Distributed photovoltaic power prediction method and device, electronic equipment and storage medium

Abstract

The application relates to a distributed photovoltaic power prediction method and device, electronic equipment and a storage medium, wherein the method comprises the following steps: generating a corresponding virtual power station around an actual photovoltaic power station to be predicted, and acquiring the longitude and latitude information of the virtual photovoltaic power station; acquiring the historical weather data of the environment around the virtual photovoltaic power station based on the longitude and latitude information; generating power data of the virtual photovoltaic power station by using the historical weather data, pre-training a prediction large model by using the power data, fine-tuning the prediction large model based on the power data of the actual photovoltaic power station, and obtaining the power prediction result of the actual photovoltaic power station based on the fine-tuned prediction large model. Thus, the problem of high-precision power prediction of distributed photovoltaic power is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing methods for prediction purposes, and in particular to a distributed photovoltaic power prediction method and device, electronic equipment and storage medium. BACKGROUND

[0002] Traditional power generation methods are highly polluting and unsustainable. In order to meet the demand for electricity and ensure sustainable development, renewable energy needs to be used to replace traditional fossil fuels on the power generation side, in order to reduce carbon dioxide emissions per unit of energy and reduce energy consumption per unit of GDP. In recent years, with the rapid development of new energy power generation technology, new energy power generation equipment represented by photovoltaics has begun to be connected to the power grid in large quantities, changing the operation mode of traditional power grids and posing new challenges to the safe operation and management of power grids.

[0003] Photovoltaics is a device that converts light energy into electrical energy using the photovoltaic effect of semiconductors, mainly composed of solar panels, controllers and inverters, and has the characteristics of no pollution and small land occupation. However, due to factors such as shading and weather, photovoltaic power generation is intermittent and volatile. In terms of safety, it will cause fluctuations in the frequency and voltage of the power grid, posing challenges to power grid control, and also causing light to be abandoned; in terms of economy, it will affect the determination of electricity prices and the investment of operators in photovoltaic construction. At present, distributed photovoltaics uses the mode of self-generation and self-use, and surplus power on the grid. Accurate distributed photovoltaic power prediction can support the safe operation of the power grid and provide a reference for the formulation of dispatching plans.

[0004] The photovoltaic power prediction method in the related art can be mainly divided into a mechanism-based method and a data-driven method. The input of the former is mainly meteorological data, and then the power is calculated according to the photovoltaic mechanism model. This type of method requires less historical data, but requires photovoltaic component parameters; the latter uses a large amount of meteorological and historical power generation data to train machine learning models such as regression and neural networks.

[0005] After a large number of distributed photovoltaics are connected to the grid, they seriously affect the safe operation of the distribution network, and high-precision prediction of their output power is urgently needed to guide the operation and dispatching of the distribution network. However, if environmental monitoring equipment and photovoltaic power prediction systems are provided separately for each power station, the cost is high, and there is a problem of a lack of historical data for newly built power stations; if photovoltaic power prediction systems are provided centrally for each region, the differences in local characteristics between different distributed power stations will seriously reduce the prediction accuracy. SUMMARY

[0006] The present application provides a distributed photovoltaic power prediction method, device, electronic equipment and storage medium to solve the problem of insufficient accuracy of distributed photovoltaic power prediction methods in the related art.

[0007] The first aspect embodiment of the present application provides a distributed photovoltaic power prediction method, comprising the following steps: generating a corresponding virtual power station around an actual photovoltaic power station to be predicted, and obtaining the longitude and latitude information of a virtual photovoltaic of the virtual power station; obtaining historical weather data of a surrounding environment of the virtual photovoltaic based on the longitude and latitude information; generating power data of the virtual photovoltaic by using the historical weather data, pre-training a large prediction model by using the power data, fine-tuning the large prediction model based on power data of the actual photovoltaic power station, and obtaining a power prediction result of the actual photovoltaic power station based on the fine-tuned large prediction model.

[0008] Optionally, in an embodiment of the present application, the fine-tuning expression of the large prediction model is:

[0009]

[0010] wherein L(w) is a target function, w is a weight of the large prediction model, is a power true value, is an input power of the large prediction model, is the large prediction model, is a power prediction value.

[0011] Optionally, in an embodiment of the present application, before the large prediction model is pre-trained by using the power data, the method further comprises: introducing time coding in an embedding layer of a Transformer model, and deleting an output layer of the Transformer model to obtain a modified Transformer model; training the modified Transformer model by using the historical power sequence to obtain the large prediction model.

[0012] Optionally, in an embodiment of the present application, the historical weather data comprises at least one of historical horizontal total radiation, historical temperature and historical rainfall.

[0013] The second aspect embodiment of the present application provides a distributed photovoltaic power prediction device, comprising: a first obtaining module configured to generate a corresponding virtual power station around an actual photovoltaic power station to be predicted, and obtain the longitude and latitude information of a virtual photovoltaic of the virtual power station; a second obtaining module configured to obtain historical weather data of a surrounding environment of the virtual photovoltaic based on the longitude and latitude information; and a prediction module configured to generate power data of the virtual photovoltaic by using the historical weather data, pre-train a large prediction model by using the power data, fine-tune the large prediction model based on power data of the actual photovoltaic power station, and obtain a power prediction result of the actual photovoltaic power station based on the fine-tuned large prediction model.

[0014] Optionally, in an embodiment of the present application, the fine-tuning expression of the large prediction model is:

[0015]

[0016] wherein L(w) is an objective function, w is a weight of a prediction large model, is a power real value, is an input power of the prediction large model, is the prediction large model, is a power prediction value.

[0017] Optionally, in an embodiment of the present application, further comprising: a control module, configured to introduce time coding in an embedding layer of a Transformer model, and delete an output layer of the Transformer model to obtain a Transformer modified model; and a training module, configured to train the Transformer modified model by using the historical power sequence to obtain the prediction large model.

[0018] Optionally, in an embodiment of the present application, the historical weather data comprises at least one of historical horizontal total radiation, historical temperature, and historical rainfall.

[0019] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the distributed photovoltaic power prediction method as described in the above embodiments.

[0020] The fourth aspect of the present application provides a computer readable storage medium, which stores computer instructions for making the computer execute the distributed photovoltaic power prediction method as described in the above embodiments.

[0021] The fifth aspect of the present application provides a computer program product, comprising a computer program, which, when executed, is used to implement the distributed photovoltaic power prediction method as described above.

[0022] The embodiments of the present application can generate a corresponding virtual power station around a real photovoltaic power station to be predicted, and obtain historical weather data of a virtual photovoltaic surrounding environment of the virtual power station, thereby generating power data of the virtual photovoltaic, and pre-training a prediction large model by using the power data, and fine-tuning the model by using power data of the real photovoltaic power station, thereby performing power prediction of the real photovoltaic power station, that is, generating a virtual power station around a real photovoltaic power station to be predicted by using a mechanism model, pre-training a prediction large model by using generated data, and then performing transfer learning on the real photovoltaic power station to fine-tune a small amount of parameters, thereby effectively coping with data scarcity while supporting lightweight deployment. Thus, the problem of insufficient accuracy of the distributed photovoltaic power prediction method in the related art is solved.

[0023] Additional aspects and advantages of the present application will be partially apparent and partially described in the following description. BRIEF DESCRIPTION OF DRAWINGS

[0024] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the appended drawings.

[0025] Figure 1 A flow chart of a distributed photovoltaic power prediction method according to an embodiment of the present application;

[0026] Figure 2 A principle schematic diagram of a distributed photovoltaic power prediction method according to an embodiment of the present application;

[0027] Figure 3 A structure schematic diagram of a distributed photovoltaic power prediction device according to an embodiment of the present application;

[0028] Figure 4 A structure schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which the same or similar elements or elements having the same or similar functions are denoted by the same or similar reference numerals throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0030] The distributed photovoltaic power prediction method, device, electronic device and storage medium of the embodiments of the present application are described below with reference to the accompanying drawings. In view of the insufficient accuracy of the related art distributed photovoltaic power prediction method mentioned in the above background art, the present application provides a distributed photovoltaic power prediction method. In the method, a corresponding virtual power station can be generated around the actual photovoltaic power station to be predicted, and historical weather data of the virtual photovoltaic power station can be obtained, so as to generate power data of the virtual photovoltaic power station, and a large prediction model can be pre-trained using the power data, and the model can be fine-tuned through the power data of the actual photovoltaic power station, so as to perform power prediction of the actual photovoltaic power station. That is, a mechanism model is used to generate a virtual power station around the real photovoltaic power station to be predicted, a large prediction model is pre-trained using the generated data, and then transfer learning is performed on the real photovoltaic power station, and the parameter quantity is small. In this way, the problem of insufficient accuracy of the related art distributed photovoltaic power prediction method is solved.

[0031] Specifically, Figure 1A flowchart of a distributed photovoltaic power prediction method provided by an embodiment of the present application.

[0032] As shown in Figure 1 The distributed photovoltaic power prediction method is applied to a model construction stage, and the method comprises the following steps:

[0033] In step S101, corresponding virtual power stations are generated around the actual photovoltaic power station to be predicted, and the latitude and longitude information of the virtual photovoltaic power stations is obtained.

[0034] In actual execution, the embodiment of the present application can generate virtual power stations around the real photovoltaic power station to be predicted by using a mechanism model, and determine the latitude and longitude of the virtual photovoltaic power stations.

[0035] Since the distributed photovoltaic layout is flexible, it is generally distributed on apartment roofs, factory roofs, industrial parks and roadsides, etc. Therefore, the determination of the photovoltaic construction site refers to the actual administrative division. For example, the embodiment of the present application can take the location of the administrative center as the center point, and classify the grid-shaped sites with a spatial resolution of 5 kilometers by the minimum distance, wherein 5 kilometers is the minimum resolution of meteorological satellites. The classified points are randomly sampled to the required number. By comparing the map, some sites that cannot be built photovoltaic, such as points located on the road, are translated by 0.05 degrees east / west / south / north, not more than two units.

[0036] In step S102, the historical weather data of the surrounding environment of the virtual photovoltaic power station is obtained based on the latitude and longitude information.

[0037] It can be understood that the photovoltaic power generation power is mainly affected by the geographical location, installation angle, self parameter and weather, etc. For a specific photovoltaic power station, its power generation power mainly depends on meteorological factors, and the irradiance and temperature have greater influence. According to the photovoltaic power generation mechanism, there is a direct relationship between the power generation power and the horizontal total radiation, and the power generation power of a photovoltaic power station is similar to the ground irradiance fluctuation law of the location. The ground irradiance can be regarded as the clear sky radiation minus the cloud attenuation, wherein the clear sky radiation is affected by the solar elevation angle and the length of sunshine, and is mainly determined by the geographical location, and the mechanism is clear, while the cloud attenuation is mainly affected by the weather, especially the cloud, and the mechanism is complex.

[0038] Further, the embodiment of the present application can obtain the historical weather forecast according to the determined latitude and longitude of the virtual photovoltaic power station, and process the historical weather forecast into historical weather data with a resolution of 15 minutes after space-time interpolation, i.e. meteorological data.

[0039] Optionally, in an embodiment of the present application, the historical weather data comprises at least one of historical horizontal total radiation, historical temperature and historical rainfall.

[0040] The historical weather data includes horizontal total radiation, temperature, rainfall and other weather data affecting the power of the photovoltaic power station.

[0041] In step S103, the power data of the virtual photovoltaic is generated by using the historical weather data, the large prediction model is pre-trained by using the power data, the large prediction model is fine-tuned based on the power data of the actual photovoltaic power station, and the power prediction result of the actual photovoltaic power station is obtained based on the fine-tuned large prediction model.

[0042] The fine-tuning expression of the large prediction model is as follows:

[0043]

[0044] L(w) is a target function, w is a weight of the large prediction model, is a real power value, is an input power of the large prediction model, is the large prediction model, is a power prediction value.

[0045] As a possible implementation manner, the embodiment of the application can set corresponding virtual photovoltaic parameters according to a mechanism model, generate corresponding power data by using weather data, and the power data is represented as follows:

[0046]

[0047] Wherein, s represents the virtual photovoltaic number, h represents the time range, and 1 year is taken here. s , lat s respectively represent the longitude and latitude of the photovoltaic, respectively represent the horizontal total radiation (W / m2), temperature (℃) and daily rainfall (mm) of the photovoltaic in 1 year, s , tilt s , cap s , op s respectively represent the azimuth angle, tilt angle, installed capacity and operation life of the photovoltaic, represents the cleaning day of the photovoltaic in 1 year, respectively represent the initial efficiency and sensitivity of the photovoltaic, represents the power generation of the photovoltaic in 1 year.

[0048] The embodiment of the application can normalize the data by using the Z-Score method, so as to better train the neural network.

[0049]

[0050] Wherein, μ and σ are the mean and variance respectively, and p' and p are the normalized value and original value respectively.

[0051] Further, the embodiment of the present application can utilize the generated data to pre-train a large prediction model, such as a Transformer model, to obtain a large prediction model, and perform transfer learning on the real photovoltaic power station to fine-tune the large prediction model, thereby obtaining a lightweight model, so as to utilize the lightweight model to predict the actual power generation of the actual photovoltaic power station to be measured.

[0052] In actual execution, the principle of time series prediction can be represented as:

[0053]

[0054] For photovoltaic power prediction, when a photovoltaic power generation sequence is given, the objective function can be defined as:

[0055]

[0056] Suppose the prediction error ∈ s ~ N(0, σ 2 ), then:

[0057]

[0058] Then,

[0059]

[0060]

[0061] Optionally, in an embodiment of the present application, before the large prediction model is pre-trained using the power data, it further includes: introducing time coding in the embedding layer of the Transformer model, and deleting the output layer of the Transformer model to obtain a modified Transformer model; training the modified Transformer model using the historical power sequence to obtain the large prediction model.

[0062] As a possible implementation manner, the embodiment of the present application makes some changes on the structure of the Transformer model. First, the embodiment of the present application can remove the Softmax output layer and directly output with a linear layer; since the power sequence has a clear time label, the time information is considered, and the time coding is introduced in the embedding layer, and the day of the year, the day of the month, the day of the week and the hour of the day are encoded respectively; the single-variable multi-step prediction manner is adopted, and the input and output of the model are only the power generation, the input of the decoder can consider the historical input information, and the output is the power generation sequence of the future multiple time points. The specific process includes:

[0063] Time coding X t= [T1, T2,..., T k ] :

[0064]

[0065] where, denotes the day of the year, denotes the day of the month, denotes the day of the week, denotes the hour of the day.

[0066] Embedding the input features:

[0067] X = val(X p ) + pos(X p ) + tem(X t ),

[0068] The value embedding is a convolutional layer that represents the input features X p = [P1, P2,..., P k ] and the position embedding represents the position of the value in the sequence; the temporal embedding is a linear layer that represents the temporal encoding X t .

[0069] Multi-head attention mechanism: define query Q = XW Q , key K = XW K , value V = XW V , and the dot product attention is defined as:

[0070]

[0071] Multi-head attention is defined as:

[0072] MHA(Q, L, V) = Concat(head1,..., head H )W O ,

[0073] where W i Q , W i K , W i V and W O are the projection weights for the query, key, value and linear transformation, respectively.

[0074] Position feed-forward network:

[0075] FFN(X') = GeLU(X'W1 + b1)W2 + b2,

[0076] Wherein, X' represents the output of the attention layer, GeLU is an activation function, W1 and W2 are projection weights, and b1 and b2 are biases.

[0077] For distributed photovoltaics, some power stations tend to operate for a short time, and have limited historical data. Therefore, the generated power data is used for pre-training, and the actual power station data is used for fine-tuning.

[0078] In combination Figure 2 As shown in FIG. 1, the working principle of the distributed photovoltaic power prediction method of the embodiments of the present application is described in detail.

[0079] The structure involved in the embodiments of the present application can be as shown in Figure 2 As shown in FIG. 1, the embodiments of the present application can deploy the pre-trained model on a cloud platform, collect and transmit distributed photovoltaic power generation data in different regions in real time, and store them on the cloud platform. The pre-trained model is fine-tuned periodically with real data to realize data assimilation.

[0080] For example, the embodiments of the present application can prove the effectiveness of the embodiments of the present application with a specific experiment.

[0081] Among them, the model data set can include: the required meteorological data mainly includes horizontal total radiation, temperature and rainfall. Among them, the accuracy of horizontal total radiation has the greatest impact on power generation, and needs to be as high resolution as possible, and the impact of temperature and rainfall is relatively small.

[0082] (1) Horizontal shortwave radiation: obtained from Himawari-8 satellite, covering 60°S-60°N, 80°E-160°W, spatial resolution 5km, time resolution 10min.

[0083] (2) Temperature: obtained from ERA5, covering the whole world, spatial resolution 25km, time resolution 1h.

[0084] (3) Rainfall: obtained from ERA5, covering the whole world, spatial resolution 25km, time resolution 1h.

[0085] (4) Power: 5 actual power stations 2022.01-2023.05 data, resolution 15min.

[0086] In the power generation part, the embodiment of the application can generate photovoltaic power generation data of 2020 at 10 positions near 5 photovoltaic power stations in Shanxi Province, China, with a resolution of 15 minutes. According to the longitude and latitude of the photovoltaic, the surface shortwave radiation data provided by the Himawari-8 satellite in 2020 is read, and the resolution of 10 minutes is processed to 15 minutes in time. The rule is that the original data corresponds to the time of 0, 10, 20, 30, 40 and 50, and the points at 0 and 30 are retained. The point at 20 represents 15, the point at 50 represents 45, and the missing time is supplemented by the next point. If the next time is also missing, it is linearly interpolated. The processed data corresponds to the time of 0, 15, 30 and 45. For temperature and rainfall, the data of 2020 is downloaded from the ERA5 website. Each photovoltaic site uses the data of the nearest known point to itself. The rainfall is the 24-hour cumulative amount, and the temperature is interpolated from 1 hour to 15 minutes. According to the mechanism model, the photovoltaic parameter value range can be as shown in Table 1, wherein Table 1 is a photovoltaic parameter and value range table.

[0087] Table 1

[0088] Photovoltaic parameters Value range Latitude and longitude Photovoltaic installation site Azimuth angle -10°～+10° Inclination angle Latitude ± 10° Installed capacity 5-15 kW Cleaning days 4～8 Initial efficiency F0 0.8～1.0 Sensitivity Ai 0.001～0.010 Operating life 3～8

[0089] In the experimental setup part, it can include two aspects of pre-training and fine-tuning (correction).

[0090] (1) Pre-training: 1 encoder layer el, 1 decoder layer dl, 8 attention heads, embedding layer dimension dm = 128, feedforward layer dimension df = 512, input length 112, output length 16, batch size batch_size = 32, training 10 times, training set, validation set, test set division is 7:1:2, and the model parameter amount is 0.47M.

[0091] (2) Fine-tuning: embedding layer, output projection layer, training 10 times, fine-tuning parameter amount is 2K.

[0092] In the evaluation index aspect, the embodiment of the application can use mean absolute error and root mean square error to evaluate the prediction effect.

[0093]

[0094]

[0095] wherein C s is the installed capacity, and n is the predicted time step.

[0096] Based on the above parameters and settings, the embodiment of the application can obtain corresponding experimental results.

[0097] Table 2 gives the prediction errors of the embodiments of the present application under different stations and fine-tuning data amounts, compares common long short-term memory network (LSTM) and linear method (DLinear), and the capacity of the five power stations ranges from 100 to 150 MW. The experimental results show that when the fine-tuning parameter amount is 2K, compared with the linear method and the LSTM method, the average absolute error is reduced by 39.60% and 36.28%, respectively, when the real historical data is scarce. When the fine-tuning data amount is 1 month, the prediction error of the embodiments of the present application is reduced more than the comparative method. With the increase of the fine-tuning data amount, the gap between the prediction error of the proposed method and the comparative method is decreasing. Table 2 is a comparison table of prediction errors under different stations and fine-tuning data amounts (the indicators are MAE (MW) and RMSE (MW), respectively).

[0098] Table 2

[0099]

[0100] The experimental results show that the embodiments of the present application can obtain prediction accuracy better than related art by fine-tuning a small amount of parameters when the real historical data is scarce.

[0101] The distributed photovoltaic power prediction method proposed in the embodiments of the present application can generate corresponding virtual power stations around the actual photovoltaic power station to be predicted, obtain historical weather data of the virtual photovoltaic surrounding environment of the virtual power station, generate power data of the virtual photovoltaic, pre-train a large prediction model using the power data, fine-tune the model through the power data of the actual photovoltaic power station, and thus predict the power of the actual photovoltaic power station. That is, a mechanism model is used to generate virtual power stations around the real photovoltaic power station to be predicted, a large prediction model is pre-trained using the generated data, and then transfer learning is performed on the real photovoltaic power station to fine-tune a small amount of parameters, which effectively copes with data scarcity while supporting lightweight deployment. Thus, the problem of insufficient accuracy of the distributed photovoltaic power prediction method in related art is solved.

[0102] Secondly, the distributed photovoltaic power prediction device according to the embodiments of the present application is described with reference to the accompanying drawings.

[0103] Figure 3 is a block schematic diagram of the distributed photovoltaic power prediction device of the embodiments of the present application.

[0104] As shown in Figure 3 , the distributed photovoltaic power prediction device 10 comprises a first acquisition module 100, a second acquisition module 200 and a prediction module 300.

[0105] Specifically, the first acquisition module 100 is configured to generate corresponding virtual power stations around the actual photovoltaic power station to be predicted, and acquire the longitude and latitude information of the virtual photovoltaic of the virtual power station.

[0106] The second acquisition module 200 is configured to acquire historical weather data of a virtual photovoltaic peripheral environment based on latitude and longitude information.

[0107] The prediction module 300 is configured to generate power data of the virtual photovoltaic by using the historical weather data, pre-train a large prediction model by using the power data, fine-tune the large prediction model based on actual power data of the actual photovoltaic power station, and obtain a power prediction result of the actual photovoltaic power station based on the fine-tuned large prediction model.

[0108] Optionally, in an embodiment of the present application, the fine-tuning expression of the large prediction model is as follows:

[0109]

[0110] wherein L(W) is a target function, W is a weight of the large prediction model, is a power real value, is an input power of the large prediction model, is the large prediction model, is a power prediction value.

[0111] Optionally, in an embodiment of the present application, the distributed photovoltaic power prediction device 10 further comprises a control module and a training module.

[0112] The control module is configured to introduce time coding in an embedding layer of a Transformer model and delete an output layer of the Transformer model to obtain a modified Transformer model.

[0113] The training module is configured to train the modified Transformer model by using a historical power sequence to obtain the large prediction model.

[0114] Optionally, in an embodiment of the present application, the historical weather data comprises at least one of historical horizontal total radiation, historical temperature and historical rainfall.

[0115] It should be noted that the foregoing explanation and description of the embodiment of the distributed photovoltaic power prediction method also apply to the distributed photovoltaic power prediction device of this embodiment, which will not be described here again.

[0116] According to the distributed photovoltaic power prediction device provided in the embodiment of the present application, a corresponding virtual power station can be generated around the actual photovoltaic power station to be predicted, and the historical weather data of the peripheral environment of the virtual photovoltaic power station is obtained, so as to generate the power data of the virtual photovoltaic power station, and the power data is used to pre-train a large prediction model, and the model is fine-tuned through the power data of the actual photovoltaic power station, so as to perform power prediction on the actual photovoltaic power station, that is, a virtual power station is generated around the real photovoltaic power station to be predicted by using a mechanism model, a large prediction model is pre-trained by using the generated data, and then migration learning is performed on the real photovoltaic power station, the parameter quantity is small, and the light-weight deployment is supported while effectively coping with the data scarcity. Therefore, the problem of insufficient accuracy of the distributed photovoltaic power prediction method in the related art is solved.

[0117] Figure 4 The structure schematic diagram of the electronic device provided in the embodiment of the present application is provided. The electronic device can include:

[0118] The memory 401, the processor 402 and the computer program stored in the memory 401 and executable on the processor 402.

[0119] The processor 402 implements the distributed photovoltaic power prediction method provided in the above embodiment when executing the program.

[0120] Further, the electronic device further includes:

[0121] The communication interface 403 is used for communication between the memory 401 and the processor 402.

[0122] The memory 401 is used to store the computer program executable on the processor 402.

[0123] The memory 401 can include a high-speed RAM memory, and can also include a non-volatile memory, for example, at least one disk memory.

[0124] If the memory 401, the processor 402 and the communication interface 403 are independently implemented, the communication interface 403, the memory 401 and the processor 402 can be connected to each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 4Only one bus or only one type of bus can exist, however.

[0125] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can complete the communication among each other through an internal interface.

[0126] The processor 402 can be a central processing unit (CPU) or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement one or more embodiments of the present application.

[0127] The embodiment further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the distributed photovoltaic power prediction method.

[0128] The embodiment of the present application further provides a computer program product, which comprises a computer program. The computer program is executed by a processor to implement the distributed photovoltaic power prediction method provided by the embodiment of the present application.

[0129] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or N embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples, without contradiction.

[0130] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one feature. In the description of the present application, the meaning of "N" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0131] Any processes or methods described in the flowcharts or otherwise described herein can be understood as representing a module, segment, or portion of code that includes one or N steps for implementing the specified logical functions or processes. The scope of a preferred embodiment of the present application encompasses combinations of one or more steps of the described processes, and alternative implementations including those that do not perform the steps in the order shown or discussed, including those that perform the steps in substantially simultaneous, reverse order, or in other orders, as will be understood by those skilled in the art of the embodiments described herein.

[0132] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a product of the manufacturing and / or processing, and can be a machine-readable storage medium (alternatively referred to as machine-readable non-transitory storage medium). The computer-readable medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Note that the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0133] It should be understood that aspects of the present application can be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, the N steps or processes can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As such, if implemented in hardware, and in another embodiment, any of the following technologies, known in the art, or combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and so forth.

[0134] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment method can be completed by programs instructing related hardware, and the programs can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0135] In addition, each functional unit in each embodiment of the present application can be integrated into one processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0136] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for predicting distributed photovoltaic power, characterized in that, Includes the following steps: A corresponding virtual power station is generated around the actual photovoltaic power station to be predicted, and the latitude and longitude information of the virtual photovoltaic power station is obtained. Historical weather data of the surrounding environment of the virtual photovoltaic system are obtained based on the latitude and longitude information. The power data of the virtual photovoltaic system is generated using the historical weather data. A large prediction model is pre-trained using the power data. The large prediction model is then fine-tuned based on the power data of the actual photovoltaic power station. The power prediction result of the actual photovoltaic power station is obtained based on the fine-tuned large prediction model. The process includes, before pre-training the large prediction model using the power data, introducing time coding into the embedding layer of the Transformer model and deleting the output layer of the Transformer model to obtain a modified Transformer model; and training the modified Transformer model using historical power sequences to obtain the large prediction model. The step of deleting the output layer of the Transformer model to obtain the modified Transformer model includes: deleting the Softmax output layer of the Transformer model and outputting it as a linear layer; introducing time coding in the embedding layer; adopting a univariate multi-step prediction method; using power generation as the input and output of the model; using historical input information as the input of the decoder; and outputting a power generation sequence for multiple future time periods.

2. The method according to claim 1, characterized in that, The fine-tuning expression for the large prediction model is as follows: ， in, Let be the objective function. To predict the weights of large models, This is the actual power value. To predict the input power of a large model, To predict large models, This is the predicted power value.

3. The method according to claim 1, characterized in that, The historical weather data includes at least one of historical total horizontal radiation, historical temperature, and historical rainfall.

4. A distributed photovoltaic power prediction device, characterized in that, include: The first acquisition module is used to generate a corresponding virtual power station around the actual photovoltaic power station to be predicted, and to acquire the latitude and longitude information of the virtual photovoltaic power station. The second acquisition module is used to acquire historical weather data of the surrounding environment of the virtual photovoltaic system based on the latitude and longitude information. The prediction module is used to generate power data of the virtual photovoltaic system using the historical weather data, pre-train a large prediction model using the power data, fine-tune the large prediction model based on the power data of the actual photovoltaic power station, and obtain the power prediction result of the actual photovoltaic power station based on the fine-tuned large prediction model. The control module is used to introduce temporal encoding into the embedding layer of the Transformer model and remove the output layer of the Transformer model to obtain the modified Transformer model; the training module is used to train the modified Transformer model using historical power sequences to obtain the large prediction model. The control module is further used to remove the Softmax output layer of the Transformer model and output it as a linear layer. It introduces time coding in the embedding layer and adopts a univariate multi-step prediction method. The power generation is used as the input and output of the model. The input of the decoder is historical input information, and the output is the power generation sequence at multiple future times.

5. The apparatus according to claim 4, characterized in that, The fine-tuning expression for the large prediction model is as follows: ， in, Let be the objective function. To predict the weights of large models, This is the actual power value. To predict the input power of a large model, To predict large models, This is the predicted power value.

6. The apparatus according to claim 4, characterized in that, The historical weather data includes at least one of historical total horizontal radiation, historical temperature, and historical rainfall.

7. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the distributed photovoltaic power prediction method as described in any one of claims 1-3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the distributed photovoltaic power prediction method as described in any one of claims 1-3.

Images