a lighter-weight model and / or a prediction model trained with less training data. The first power prediction model is more stable than the second power prediction model and encapsulates more data; generally, the prediction error of the first power prediction model is better than that of the second power prediction model.

[0068] In this embodiment of the invention, the second power prediction model is a lightweight prediction model trained on the first power prediction network using knowledge distillation; the second power prediction model does not undergo data optimization during training. Compared to the first power prediction model, the second power prediction model trained through knowledge distillation encapsulates less data and performs worse in predicting overall data.

[0069] S108. Input historical multidimensional parameter time series data into the first power prediction model and the second power prediction model, and determine whether the first power prediction model has concept drift through the time window.

[0070] The main principle behind using a time window to determine whether the first power prediction model has experienced concept drift is to try to find a time window in which the second power prediction model outperforms the first power prediction network. Since the first power prediction network is more stable and encapsulates more data, it generally outperforms the second. However, if the second power prediction model outperforms the first power prediction network within a certain time window, it may indicate that concept drift has occurred.

[0071] In one exemplary embodiment, Figure 2 The flowchart for determining whether the first power prediction model has experienced concept drift through a time window includes the following steps in step S108:

[0072] S802, Set the length of the time window to T;

[0073] S804. Extract the time series data of the most recent T time nodes from the historical multidimensional parameter time series data and input it into the time window;

[0074] S806. Input the time series data within the time window into the first power prediction model to output the first photovoltaic power prediction value at the next moment of the time window; input the time series data within the time window into the second power prediction model to output the second photovoltaic power prediction value at the next moment of the time window;

[0075] S808. Obtain real-time multidimensional parameter data at the next moment of the time window; wherein the real-time power data is the actual value of photovoltaic power; based on the first photovoltaic power prediction value, the second photovoltaic power prediction value and the actual value of photovoltaic power, obtain the prediction error ε1 of the first power prediction model and the prediction error ε2 of the second power prediction model;

[0076] S810. Determine whether concept drift occurs in the current time window based on the relationship between the prediction error ε1 and the prediction error ε2. If ε2 < ε1, it is determined that concept drift occurs in the current time window, and step S212 is executed. If ε2 ≥ ε1, it is determined that concept drift does not occur in the current time window, and step S214 is executed.

[0077] S812. Obtain the time series data of the K time nodes after the current time window, and input the time series data of the T time nodes of the current time window and the time series data of the K time nodes after the time window into the first power prediction model for training to update the first power prediction model.

[0078] S814. Slide the time window to the next time point; wherein, the data in the slidable time window includes the time series data of the most recent T-1 time nodes extracted from the historical multidimensional parameter time series data and one real-time multidimensional parameter data and execute step S808.

[0079] According to another aspect of the present invention, a new energy photovoltaic short-term power prediction system 200 is also provided, such as... Figure 3 As shown, it includes:

[0080] The photovoltaic data acquisition unit 202 is used to acquire historical multi-dimensional parameter time-series data of distributed photovoltaic power station groups; the multi-dimensional parameter time-series data includes power data and meteorological data.

[0081] Optionally, the multidimensional parameter time series data is obtained by acquiring historical power data and historical meteorological data of the distributed photovoltaic power station cluster, and performing missing data completion and data normalization processing on the historical power data and historical meteorological data to obtain multidimensional time series data; wherein, meteorological data includes, but is not limited to, data such as irradiance, cloud cover, humidity, temperature, and wind speed. In the specific implementation process, there is no limitation on the time series length of the historical multidimensional parameter time series data.

[0082] The first model training unit 204 is used to train a first power prediction model through a neural network; the first power prediction model is used for the final prediction of short-term photovoltaic power.

[0083] Optionally, the neural network used to train the first power prediction model can be a recurrent neural network, including but not limited to LSTM, biLSTM, etc., capable of analyzing and processing time series inputs. Before training the first power prediction model, it can be initialized to improve convergence speed and accuracy during training.

[0084] Furthermore, due to the strong fluctuations and randomness of photovoltaic (PV) output power, traditional PV power prediction processes do not perform noise reduction on historical data, and the large amount of noise may prevent accurate prediction and analysis of PV output. Therefore, before training with a recurrent neural network, machine learning can be used to filter key features of multi-dimensional parameter time-series data, reducing the PV time-series dimension and thus optimizing the training process. The machine learning models used include, but are not limited to, linear regression, support vector machines (SVM), nearest neighbor (KNN), logistic regression, decision trees, k-means, and random forests.

[0085] The second model training unit 206 is used to train a lightweight second power prediction model; the second power prediction model is used to predict short-term photovoltaic power.

[0086] In this embodiment, the second power prediction model primarily serves as a comparison model with the first power prediction model. Compared to the first power prediction model, the second power prediction model is a lighter-weight model and / or a prediction model trained with less training data. The first power prediction model is more stable than the second power prediction model and encapsulates more data; generally, the prediction error of the first power prediction model is better than that of the second power prediction model.

[0087] In this embodiment of the invention, the second power prediction model is a lightweight prediction model trained on the first power prediction network using knowledge distillation; the second power prediction model does not undergo data optimization during training. Compared to the first power prediction model, the second power prediction model trained through knowledge distillation encapsulates less data and performs worse in predicting overall data.

[0088] The concept drift judgment unit 208 is used to input historical multidimensional parameter time series data into the first power prediction model and the second power prediction model, and to judge whether the first power prediction model has concept drift by using a sliding time window.

[0089] The concept drift judgment unit 208 determines whether the first power prediction model has experienced concept drift through a time window by attempting to find a time window in which the second power prediction model outperforms the first power prediction network. Since the first power prediction network is more stable and encapsulates more data than the second power prediction model, it generally outperforms the second power prediction model. However, if the second power prediction model outperforms the first power prediction network within a certain time window, it may indicate that concept drift has occurred.

[0090] According to another aspect of the present invention, an electronic device 300 is also provided, such as... Figure 4 As shown, it includes: a processor 301; and a memory 302 for storing processor-executable instructions; wherein the processor is configured to execute any of the above-described methods for short-term power prediction of new energy photovoltaic power. For a detailed description of the method, please refer to the corresponding description in the above method embodiments, which will not be repeated here.

[0091] According to another aspect of the present invention, a storage medium 400 is also provided, such as... Figure 5 As shown, the storage medium includes a stored program, wherein, during program execution, the device containing the computer-readable storage medium executes any of the above-described methods for short-term power prediction of new energy photovoltaic power. For a detailed description of the method, please refer to the corresponding description in the above-described method embodiments, which will not be repeated here.

[0092] The program instructions are stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, or external hard drive) or on a network, and include several computer program instructions to cause a computing device (such as a personal computer, server, or network device) to execute the method described above according to the embodiments of this application.

[0093] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention, and other modifications can be easily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

[0094] The apparatus, electronic device, and non-volatile computer storage medium and method provided in the embodiments of this specification are corresponding. Therefore, the apparatus, electronic device, and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, electronic device, and non-volatile computer storage medium will not be repeated here.

[0095] Those skilled in the art will also know that, besides implementing the controller in the form of purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller take the form of logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices included within it for implementing various functions can also be considered structures within that hardware component. Alternatively, the devices for implementing various functions can be considered as both software modules implementing the method and structures within a hardware component.

[0096] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0097] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0098] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects.

[0099] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0103] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0104] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0105] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside on local and remote computer storage media, including storage devices.

[0106] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0107] The above description is merely an embodiment of this specification and is not intended to limit the scope of one or more embodiments of this specification. Various modifications and variations can be made to one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of one or more embodiments of this specification.

Claims

1. A method for short-term power prediction of new energy photovoltaic power, characterized in that, include: Obtain historical multi-dimensional parameter time-series data of distributed photovoltaic power station groups; The multidimensional parameter time-series data includes power data and meteorological data; Based on the acquired historical multidimensional parameter time-series data, a first power prediction model is trained through a neural network; the first power prediction model is used for the final prediction of short-term photovoltaic power. Train a lightweight second power prediction model; the second power prediction model is used to predict short-term photovoltaic power. Historical multidimensional parameter time series data are input into the first power prediction model and the second power prediction model. The sliding time window is used to determine whether the first power prediction model has concept drift. Determining whether concept drift has occurred using a sliding time window includes the following steps: Set the length of the time window to T; Extract the most recent T time points from historical multidimensional parameter time series data and input them into the time window; The time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window. The time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window. Specifically, the steps include: The time series data within the time window is input into the first power prediction model to output the first photovoltaic power prediction value at the next moment of the time window; The time series data within the time window is input into the second power prediction model to output the second photovoltaic power prediction value at the next moment of the time window. Acquire real-time multidimensional parameter data at the next moment of the time window; wherein the real-time multidimensional parameter data is the actual value of photovoltaic power; The prediction error of the first power prediction model and the prediction error of the second power prediction model are obtained based on the first photovoltaic power prediction value, the second photovoltaic power prediction value and the actual photovoltaic power value. The determination of whether concept drift occurs in the current time window is based on the relationship between the prediction error and the magnitude of the prediction error; if the prediction error is positive, the current time window is determined to have concept drift, and if the prediction error is negative, the current time window is determined not to have concept drift.

2. The method for short-term power prediction of new energy photovoltaic power according to claim 1, characterized in that, Obtaining historical multidimensional parameter time-series data of a distributed photovoltaic power station group includes the following steps: Acquire historical power data and historical meteorological data of distributed photovoltaic power station clusters; The historical power data and historical meteorological data are subjected to missing data completion processing and data normalization processing to obtain multidimensional time series data.

3. The method for short-term power prediction of new energy photovoltaic power according to claim 1, characterized in that, The second power prediction model is a lightweight prediction model trained on the first prediction network through knowledge distillation; the second power prediction model does not perform data optimization during training.

4. The method for short-term power prediction of new energy photovoltaic power according to claim 1, characterized in that, If no concept drift occurs in the current time window, the following steps are also included: The time window is slid to the next time point; wherein, the data in the slidable time window includes time series data of the most recent T-1 time nodes extracted from the historical multidimensional parameter time series data and one real-time multidimensional parameter data; The time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window.

5. The method for short-term power prediction of new energy photovoltaic power according to claim 1, characterized in that, If concept drift occurs in the current time window, the following steps are also included: Obtain the time series data of the K time nodes following the current time window; The time series data of the current time window T time nodes and the time series data of the time nodes after the time window K time nodes are input into the first power prediction model for training and updating the first power prediction model. The time window is then slid to the next time point; wherein the data within the slidable time window includes time-series data extracted from the most recent T-1 time nodes in the historical multidimensional parameter time-series data and one real-time multidimensional parameter data; The time series data within the time window is input into the first power prediction model and the second power prediction model to determine whether concept drift occurs in the current time window.

6. A short-term power prediction system for new energy photovoltaics, characterized in that, The system for performing the method as described in any one of claims 1 to 5 includes: A photovoltaic data acquisition unit is used to acquire historical multi-dimensional parameter time-series data of distributed photovoltaic power station groups; the multi-dimensional parameter time-series data includes power data and meteorological data. The first model training unit is used to train a first power prediction model through a neural network; the first power prediction model is used for the final prediction of short-term photovoltaic power. The second model training unit is used to train a lightweight second power prediction model; the second power prediction model is used to predict short-term photovoltaic power. The concept drift judgment unit is used to input historical multidimensional parameter time series data into the first power prediction model and the second power prediction model, and to judge whether the first power prediction model has concept drift by using a sliding time window.

7. An electronic device comprising a memory and a processor; wherein, The memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the steps of the method according to any one of claims 1-5.

8. A storage medium having computer instructions stored thereon; wherein, When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1-5.

Images