A distributed photovoltaic power prediction method and device based on transfer learning

Abstract

The present application relates to the technical field of power system new energy power prediction, and more particularly to a distributed photovoltaic power prediction method and device based on transfer learning, comprising: acquiring meteorological data of a photovoltaic power station, performing interpolation calculation on the meteorological data of the photovoltaic power station to obtain interpolation data, performing classification processing on the interpolation data, acquiring a source domain and a target domain, constructing a source domain power prediction model, acquiring a preliminary prediction value, acquiring an error sequence based on the preliminary prediction value, constructing an error prediction model based on the error sequence, acquiring final prediction data of the source domain, performing double transfer learning on the source domain power prediction model and the error prediction model based on the final prediction data of the source domain, fine-tuning model parameters of the target domain, and acquiring a power prediction result of the photovoltaic power station. Transfer learning is performed on multiple target domain photovoltaic power stations within a certain range, which improves photovoltaic power prediction accuracy and effectively makes up for the problem of insufficient power prediction accuracy caused by data scarcity.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy power prediction for electric power systems, and in particular to a distributed photovoltaic power prediction method and device based on transfer learning. Background Art

[0002] Photovoltaic power forecasting involves predicting the active power of a photovoltaic power station for a specific period in the future based on meteorological conditions, statistical laws, and other technologies and methods. Due to the diurnal cycle of sunlight, photovoltaic power stations can only generate electricity during the day, making them a typical intermittent power source. Photovoltaic power is affected by meteorological and environmental conditions and exhibits significant volatility and randomness. These characteristics can negatively impact the grid when large-scale photovoltaic power generation is connected to the grid. Accurate and timely prediction of photovoltaic power will be of great significance to grid scheduling and photovoltaic power station operation.

[0003] Transfer learning is a machine learning concept that allows existing models to be fine-tuned for application in new domains or for new functions. In transfer learning, the data domain is divided into a source domain and a target domain. Typically, a model is pre-trained in the source domain, which has abundant data, and then fine-tuned in the target domain, which has less data. This allows the model to fully utilize the source domain data and improve its performance on the target data.

[0004] Currently, distributed photovoltaic systems often lack specialized meteorological monitoring equipment, and a large number of newly built photovoltaic power stations each year lack sufficient historical data, making it difficult to establish high-precision prediction models.

[0005] In summary, how to design a photovoltaic power prediction method with high accuracy and high efficiency is an urgent problem that needs to be solved. Summary of the Invention

[0006] The purpose of the present invention is to provide a distributed photovoltaic power prediction method based on transfer learning to solve the problems of low accuracy and poor efficiency of existing distributed photovoltaic prediction methods.

[0007] To solve the above technical problems, the present invention provides a distributed photovoltaic power prediction method based on transfer learning, comprising:

[0008] Acquiring meteorological data of the photovoltaic power station, and performing interpolation calculation on the meteorological data of the photovoltaic power station to obtain interpolation data;

[0009] Classify the interpolation data to obtain a source domain and a target domain;

[0010] Build a source domain power prediction model and obtain preliminary prediction values;

[0011] Based on the preliminary prediction value, obtaining an error sequence;

[0012] Building an error prediction model based on the error sequence to obtain final prediction data of the source domain;

[0013] Based on the source domain final prediction data, the source domain power prediction model and the error prediction model are double-transferred learned, the model parameters of the target domain are fine-tuned, and the power prediction result of the photovoltaic power station is obtained.

[0014] Preferably, the meteorological data of the photovoltaic power station is obtained, and the interpolation calculation is performed on the meteorological data of the photovoltaic power station to obtain the interpolation data.

[0015] Based on geographic information and coarse-grained meteorological data, the meteorological resource data in a wide range is grid processed, and based on the Kriging interpolation method, the meteorological data of the latitude and longitude position of each photovoltaic power station is geographically interpolated to obtain the interpolation data.

[0016] Preferably, the Kriging interpolation method calculation formula is:

[0017]

[0018] Wherein, z(S0) is the Kriging interpolation result at S0, n is the number of data samples, z i is a sample point value, and λ i is a sample point weight coefficient.

[0019] Preferably, the interpolation data is classified and processed to obtain the source domain and the target domain.

[0020] Based on the interpolation data, the photovoltaic power stations with the same meteorological characteristics are clustered by using a self-organizing mapping network, all photovoltaic power stations are classified, the correlation between each photovoltaic power station and the remaining photovoltaic power stations is calculated in each class, and in the case of meeting the preset data requirements, one or several power stations with the highest correlation average are selected as the source domain, and the remaining photovoltaic power stations are selected as the target domain.

[0021] Preferably, the correlation calculation formula is:

[0022]

[0023] Wherein, r is the correlation coefficient of the meteorological resource data between different photovoltaic power stations, n is the number of data samples, X i is the X number of photovoltaic power station meteorological data samples, Y i is the Y number of photovoltaic power station meteorological data samples, is the average value of the X number of photovoltaic power station meteorological data samples, is the average value of the Y number of photovoltaic power station meteorological data samples.

[0024] Preferably, the error prediction model is constructed based on the error sequence, and the source domain final prediction data is obtained.

[0025] An error prediction model is established for the error sequence using a long short-term memory network algorithm, and final prediction data of the source domain is obtained based on the preliminary prediction value and the error prediction value.

[0026] Preferably, performing dual transfer learning on the source domain power prediction model and the error prediction model based on the source domain final prediction data, fine-tuning the model parameters of the target domain, and obtaining the photovoltaic power station power prediction result includes:

[0027] Based on the power prediction model features of the source domain photovoltaic power station under similar meteorological resources, a rough image and parameter fine-tuning of the target domain photovoltaic power station prediction model are performed, and the source domain features are converted to target domain features to obtain a trained prediction model;

[0028] The trained prediction model is used to predict the photovoltaic power station data to obtain a photovoltaic power station power prediction result.

[0029] The present invention also provides a distributed photovoltaic power prediction device based on transfer learning, comprising:

[0030] A data acquisition module is used to acquire meteorological data of the photovoltaic power station and perform interpolation calculation on the meteorological data of the photovoltaic power station to obtain interpolation data;

[0031] A classification processing module, which performs classification processing on the interpolation data to obtain a source domain and a target domain;

[0032] The preliminary prediction module builds a source domain power prediction model and obtains preliminary prediction values;

[0033] An error sequence acquisition module, which acquires an error sequence based on the preliminary prediction value;

[0034] A source domain prediction module constructs an error prediction model based on the error sequence to obtain final source domain prediction data;

[0035] The power prediction module performs dual transfer learning on the source domain power prediction model and the error prediction model based on the final prediction data of the source domain, fine-tunes the model parameters of the target domain, and obtains the power prediction result of the photovoltaic power station.

[0036] The present invention also provides a distributed photovoltaic power prediction device based on transfer learning, comprising:

[0037] Memory for storing computer programs;

[0038] A processor is configured to implement the steps of the above-mentioned method for distributed photovoltaic power prediction based on transfer learning when executing the computer program.

[0039] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned distributed photovoltaic power prediction method based on transfer learning are implemented.

[0040] The present invention provides a distributed photovoltaic power prediction method based on transfer learning. The method uses the Kriging interpolation method to associate the spatial position of a sample point with the value of the sample point, and uses the distance between the sample point and the interpolation point to measure the weight value of each sample point. The value of the interpolation point is calculated in a weighted average manner. The existing model is then fine-tuned based on transfer learning. The data domain is divided into a source domain and a target domain. The pre-trained model is fine-tuned in the target domain with a smaller data volume. The source domain data is fully utilized to improve the model's performance on the target data. This realizes the rapid and large-scale migration of distributed photovoltaic power station power prediction models, quickly learns the common features of prediction models in similar scenarios, improves the accuracy of photovoltaic power prediction, and effectively compensates for the problem of insufficient power prediction accuracy caused by data scarcity. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0042] Figure 1 This is a flowchart of a first specific embodiment of a distributed photovoltaic power prediction method based on transfer learning provided by the present invention;

[0043] Figure 2 This is a structural block diagram of a distributed photovoltaic power prediction device based on transfer learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The core of the present invention is to provide a distributed photovoltaic power prediction method and device based on transfer learning. Transfer learning is used to extract power prediction model features under similar meteorological resources from source-domain photovoltaic power stations with higher data quality, and transfer learning is performed on multiple target-domain photovoltaic power stations within a certain range, thereby improving the photovoltaic power prediction accuracy and effectively compensating for the problem of insufficient power prediction accuracy caused by data scarcity.

[0045] In order to enable those skilled in the art to better understand the present invention, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0046] Please refer to Figure 1 , Figure 1This is a flowchart of a first specific embodiment of a distributed photovoltaic power prediction method based on transfer learning provided by the present invention; the specific operation steps are as follows:

[0047] Step S101: Acquire meteorological data of a photovoltaic power station, and perform interpolation calculation on the meteorological data of the photovoltaic power station to obtain interpolation data;

[0048] Based on geographic information and coarse-grained meteorological data, the meteorological resource data in a wide area is gridded. Based on the Kriging interpolation method, the meteorological data at the latitude and longitude of each photovoltaic power station is geographically interpolated to obtain interpolated data.

[0049] The Kriging interpolation calculation formula is:

[0050]

[0051] Among them, z(S0) is the Kriging interpolation result at S0, n is the number of data samples, z i is the sample point value, λ i is the sample point weight coefficient.

[0052] Step S102: performing classification processing on the interpolation data to obtain a source domain and a target domain;

[0053] Based on the interpolation data, a self-organizing map network clustering is performed on photovoltaic power stations with the same meteorological characteristics, all photovoltaic power stations are classified, and the correlation between each photovoltaic power station and the remaining photovoltaic power stations in each category is calculated. When preset data requirements are met, one or several photovoltaic power stations with the highest correlation mean are selected as source domains, and the remaining photovoltaic power stations are selected as target domains;

[0054] The correlation calculation formula is:

[0055]

[0056] Among them, r is the correlation coefficient of meteorological resource data between different photovoltaic power stations, n is the number of data samples, X i is the meteorological data sample of power station X, Y i is the meteorological data sample of Power Station Y, is the average value of the meteorological data sample of power station X, is the average value of meteorological data samples of Power Station Y.

[0057] Step S103: constructing a source domain power prediction model and obtaining a preliminary prediction value;

[0058] Step S104: obtaining an error sequence based on the preliminary prediction value;

[0059] Step S105: constructing an error prediction model based on the error sequence to obtain final prediction data of the source domain;

[0060] An error prediction model is established for the error sequence using a long short-term memory network algorithm, and final prediction data of the source domain is obtained based on the preliminary prediction value and the error prediction value.

[0061] Step S106: performing dual transfer learning on the source domain power prediction model and the error prediction model based on the final prediction data of the source domain, fine-tuning the model parameters of the target domain, and obtaining the power prediction result of the photovoltaic power station.

[0062] Based on the power prediction model features of the source domain photovoltaic power station under similar meteorological resources, a rough image and parameter fine-tuning of the target domain photovoltaic power station prediction model are performed, and the source domain features are converted to target domain features to obtain a trained prediction model;

[0063] The trained prediction model is used to predict the photovoltaic power station data to obtain a photovoltaic power station power prediction result.

[0064] This embodiment provides a distributed photovoltaic power prediction method based on transfer learning. This method extracts power prediction model features from source-domain photovoltaic power stations with high-quality data under similar meteorological conditions. It then roughly "profiles" the target-domain photovoltaic power station prediction model and fine-tunes its parameters, thereby achieving feature transfer from the source domain to the target domain. After selecting one or more photovoltaic power stations as the source domain, transfer learning is performed on multiple target-domain photovoltaic power stations within a certain range, enabling rapid and large-scale migration of distributed photovoltaic power station power prediction models. For photovoltaic power stations with relatively recent construction and limited data, using transfer learning to build power prediction models can quickly learn the common features of prediction models under similar scenarios, effectively compensating for the lack of power prediction accuracy caused by data scarcity.

[0065] Based on the above embodiment, this embodiment describes the distributed photovoltaic power prediction method based on transfer learning, as follows:

[0066] Based on geographic information and coarse-grained meteorological data, the meteorological resource data in a wide area is gridded and processed. Combined with the Kriging interpolation method, the meteorological data at the latitude and longitude of each photovoltaic power station is geographically interpolated.

[0067] The Kriging interpolation method is based on the first law of geography. It associates the spatial position of the sample point with the value of the sample point, and uses the distance between the sample point and the interpolation point to calculate the weight value of each sample point, and calculates the value of the interpolation point in a weighted average manner.

[0068] Based on the interpolation results, the PV power stations with the same meteorological characteristics are clustered using the Self-Organizing Maps (SOM) network, and all PV power stations are divided into n categories.

[0069] The SOM algorithm is an unsupervised learning algorithm for clustering and high-dimensional visualization. It is an artificial neural network developed by simulating signal processing in the human brain. The SOM algorithm is a simple neural network consisting of only an input layer and a competitive layer. Neurons in the competitive layer promote learning with adjacent neurons until they have similar weight vectors. Therefore, sample data with similar input features can be mapped to adjacent neurons, achieving dimensionality reduction from the input space (n-dimensional) to the output plane (2-dimensional).

[0070] In each category, the correlation between each PV power station and the rest is calculated. When the data requirements are met, one or several PV stations with the highest correlation mean are selected as the source domain, and the rest of the PV power stations are selected as the target domain. After obtaining the SOM clustering results, the correlation analysis is performed on the meteorological resource data of the PV power stations in each category.

[0071] The source domain data is segmented and processed, and a source domain power prediction model is established using the Long Short Term Memory Network (LSTM) algorithm to obtain preliminary prediction results.

[0072] Subtract the initial prediction result from the original sequence to obtain the error sequence;

[0073] The error prediction model is established for the error sequence using the Long Short Term Memory Network (LSTM) algorithm. The final prediction result of the source domain is the sum of the preliminary prediction value and the error prediction value.

[0074] First, the sample data is divided into three segments a, b, and c. The three segments are represented by x a (t), x b (t), x c (t), using the data of segment a to establish a preliminary power prediction model based on the Long Short Term Memory Network (LSTM) algorithm, predict segment b and obtain the power prediction value x' of segment b b (t);

[0075] Secondly, the original sample data of segment b is compared with the power prediction value x b (t) Make the difference to get the error sequence e of power prediction b (t);

[0076] Using the error sequence to establish an error prediction model based on the Long Short Term Memory Network (LSTM) algorithm;

[0077] For the test set c segment data, the final prediction result is equal to the power prediction value x' obtained by the preliminary power prediction model c (t) and the error prediction value e obtained by the error prediction model c (t), since the error prediction model is also a neural network structure, it can also perform transfer learning from the source domain to the target domain, thereby establishing a dual transfer model of the power prediction model and the error correction model.

[0078] Dual transfer learning is performed on the power prediction model and error prediction model trained in the source domain. Based on the input target domain data, the model parameters of the target domain are fine-tuned. The above steps are repeated repeatedly to finally obtain the prediction results of each photovoltaic power station in the target domain.

[0079] An embodiment of the present invention provides a distributed photovoltaic power prediction method based on transfer learning, which utilizes neurons in the competition layer of the SOM algorithm and adjacent neurons to promote learning with each other until they have similar weight vectors, maps sample data with similar input features to adjacent neurons, and realizes dimensionality reduction mapping from the input space to the output plane, solves the problems of unclear input features, interlaced and mixed feature parameters, and knowledge discovery, data classification, and fault identification of nonlinear distribution, establishes a dual migration model of power prediction model and error correction model based on the transfer learning method, and realizes rapid and large-scale migration of distributed photovoltaic power station power prediction models. For some photovoltaic power stations with a short construction time and less data, the power prediction model is built using transfer learning, which can quickly learn the common features of the prediction model in similar scenarios, effectively make up for the problem of insufficient power prediction accuracy caused by lack of data, and realize efficient and accurate photovoltaic power prediction.

[0080] Please refer to Figure 2 , Figure 2 This is a block diagram of a distributed photovoltaic power prediction device based on transfer learning provided by an embodiment of the present invention; the specific device may include:

[0081] The data acquisition module 100 acquires meteorological data of the photovoltaic power station and performs interpolation calculation on the meteorological data of the photovoltaic power station to obtain interpolation data;

[0082] A classification processing module 200 performs classification processing on the interpolation data to obtain a source domain and a target domain;

[0083] A preliminary prediction module 300 constructs a source domain power prediction model and obtains a preliminary prediction value;

[0084] An error sequence acquisition module 400 acquires an error sequence based on the preliminary prediction value;

[0085] A source domain prediction module 500 constructs an error prediction model based on the error sequence to obtain final source domain prediction data;

[0086] The power prediction module 600 performs dual transfer learning on the source domain power prediction model and the error prediction model based on the source domain final prediction data, fine-tunes the model parameters of the target domain, and obtains the photovoltaic power station power prediction result.

[0087] A distributed photovoltaic power prediction device based on transfer learning in this embodiment is used to implement the aforementioned distributed photovoltaic power prediction method based on transfer learning. Therefore, the specific implementation method of a distributed photovoltaic power prediction device based on transfer learning can be seen in the embodiment part of a distributed photovoltaic power prediction method based on transfer learning in the previous text. For example, the data acquisition module 100, the classification processing module 200, the preliminary prediction module 300, the error sequence acquisition module 400, the source domain prediction module 500, and the power prediction module 600 are respectively used to implement steps S101, S102, S103, S104, S105, and S106 in the aforementioned distributed photovoltaic power prediction method based on transfer learning. Therefore, its specific implementation method can refer to the description of the corresponding embodiments of each part and will not be repeated here.

[0088] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0089] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0090] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0091] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0092] The above is a detailed introduction to the distributed photovoltaic power prediction method and device based on transfer learning provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0093] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0094] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0095] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

Claims

1. A distributed photovoltaic power prediction method based on transfer learning, characterized in that: include: Based on geographic information and coarse-grained meteorological data, the meteorological resource data in a wide area is gridded. Based on the Kriging interpolation method, the meteorological data at the latitude and longitude of each photovoltaic power station is geographically interpolated to obtain interpolated data. Based on the interpolation data, a self-organizing map network clustering is performed on photovoltaic power stations with the same meteorological characteristics, all photovoltaic power stations are classified, and the correlation between each photovoltaic power station and the remaining photovoltaic power stations in each category is calculated. When preset data requirements are met, one or several photovoltaic power stations with the highest correlation mean are selected as source domains, and the remaining photovoltaic power stations are selected as target domains; Build a source domain power prediction model and obtain preliminary prediction values; Based on the preliminary prediction value, obtaining an error sequence; Establishing an error prediction model for the error sequence using a long short-term memory network algorithm, and obtaining final prediction data of the source domain based on the preliminary prediction value and the error prediction value; Based on the final prediction data of the source domain, dual transfer learning is performed on the source domain power prediction model and the error prediction model, and the model parameters of the target domain are fine-tuned to obtain the power prediction result of the photovoltaic power station.

2. The distributed photovoltaic power prediction method based on transfer learning according to claim 1, characterized in that: The Kriging interpolation calculation formula is: in, for The Kriging interpolation result at , is the number of data samples, is the sample point value, is the sample point weight coefficient.

3. The distributed photovoltaic power prediction method based on transfer learning according to claim 1, characterized in that: The correlation calculation formula is: in, is the correlation coefficient of meteorological resource data between different photovoltaic power stations, is the number of data samples, is the meteorological data sample of power station X, is the meteorological data sample of Power Station Y, is the average value of the meteorological data sample of power station X, is the average value of meteorological data samples of Power Station Y.

4. The distributed photovoltaic power prediction method based on transfer learning according to claim 1, characterized in that: The performing dual transfer learning on the source domain power prediction model and the error prediction model based on the source domain final prediction data, fine-tuning the model parameters of the target domain, and obtaining the photovoltaic power station power prediction result includes: Based on the power prediction model features of the source-domain photovoltaic power station under similar meteorological resources, a rough image and parameter fine-tuning of the target-domain photovoltaic power station prediction model are performed, and the source-domain features are converted to the target-domain features to obtain the trained prediction model. The trained prediction model is used to predict the photovoltaic power station data to obtain a photovoltaic power station power prediction result.

5. A distributed photovoltaic power prediction device based on transfer learning, characterized in that: include: The data acquisition module performs grid processing on meteorological resource data over a wide area based on geographic information and coarse-grained meteorological data. Based on the Kriging interpolation method, it performs geographic interpolation calculations on the meteorological data at the latitude and longitude of each photovoltaic power station to obtain interpolated data. a classification processing module, which performs a self-organizing map network clustering on photovoltaic power stations with the same meteorological characteristics based on the interpolation data, classifies all photovoltaic power stations, calculates the correlation between each photovoltaic power station and the rest of the photovoltaic power stations in each class, and selects one or several power stations with the highest correlation mean as the source domain, and the rest of the photovoltaic power stations as the target domain, provided that the preset data requirements are met; The preliminary prediction module builds a source domain power prediction model and obtains preliminary prediction values; An error sequence acquisition module, which acquires an error sequence based on the preliminary prediction value; A source domain prediction module, which uses a long short-term memory network algorithm to establish an error prediction model for the error sequence, and obtains final source domain prediction data based on the preliminary prediction value and the error prediction value; The power prediction module performs dual transfer learning on the source domain power prediction model and the error prediction model based on the final prediction data of the source domain, fine-tunes the model parameters of the target domain, and obtains the power prediction result of the photovoltaic power station.

6. A distributed photovoltaic power prediction device based on transfer learning, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of a distributed photovoltaic power prediction method based on transfer learning as described in any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the distributed photovoltaic power prediction method based on transfer learning as described in any one of claims 1 to 4 are implemented.

Images