Prediction method and device for photovoltaic combined power and computer equipment

Abstract

The invention discloses a photovoltaic joint power prediction method and device and computer equipment. The method comprises the steps that multi-dimensional feature data of each photovoltaic station are acquired, and the multi-dimensional feature data comprise historical power data, current meteorological data and weather forecast data; performing feature extraction on the multi-dimensional feature data by adopting a power prediction model to obtain a multi-dimensional feature vector, and performing coding of different dimensions on the multi-dimensional feature vector in sequence to obtain a position coding vector, a time coding vector and a similarity coding vector, the position coding vector being used for representing the position feature of each station, the time coding vector being used for representing the similarity coding vector, and the time coding vector being used for representing the similarity coding vector; the time coding vector is used for representing the time feature of the multi-dimensional feature data of each station, and the similarity coding vector is used for representing the similarity degree among the plurality of photovoltaic stations; and analyzing the position coding vector, the time coding vector and the similarity coding vector by adopting a power prediction model to obtain a prediction result.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic power generation, and more specifically, to a photovoltaic combined power prediction method, device and computer equipment. Background Art

[0002] With the rapid development of renewable energy power generation, photovoltaic power prediction technology has received increasing attention. Traditional prediction methods are mostly based on statistical analysis and machine learning algorithms (such as support vector machines, neural networks, etc.), and are mostly independent predictions for single energy sites. In recent years, deep learning technology has emerged, but its application in photovoltaic power prediction is still limited: 1. Insufficient use of spatiotemporal correlation: Existing photovoltaic power prediction methods often do not fully consider the complex spatiotemporal coupling characteristics between sites in the region. When traditional machine learning methods process spatial information, they rely on simple distance metrics and are difficult to effectively model the impact of spatial relationships such as geographical distance and meteorological correlation between sites on power output. In the time dimension, its modeling ability is limited, and its ability to capture long-term dependencies is insufficient, which affects the prediction accuracy. 2. The prediction accuracy needs to be improved: Some deep learning-based models are prone to overfitting or underfitting when processing complex photovoltaic power station power data. Overfitting makes the model overly dependent on training data and has poor generalization ability; underfitting does not fully capture the potential laws of the data, resulting in low actual prediction accuracy, affecting the reliability in photovoltaic power prediction. 3. Lack of adaptability and flexibility: Existing technologies lack adaptability and flexibility when facing photovoltaic sites of different regions and sizes. Factors such as meteorological conditions in different regions and the geographical distribution of stations have a significant impact on prediction accuracy, but traditional models are difficult to adjust according to specific station layouts and optical resource characteristics, and have poor universality. 4. Insufficient cross-station information fusion: Many existing methods do not effectively integrate the temporal and spatial characteristics of different stations. The model often processes the historical power data of each station independently or ignores the spatial relationship between stations, and is unable to fully utilize cross-station information to improve prediction accuracy. In multi-station prediction tasks, the association between meteorological data and historical power data of different stations is not effectively utilized, and the information fusion efficiency is low, which affects the prediction effect. Summary of the invention

[0003] The embodiments of the present application provide a photovoltaic combined power prediction method, device and computer equipment to at least solve the technical problem of low prediction accuracy in the related art.

[0004] According to one aspect of an embodiment of the present application, a method for predicting photovoltaic combined power is provided, comprising: obtaining multi-dimensional feature data of each photovoltaic site, wherein the multi-dimensional feature data comprises: historical power data, current meteorological data and weather forecast data; using a power prediction model to perform feature extraction on the multi-dimensional feature data to obtain a multi-dimensional feature vector, and encoding the multi-dimensional feature vector in different dimensions in turn to obtain a position coding vector, a time coding vector and a similarity coding vector, wherein the position coding vector is used to represent the position feature of each site, the time coding vector is used to represent the time feature of the multi-dimensional feature data of each site, and the similarity coding vector is used to represent the degree of similarity between multiple photovoltaic sites; using a power prediction model to analyze the position coding vector, the time coding vector and the similarity coding vector to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic sites in the prediction period.

[0005] Optionally, a power prediction model is used to predict a prediction result based on the position coding, time coding and similarity coding, including: position coding the multidimensional feature vector to obtain the position coding vector; time coding the multidimensional feature vector to obtain the time coding vector; similarity coding the multidimensional feature vector to obtain the similarity coding vector; fusing the position coding vector, the time coding vector and the similarity coding vector to obtain a fused vector; and predicting the fused vector using a power prediction model to obtain the prediction result.

[0006] Optionally, position encoding is performed on the multi-dimensional feature vector to obtain the position encoding vector, including: obtaining the position information of each photovoltaic field station and the total dimension of the position encoding vector; dividing the dimension of the position encoding into an even dimension and an odd dimension; when the dimension of the position encoding is an even dimension, determining the even dimension encoding value of the position encoding based on a sine function, the position information of each photovoltaic field station and the total dimension of the position encoding; when the dimension of the position encoding is an odd dimension, determining the odd dimension encoding value of the position encoding based on a cosine function, the position information of each photovoltaic field station and the total dimension of the position encoding vector; and determining the position encoding vector based on the even dimension encoding value of the position encoding and the odd dimension encoding value of the position encoding.

[0007] Optionally, the multidimensional feature vector is time-encoded to obtain the time-encoded vector, including: obtaining the time dimension of the multidimensional feature vector, the time dimension including at least: month, day, hour and minute, and obtaining the value range of each time dimension; determining the embedding matrix of each time dimension based on the value range of each time dimension, the number of columns of the embedding matrix is ​​the time dimension of the multidimensional feature vector, the number of rows of the embedding matrix is ​​the number of values ​​of each time dimension, and the elements of the embedding matrix represent the embedding vector corresponding to each time dimension value; obtaining the time feature of the multidimensional feature vector, and obtaining the embedding vector corresponding to the time feature from the embedding matrix; adding the embedding vectors of different time dimensions corresponding to the time feature to obtain the time-encoded vector.

[0008] Optionally, similarity encoding is performed on the multi-dimensional feature vector to obtain the similarity encoding vector, including: determining the similarity of each photovoltaic station in each dimensional feature based on the multi-dimensional feature data of each photovoltaic station; determining the similarity between multiple stations based on the similarity of each dimensional feature of each photovoltaic station; encoding the feature representation of the similarity between the multiple stations to obtain the similarity encoding vector.

[0009] Optionally, the similarity of the characteristics of multiple photovoltaic stations in each dimension is determined according to the multi-dimensional feature data of each photovoltaic station, including: determining the distance between any two photovoltaic stations according to the location information of each photovoltaic station; constructing a distance similarity matrix based on the distance between any two photovoltaic stations, and the distance similarity matrix is ​​used to represent the distance similarity between each photovoltaic station; extracting meteorological features from the multi-dimensional feature vector, and the meteorological features include: current meteorological features and historical meteorological features; performing similarity comparison on the meteorological features of each photovoltaic station to obtain a meteorological feature similarity matrix, and the meteorological feature similarity matrix is ​​used to represent the meteorological feature similarity between each photovoltaic station; extracting weather forecast features from the multi-dimensional feature vector, and converting each The meteorological characteristics of the photovoltaic stations are compared for similarity to obtain a weather forecast characteristic similarity matrix, and the weather forecast characteristic similarity matrix is ​​used to represent the weather forecast characteristic similarity between each photovoltaic station; the meteorological characteristic similarity matrix and the weather forecast characteristic similarity matrix are determined as a meteorological similarity matrix; power characteristics are extracted from the multi-dimensional feature vector, and the power characteristics of each photovoltaic station are compared for similarity to obtain a power characteristic similarity matrix, and the power characteristic similarity matrix is ​​used to represent the power characteristic similarity between each photovoltaic station; the distance similarity matrix, the meteorological similarity matrix and the power characteristic similarity matrix are weightedly summed to determine a weighted similarity matrix, wherein the weighted similarity matrix represents the degree of similarity between the multiple stations.

[0010] Optionally, the prediction result is obtained based on the fusion vector, including: encoding the fusion vector through the encoding layer in the preset power prediction model to obtain an initial feature vector; performing feature processing on the initial feature vector through the fully connected layer in the preset power prediction model to obtain a target feature vector; and outputting the prediction result of the target feature vector through the output layer in the preset power prediction model.

[0011] According to another aspect of an embodiment of the present application, a photovoltaic combined power prediction device is also provided, including: an acquisition module, used to acquire multi-dimensional feature data of each photovoltaic site, wherein the multi-dimensional feature data includes: historical power data, current meteorological data and weather forecast data; a feature module, used to use a power prediction model to perform feature extraction on the multi-dimensional feature data to obtain a multi-dimensional feature vector, and encode the multi-dimensional feature vector in different dimensions in turn to obtain a position coding vector, a time coding vector and a similarity coding vector, wherein the position coding vector is used to represent the position feature of each site, the time coding vector is used to represent the time feature of the multi-dimensional feature data of each site, and the similarity coding vector is used to represent the degree of similarity between multiple photovoltaic sites; a prediction module, used to use a power prediction model to analyze the position coding vector, the time coding vector and the similarity coding vector to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic sites in the prediction period.

[0012] According to another aspect of the embodiment of the present application, a communication device is provided, including: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and is used to execute the above-mentioned photovoltaic combined power prediction method.

[0013] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned photovoltaic combined power prediction method by running the computer program.

[0014] According to another aspect of the embodiments of the present application, a computer program product is provided, including computer instructions, which implement the above-mentioned photovoltaic combined power prediction method when executed by a processor.

[0015] In an embodiment of the present application, multi-dimensional feature data of each photovoltaic station is obtained, wherein the multi-dimensional feature data includes: historical power data, current meteorological data and weather forecast data; a power prediction model is used to extract features from the multi-dimensional feature data to obtain a multi-dimensional feature vector, and the multi-dimensional feature vector is encoded in different dimensions in turn to obtain a position encoding vector, a time encoding vector and a similarity encoding vector, wherein the position encoding vector is used to represent the position feature of each station, the time encoding vector is used to represent the time feature of the multi-dimensional feature data of each station, and the similarity encoding vector is used to represent the degree of similarity between multiple photovoltaic stations; a power prediction model is used to analyze the position encoding vector, the time encoding vector and the similarity encoding vector to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic stations in the prediction period, thereby achieving the purpose of jointly predicting the encoding vectors based on the position, time and similarity of the multi-dimensional features of the photovoltaic stations, thereby achieving the technical effect of improving the accuracy of power prediction, and thus solving the technical problem of low prediction accuracy in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 It is a hardware structure block diagram of a computer terminal for implementing a photovoltaic combined power prediction method according to an embodiment of the present application;

[0018] Figure 2 is a flow chart of a photovoltaic combined power prediction method according to an embodiment of the present application;

[0019] Figure 3 is a flow chart of another photovoltaic combined power prediction method according to an embodiment of the present application;

[0020] Figure 4 is a flowchart of a power prediction model prediction according to an embodiment of the present application;

[0021] Figure 5 It is a structural diagram of a photovoltaic combined power prediction device according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] The information collected in the embodiments of the present application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or reject automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered.

[0025] In order to solve the problems existing in the related art, the present application embodiment provides a photovoltaic combined power prediction method, which can be run on Figure 1 In the computer terminal shown, the computer terminal is explained below.

[0026] The photovoltaic combined power prediction method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal for implementing a photovoltaic combined power prediction method is shown. Figure 1As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0027] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0028] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the photovoltaic combined power prediction method in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned photovoltaic combined power prediction method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0029] The transmission module 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0031] It should be noted that, in some optional embodiments, the above Figure 1 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0032] Under the above-mentioned operating environment, an embodiment of the present application provides an embodiment of a method for predicting photovoltaic combined power. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0033] Figure 2 is a flow chart of a photovoltaic combined power prediction method according to an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0034] Step S202, obtaining multi-dimensional characteristic data of each photovoltaic station, wherein the multi-dimensional characteristic data includes: historical power data, current meteorological data and weather forecast data;

[0035] In step S202, the weather forecast data may be NWP (Numerical Weather Prediction) data. Meteorological data mainly include solar radiation, temperature, etc. Taking T time steps as an example, the multi-dimensional feature data may be expressed as N×T×D, where N represents the amount of data and D represents the feature dimension of the multi-dimensional feature data.

[0036] Step S204: extracting features from the multi-dimensional feature data using a power prediction model to obtain a multi-dimensional feature vector, and encoding the multi-dimensional feature vector in different dimensions in turn to obtain a position encoding vector, a time encoding vector, and a similarity encoding vector, wherein the position encoding vector is used to represent the position feature of each station, the time encoding vector is used to represent the time feature of the multi-dimensional feature data of each station, and the similarity encoding vector is used to represent the degree of similarity between multiple photovoltaic stations;

[0037] In step S204, the location feature may be the distance between multiple stations, the longitude and latitude of the stations, etc.

[0038] Step S206: using a power prediction model to analyze the position coding vector, the time coding vector and the similarity coding vector to obtain a prediction result, wherein the prediction result represents the power of the plurality of photovoltaic stations in a prediction period.

[0039] Through the above steps S202 to S206, by obtaining multi-dimensional feature data of each photovoltaic station, wherein the multi-dimensional feature data includes: historical power data, current meteorological data and weather forecast data; using a power prediction model to extract features from the multi-dimensional feature data to obtain a multi-dimensional feature vector, and encoding the multi-dimensional feature vector in different dimensions in turn to obtain a position coding vector, a time coding vector and a similarity coding vector, wherein the position coding vector is used to represent the position feature of each station, the time coding vector is used to represent the time feature of the multi-dimensional feature data of each station, and the similarity coding vector is used to represent the degree of similarity between multiple photovoltaic stations; using a power prediction model to analyze the position coding vector, the time coding vector and the similarity coding vector to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic stations in the prediction period, thereby achieving the purpose of jointly predicting the coding vectors based on the position, time and similarity of the multi-dimensional features of the photovoltaic station, thereby achieving the technical effect of improving the accuracy of power prediction, and thus solving the technical problem of low prediction accuracy in related technologies.

[0040] In the technical solution provided in step S206 of the above-mentioned photovoltaic combined power prediction method, a power prediction model is used to obtain a prediction result based on the position coding, time coding and similarity coding prediction, including: position coding the multidimensional feature vector to obtain the position coding vector; time coding the multidimensional feature vector to obtain the time coding vector; similarity coding the multidimensional feature vector to obtain the similarity coding vector; fusing the position coding vector, the time coding vector and the similarity coding vector to obtain a fused vector; and using the power prediction model to predict the fused vector to obtain the prediction result.

[0041] There are many ways to fuse coding vectors. Among them, one optional fusion method is that when the feature dimensions of the position coding vector, the time coding vector and the similarity coding vector are the same, the three coding vectors can be directly spliced ​​and fused by splicing, and the feature dimensions after splicing remain unchanged. In another method, the three coding vectors can also be directly added to obtain a fused vector.

[0042] In some embodiments of the present application, position encoding is performed on the multi-dimensional feature vector to obtain the position encoding vector, including: obtaining the position information of each photovoltaic station and the total dimension of the position encoding vector; dividing the dimension of the position encoding into an even dimension and an odd dimension; when the dimension of the position encoding is an even dimension, determining the even dimension encoding value of the position encoding based on a sine function, the position information of each photovoltaic station and the total dimension of the position encoding; when the dimension of the position encoding is an odd dimension, determining the odd dimension encoding value of the position encoding based on a cosine function, the position information of each photovoltaic station and the total dimension of the position encoding vector; and determining the position encoding vector based on the even dimension encoding value of the position encoding and the odd dimension encoding value of the position encoding.

[0043] For example: determine the total dimension of the position encoding vector as d_model, and for each position pos, calculate the value of each pos in the position encoding vector.

[0044] Specifically, the dimensions of the position encoding vector are divided into even dimensions and odd dimensions, and the values ​​are calculated using the sine function and the cosine function respectively.

[0045] For even dimensions 2i, the sine function is used to generate values, with the following formula:

[0046] PE(pos,2i)=sin(pos / 10000^(2i / d_model));

[0047] For odd dimensions 2i+1, the cosine function is used, and its formula is:

[0048] PE(pos,2i+1)=cos(pos / 10000^(2i / d_model));

[0049] In the formula, PE(pos,i) represents the position encoding value of position pos in the i-th dimension, pos represents the position index value, and 10000^(2i / d_model) is a basic frequency adjustment term, which increases with the increase of i, thereby introducing different frequencies between different dimensions, so that the final position encoding vector shows periodic changes in different dimensions, but the period lengths are different. This provides the model with the ability to distinguish different positions, while also ensuring that the encoding vector changes smoothly with the position, which is especially important when processing long sequence data. After the calculation is completed, a d_model-dimensional position encoding vector is obtained, in which the value of each dimension is calculated based on pos and the above formula. These values ​​have a specific periodicity, that is, the frequency gradually changes between adjacent dimensions in the vector. This design makes the encoding easier for the model to learn and avoids the learning difficulty that may be caused by fixed position encoding. Finally, the position encoding vector is directly added to each multi-dimensional feature vector, so that each input data point carries its position information, allowing the model to perceive the relative position of the data point in the sequence when processing, which is crucial for capturing the dependency of sequence data.

[0050] The entire process ensures that the generation of position encoding vectors takes into account both the uniqueness of the position and the continuity and periodicity between positions through dynamic calculation of sine and cosine functions, providing the model with simple yet powerful position information, which helps to improve the model's processing performance on sequence data.

[0051] The specific steps of time encoding the multidimensional feature vector to obtain the time encoding vector are as follows: obtain the time dimension of the multidimensional feature vector, the time dimension at least includes: month, day, hour and minute, and obtain the value range of each time dimension; determine the embedding matrix of each time dimension based on the value range of each time dimension, the number of columns of the embedding matrix is ​​the time dimension of the multidimensional feature vector, the number of rows of the embedding matrix is ​​the number of values ​​of each time dimension, and the elements of the embedding matrix represent the embedding vector corresponding to each time dimension value; obtain the time feature of the multidimensional feature vector, and obtain the embedding vector corresponding to the time feature from the embedding matrix; add the embedding vectors of different time dimensions corresponding to the time feature to obtain the time encoding vector.

[0052] For example, add time information to the data at each time point, and in specific operations, cover the five time dimensions of minutes, hours, working days, days, and months. Use a fixed embedding method to construct embedding layers for hours, working days, days, months, and minutes, and their dimensions correspond to their respective value ranges to d_model. After converting the input data to long integers, calculate the encoding vectors of hours, working days, days, months, and minutes respectively, and then add these vectors to get the final time encoding representation. After time encoding, the data dimension becomes N×T×d_model, so that each data point is rich in time information, so that the model can better learn the law of power changes over time.

[0053] Specifically, for each time dimension (minute, hour, weekday, day, month), create a corresponding embedding matrix. For example, there are 24 possible values ​​for hours, 7 for weekdays (Monday to Sunday), 31 for days (days in a month), 12 for months (months in a year), and 1440 for minutes in a day (24 hours × 60 minutes). The number of columns in each embedding matrix is ​​the time dimension of the multidimensional feature vector, and the number of rows corresponds to the number of possible values ​​for that time dimension. The elements of each matrix are pre-computed fixed values, which can be randomly initialized and then fine-tuned as the model is trained, or they can be hand-designed, such as values ​​generated by sine and cosine functions.

[0054] Time features can be extracted from the timestamp of multi-dimensional feature data, for example: the minute, hour, weekday, day, and month values ​​corresponding to each multi-dimensional data. Conversion data type:

[0055] Each time dimension value is used as an index, and the corresponding embedding vector is extracted from the embedding matrix of each time dimension. For example, if the time feature of a multi-dimensional feature data point is "14:30 on Friday", then the embedding vectors of "Friday", "14", and "30" will be extracted from the embedding matrices of weekdays, hours, and minutes respectively.

[0056] The embedding vectors obtained from different time dimensions are added together to generate the final time encoding representation. The addition is usually done element by element to ensure that the information of each time dimension is fully preserved.

[0057] The generated time encoding vector is merged with the multi-dimensional feature vector of the original input. In the Transformer model, this is usually done by adding the time encoding vector directly to the embedding vector of the data point. In this way, the feature vector of each data point contains time information, which helps the model learn the law of power output changing over time.

[0058] Through the above steps, each data point is given information related to the time dimension, which will help the model better understand the seasonality, periodicity and more detailed time variation patterns of the data, thereby improving the accuracy of PV power prediction. This time encoding strategy is particularly suitable for processing data with obvious periodic changes, such as PV power, because power output is affected by factors such as time of day, season, and working day.

[0059] The specific steps of similarity encoding the multi-dimensional feature vector to obtain the similarity encoding vector include: determining the similarity of each photovoltaic station in each dimensional feature based on the multi-dimensional feature data of each photovoltaic station; determining the similarity between multiple stations based on the similarity of each dimensional feature of each photovoltaic station; encoding the feature representation of the similarity between the multiple stations to obtain the similarity encoding vector.

[0060] Among them, the step of determining the similarity of the characteristics of multiple photovoltaic stations in each dimension according to the multi-dimensional feature data of each photovoltaic station includes: determining the distance between any two photovoltaic stations according to the location information of each photovoltaic station; constructing a distance similarity matrix based on the distance between any two photovoltaic stations, and the distance similarity matrix is ​​used to represent the distance similarity between each photovoltaic station; extracting meteorological features from the multi-dimensional feature vector, and the meteorological features include: current meteorological features and historical meteorological features; performing similarity comparison on the meteorological features of each photovoltaic station to obtain a meteorological feature similarity matrix, and the meteorological feature similarity matrix is ​​used to represent the meteorological feature similarity between each photovoltaic station; extracting weather forecast features from the multi-dimensional feature vector, and converting each The meteorological characteristics of each photovoltaic station are compared for similarity to obtain a weather forecast characteristic similarity matrix, and the weather forecast characteristic similarity matrix is ​​used to represent the weather forecast characteristic similarity between each photovoltaic station; the meteorological characteristic similarity matrix and the weather forecast characteristic similarity matrix are determined as a meteorological similarity matrix; power characteristics are extracted from the multi-dimensional feature vector, and the power characteristics of each photovoltaic station are compared for similarity to obtain a power characteristic similarity matrix, and the power characteristic similarity matrix is ​​used to represent the power characteristic similarity between each photovoltaic station; the distance similarity matrix, the meteorological similarity matrix and the power characteristic similarity matrix are weightedly summed to determine a weighted similarity matrix, wherein the weighted similarity matrix represents the degree of similarity between the multiple stations.

[0061] For distance similarity: Calculate the geographic distance between each station based on the latitude and longitude information of each station. Common distance calculation formulas can be used, such as Euclidean distance or great circle distance. Get an N×N distance matrix to represent the distance similarity between each station. Normalize the distance matrix so that its value is in the interval [0,1] so that it can be better integrated with other similarity indicators.

[0062] For meteorological similarity: analyze the similarity of meteorological conditions at different locations. Here, consider the actual meteorological data and NWP data separately. For each time point, compare and analyze the current meteorological characteristics with the historical meteorological characteristics, and also perform similarity analysis on the NWP data. The meteorological similarity between two stations can be calculated using cosine similarity or other similarity measurement methods to obtain two meteorological similarity matrices (one for actual meteorology and one for NWP). The two similarity matrices are normalized separately.

[0063] For power similarity: analyze the similarity of power output at different locations. The power correlation coefficient or other similarity measures between two stations can be calculated to obtain an N×N power feature similarity matrix. Perform normalization. The latitude and longitude distance similarity matrix, the actual meteorological similarity matrix, the NWP meteorological similarity matrix and the power feature similarity matrix are weighted and summed to obtain the final weighted similarity matrix. The weights can be adjusted according to actual conditions. The weighted similarity matrix is ​​fused with the multi-dimensional feature vector. The similarity matrix can be expanded to a tensor with the same dimension as the multi-dimensional feature vector (i.e., N×T×d_model), and element-by-element multiplication or addition is performed with the original data. After weighted similarity encoding, the data dimension is still N×T×d_model, but the data contains weighted similarity information.

[0064] In some embodiments of the present application, the specific steps for obtaining the prediction result according to the fusion vector are as follows: encoding the fusion vector through the encoding layer in the preset power prediction model to obtain an initial feature vector; performing feature processing on the target feature vector through the fully connected layer in the preset power prediction model to obtain a target feature vector; and outputting the prediction result of the target feature vector through the output layer in the preset power prediction model.

[0065] Figure 3 A method for predicting photovoltaic combined power is also shown, such as Figure 3 As shown, the data input into the photovoltaic power prediction model includes: photovoltaic real-time meteorological data, historical power data and NWP, wherein the real-time meteorological data includes: current meteorological data and historical meteorological data, and the prediction result is obtained based on the above data.

[0066] Figure 4A prediction process of a photovoltaic power prediction model is also shown, such as Figure 4 As shown in the figure, the multi-dimensional feature data received by the input layer is encoded into weighted similarity coding, time coding and position coding through the embedding layer, and finally the prediction result is obtained.

[0067] The fusion vector is input into the encoder of the photovoltaic power prediction model. The photovoltaic power prediction model can be a Transformer model. The Transformer encoder has powerful capabilities. It mainly operates with the help of self-attention mechanism and multi-head attention mechanism, and can automatically mine and learn the intricate dependencies in the data.

[0068] The Transformer encoder is designed to maintain the consistency of input and output dimensions, that is, its output dimension is also N×T×d_model, and the dimensional form of the input data will not be changed (where N represents the number of stations, T represents the time step, and d_model is the feature dimension). Of course, although the dimension remains unchanged, the encoder can extract more valuable feature representations that better reflect the internal connection of the data through deep processing of the data, laying the foundation for further analysis and prediction. The specific internal processing effect and feature learning are determined by the structure of the Transformer and related parameter settings. The data processed by the Transformer encoder enters the fully connected layer, which further refines the features and learns the nonlinear relationship in the data. The fully connected layer captures the complex interactions between the features by weighted summing the features output by the Transformer encoder, thereby extracting more discriminative features. The main goal of the fully connected layer is to deeply process the feature vector processed by the encoder to make it more suitable for the power prediction task. The output of the fully connected layer enters the multi-station power output layer, which outputs the power prediction results of multiple stations. The final output dimension is N×T×1, where each element corresponds exactly to the power prediction value of a station at a specific time point. The function of this layer is to "convert" features into reality, that is, to directly derive physically meaningful prediction values ​​from high-dimensional features, which is the final key link of the model.

[0069] In the whole process, the method of joint training of data from multiple sites is adopted, which has unique advantages. During the training process, the model can fully tap the inherent correlation and complementarity between different sites, allowing the data of each site to "collaborate" with each other, thereby effectively improving the accuracy of power prediction. Moreover, by continuously fine-tuning and optimizing the model parameters (such as the weights and biases of each layer of the neural network) and hyperparameters (such as learning rate, batch size, number of iterations, etc.), the potential of the model can be further tapped to achieve better performance.

[0070] It should be noted that real-time meteorological data and numerical weather forecast (NWP) data were introduced in the data input stage. These two types of data provide richer and more comprehensive meteorological information. With this additional information, the model can more keenly and accurately capture the impact of changing meteorological conditions on power output, and gain a deeper insight into the subtle relationship between different sites, thereby making more accurate and reliable judgments in power forecasting tasks.

[0071] It is also necessary to explain that the embodiment of the present application provides a photovoltaic joint power prediction method, which makes full use of the advantages of Transformer to predict photovoltaic power. The Transformer architecture can automatically mine and learn the complex dependencies in photovoltaic field station related data (such as power data, meteorological data, etc.) by virtue of its self-attention mechanism and multi-head attention mechanism, and is particularly good at processing long-distance dependencies. In the present invention, these data are input into Transformer for time series feature modeling, so that it can effectively capture the long-term dependency features in the data, thereby improving the accuracy and stability of photovoltaic power prediction. In the model construction process, the data is input into Transformer after multi-step processing in turn, covering time coding (comprehensively considering multiple time dimensions to add time information), position coding (using sine and cosine functions to generate position coding vectors to integrate position information), weighted similarity coding (calculating similarity from multiple aspects and weighted fusion) and feature fusion operations, providing Transformer with high-quality input data rich in multi-dimensional features to ensure that it plays a better role. Construct a unified prediction model and set up a photovoltaic field group prediction module to carry out joint modeling so that each part works together. In terms of data, the model input is rich and targeted, including historical power data and numerical weather forecast (NWP) data. Historical power data records the power change law of photovoltaic stations. NWP data predicts the weather based on atmospheric science theory and calculation, provides meteorological information for prediction, and can also include station geographic information. This joint modeling takes advantage of the characteristics of photovoltaics. Photovoltaics are greatly affected by natural conditions. Different photovoltaic stations have power complementary characteristics in time and condition dimensions. For example, coastal and inland photovoltaic fields have different power performances due to wind factors at different times, and power complementarity is more significant in different seasons and weather conditions. This enables the data of each station to cooperate with each other during the training phase. Multi-task joint modeling also gives the model a strong generalization ability, so that it can quickly adapt to complex and changeable actual application environments, such as changes in meteorology, geography, and power output scenarios, and stably output reliable prediction results based on the learned correlation and complementarity, so as to achieve efficient and accurate prediction. A series of coding processes such as time coding, location coding, and weighted similarity coding are integrated. Time coding adds time-related features to the data based on different time dimensions (such as minutes, hours, working days, days, and months); location coding generates location coding vectors with the help of specific functions to integrate location information into the data; weighted similarity coding calculates similarity matrices from multiple angles such as longitude and latitude distance, weather (live weather and NWP data), and power, and integrates this information into the original data through reasonable weighting, fully integrating the spatiotemporal characteristics of the data and providing rich and effective feature representation for subsequent model learning. At the same time, numerical weather forecast (NWP) data and historical power data are organically combined and applied to the power prediction model.NWP data can reflect the state of atmospheric movement and weather phenomena in a certain period of time in the future. As important input data, it participates in the training of the model and each link of the prediction together with the historical power data, so that the model can accurately model and analyze how meteorological changes affect power output. Through the joint application of the two, the model can keenly capture the complex relationship between changes in meteorological conditions and power output, and can more accurately predict power in different meteorological scenarios, improving the reliability and accuracy of the prediction.

[0072] Figure 5 A photovoltaic combined power prediction device according to an embodiment of the present application includes:

[0073] An acquisition module 50 is used to acquire multi-dimensional characteristic data of each photovoltaic station, wherein the multi-dimensional characteristic data includes: historical power data, current meteorological data and weather forecast data;

[0074] The feature module 52 is used to extract features from the multi-dimensional feature data using a power prediction model to obtain a multi-dimensional feature vector, and to encode the multi-dimensional feature vector in different dimensions in turn to obtain a position encoding vector, a time encoding vector and a similarity encoding vector, wherein the position encoding vector is used to represent the position feature of each station, the time encoding vector is used to represent the time feature of the multi-dimensional feature data of each station, and the similarity encoding vector is used to represent the degree of similarity between multiple photovoltaic stations;

[0075] The prediction module 54 is used to analyze the position coding vector, the time coding vector and the similarity coding vector using a power prediction model to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic stations in the prediction period.

[0076] Through the above-mentioned photovoltaic combined power prediction device, by obtaining multi-dimensional feature data of each photovoltaic station, wherein the multi-dimensional feature data includes: historical power data, current meteorological data and weather forecast data; using a power prediction model to extract features from the multi-dimensional feature data to obtain a multi-dimensional feature vector, and encoding the multi-dimensional feature vector in different dimensions in turn to obtain a position coding vector, a time coding vector and a similarity coding vector, wherein the position coding vector is used to represent the position feature of each station, the time coding vector is used to represent the time feature of the multi-dimensional feature data of each station, and the similarity coding vector is used to represent the similarity between multiple photovoltaic stations; using a power prediction model to analyze the position coding vector, the time coding vector and the similarity coding vector to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic stations in the prediction period, thereby achieving the purpose of jointly predicting the coding vectors based on the position, time and similarity of the multi-dimensional features of the photovoltaic station, thereby achieving the technical effect of improving the accuracy of power prediction, and thus solving the technical problem of low prediction accuracy in related technologies.

[0077] The prediction module 54 includes: a prediction submodule, which is used to use a power prediction model to predict based on the position coding, time coding and similarity coding to obtain a prediction result, including: position coding the multi-dimensional feature vector to obtain the position coding vector; time coding the multi-dimensional feature vector to obtain the time coding vector; similarity coding the multi-dimensional feature vector to obtain the similarity coding vector; fusing the position coding vector, the time coding vector and the similarity coding vector to obtain a fused vector; and using a power prediction model to predict the fused vector to obtain the prediction result.

[0078] The prediction submodule includes: a first encoding unit, a second encoding unit and a third encoding unit, wherein the first encoding unit is used to perform position encoding on the multi-dimensional feature vector to obtain the position encoding vector, including: obtaining the position information of each photovoltaic field station and the total dimension of the position encoding vector; dividing the dimension of the position encoding into an even dimension and an odd dimension; when the dimension of the position encoding is an even dimension, determining the even dimension encoding value of the position encoding based on a sine function, the position information of each photovoltaic field station and the total dimension of the position encoding; when the dimension of the position encoding is an odd dimension, determining the odd dimension encoding value of the position encoding based on a cosine function, the position information of each photovoltaic field station and the total dimension of the position encoding vector; determining the position encoding vector based on the even dimension encoding value of the position encoding and the odd dimension encoding value of the position encoding.

[0079] The second encoding unit is used to perform time encoding on the multidimensional feature vector to obtain the time encoding vector, including: obtaining the time dimension of the multidimensional feature vector, the time dimension at least including: month, day, hour and minute, and obtaining the value interval of each time dimension; determining the embedding matrix of each time dimension based on the value interval of each time dimension, the number of columns of the embedding matrix is ​​the time dimension of the multidimensional feature vector, the number of rows of the embedding matrix is ​​the number of values ​​of each time dimension, and the elements of the embedding matrix represent the embedding vector corresponding to each time dimension value; obtaining the time feature of the multidimensional feature vector, and obtaining the embedding vector corresponding to the time feature from the embedding matrix; adding the embedding vectors of different time dimensions corresponding to the time feature to obtain the time encoding vector.

[0080] The third encoding unit is used to perform similarity encoding on the multi-dimensional feature vector to obtain the similarity encoding vector, including: determining the similarity of each photovoltaic station in each dimensional feature according to the multi-dimensional feature data of each photovoltaic station; determining the similarity between multiple stations based on the similarity of each photovoltaic station in each dimensional feature; encoding the feature representation of the similarity between the multiple stations to obtain the similarity encoding vector.

[0081] The third encoding unit includes: a similarity subunit and a fusion unit, wherein the similarity subunit is used to determine the similarity of multiple photovoltaic stations in each dimensional feature according to the multi-dimensional feature data of each photovoltaic station, including: determining the distance between any two photovoltaic stations according to the location information of each photovoltaic station; constructing a distance similarity matrix based on the distance between any two photovoltaic stations, wherein the distance similarity matrix is ​​used to represent the distance similarity between each photovoltaic station; extracting meteorological features from the multi-dimensional feature vector, wherein the meteorological features include: current meteorological features and historical meteorological features; performing similarity comparison on the meteorological features of each photovoltaic station to obtain a meteorological feature similarity matrix, wherein the meteorological feature similarity matrix is ​​used to represent the meteorological feature similarity between each photovoltaic station; extracting meteorological features from the multi-dimensional feature vector Weather forecast features are extracted, and the meteorological features of each photovoltaic station are compared for similarity to obtain a weather forecast feature similarity matrix, wherein the weather forecast feature similarity matrix is ​​used to represent the weather forecast feature similarity between each photovoltaic station; the meteorological feature similarity matrix and the weather forecast feature similarity matrix are determined as a meteorological similarity matrix; power features are extracted from the multi-dimensional feature vector, and the power features of each photovoltaic station are compared for similarity to obtain a power feature similarity matrix, wherein the power feature similarity matrix is ​​used to represent the power feature similarity between each photovoltaic station; weighted summation is performed on the distance similarity matrix, the meteorological similarity matrix and the power feature similarity matrix to determine a weighted similarity matrix, wherein the weighted similarity matrix represents the degree of similarity between the multiple stations.

[0082] A fusion unit is used to obtain the prediction result based on the fusion vector, including: encoding the fusion vector through the encoding layer in the preset power prediction model to obtain an initial feature vector; performing feature processing on the target feature vector through the fully connected layer in the preset power prediction model to obtain a target feature vector; and outputting the prediction result of the target feature vector through the output layer in the preset power prediction model.

[0083] It should be noted that Figure 5 The photovoltaic combined power prediction device shown is used to perform Figure 2 The photovoltaic combined power prediction method shown in the figure, therefore the relevant explanations in the above photovoltaic combined power prediction method are also applicable to the photovoltaic combined power prediction device, and will not be repeated here.

[0084] An embodiment of the present application further provides a computer device, comprising: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and is used to execute the above-mentioned photovoltaic combined power prediction method.

[0085] An embodiment of the present application further provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned photovoltaic combined power prediction method by running the computer program.

[0086] The embodiment of the present application further provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the photovoltaic combined power prediction method in the present application.

[0087] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0088] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0090] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0091] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0092] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a communication device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0093] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting photovoltaic combined power, characterized in that: include: Acquire multi-dimensional characteristic data of each photovoltaic station, wherein the multi-dimensional characteristic data includes: historical power data, current meteorological data and weather forecast data; A power prediction model is used to extract features from the multi-dimensional feature data to obtain a multi-dimensional feature vector, and the multi-dimensional feature vector is encoded in different dimensions in turn to obtain a position encoding vector, a time encoding vector and a similarity encoding vector, wherein the position encoding vector is used to represent the position feature of each station, the time encoding vector is used to represent the time feature of the multi-dimensional feature data of each station, and the similarity encoding vector is used to represent the degree of similarity between multiple photovoltaic stations; The position coding vector, the time coding vector and the similarity coding vector are analyzed by using a power prediction model to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic stations in the prediction period.

2. The method according to claim 1, characterized in that The power prediction model is used to predict the position code, time code and similarity code to obtain the prediction result, including: Performing position encoding on the multi-dimensional feature vector to obtain the position encoding vector; Performing time coding on the multi-dimensional feature vector to obtain the time coding vector; Performing similarity coding on the multi-dimensional feature vector to obtain the similarity coding vector; Fusing the position encoding vector, the time encoding vector and the similarity encoding vector to obtain a fused vector; The power prediction model is used to predict the fusion vector to obtain the prediction result.

3. The method according to claim 2, characterized in that Position encoding the multi-dimensional feature vector to obtain the position encoding vector includes: Obtaining the location information of each photovoltaic station and the total dimension of the location encoding vector; Dividing the dimension of the position encoding into an even dimension and an odd dimension; In the case where the dimension of the position code is an even dimension, determining an even-dimensional code value of the position code based on a sine function, the position information of each photovoltaic station and the total dimension of the position code; In the case where the dimension of the position code is an odd dimension, determining the odd-dimensional code value of the position code based on the cosine function, the position information of each photovoltaic station and the total dimension of the position code vector; The position encoding vector is determined based on the even-dimension encoding value of the position encoding and the odd-dimension encoding value of the position encoding.

4. The method according to claim 3, characterized in that Performing time coding on the multi-dimensional feature vector to obtain the time coding vector includes: Obtaining a time dimension of the multidimensional feature vector, where the time dimension includes at least: month, day, hour, and minute, and obtaining a value interval for each time dimension; Determine an embedding matrix for each time dimension based on the value interval of each time dimension, wherein the number of columns of the embedding matrix is ​​the time dimension of the multi-dimensional feature vector, the number of rows of the embedding matrix is ​​the number of values ​​of each time dimension, and the elements of the embedding matrix represent the embedding vector corresponding to each time dimension value; Obtaining a time feature of the multi-dimensional feature vector, and obtaining an embedding vector corresponding to the time feature from the embedding matrix; The embedding vectors of different time dimensions corresponding to the time feature are added together to obtain the time encoding vector.

5. The method according to claim 3, characterized in that: Performing similarity coding on the multi-dimensional feature vector to obtain the similarity coding vector includes: Determine the similarity of each photovoltaic station in each dimension based on the multi-dimensional feature data of each photovoltaic station; Determine the similarity between the multiple photovoltaic stations based on the similarity of the characteristics of each dimension of each photovoltaic station; The feature representation of the similarity between the multiple stations is encoded to obtain the similarity encoding vector.

6. The method according to claim 5, characterized in that The similarity of the characteristics of multiple photovoltaic stations in each dimension is determined based on the multi-dimensional characteristic data of each photovoltaic station, including: Determine the distance between any two photovoltaic stations according to the location information of each photovoltaic station; Constructing a distance similarity matrix based on the distance between any two photovoltaic stations, wherein the distance similarity matrix is ​​used to represent the distance similarity between each photovoltaic station; Extracting meteorological features from the multi-dimensional feature vector, the meteorological features including: current meteorological features and historical meteorological features; Comparing the meteorological characteristics of each photovoltaic station by similarity to obtain a meteorological characteristic similarity matrix, wherein the meteorological characteristic similarity matrix is ​​used to represent the meteorological characteristic similarity between each photovoltaic station; Extracting weather forecast features from the multi-dimensional feature vector, and performing similarity comparison on the meteorological features of each photovoltaic station to obtain a weather forecast feature similarity matrix, wherein the weather forecast feature similarity matrix is ​​used to represent the weather forecast feature similarity between each photovoltaic station; Determining the meteorological feature similarity matrix and the weather forecast feature similarity matrix as a meteorological similarity matrix; Extracting power features from the multi-dimensional feature vector, performing similarity comparison on the power features of each photovoltaic station to obtain a power feature similarity matrix, wherein the power feature similarity matrix is ​​used to represent the power feature similarity between each photovoltaic station; The distance similarity matrix, the meteorological similarity matrix and the power characteristic similarity matrix are weightedly summed to determine a weighted similarity matrix, wherein the weighted similarity matrix represents the degree of similarity between the multiple stations.

7. The method according to claim 6, characterized in that Obtaining the prediction result according to the fusion vector includes: Encoding the fused vector through a coding layer in a preset power prediction model to obtain an initial feature vector; Performing feature processing on the initial feature vector through a fully connected layer in the preset power prediction model to obtain a target feature vector; The prediction result of the target feature vector is outputted through the output layer in the preset power prediction model.

8. A photovoltaic combined power prediction device, characterized in that: include: An acquisition module is used to acquire multi-dimensional characteristic data of each photovoltaic station, wherein the multi-dimensional characteristic data includes: historical power data, current meteorological data and weather forecast data; A feature module, used to extract features from the multi-dimensional feature data using a power prediction model to obtain a multi-dimensional feature vector, and to sequentially encode the multi-dimensional feature vector in different dimensions to obtain a position encoding vector, a time encoding vector and a similarity encoding vector, wherein the position encoding vector is used to represent the position feature of each station, the time encoding vector is used to represent the time feature of the multi-dimensional feature data of each station, and the similarity encoding vector is used to represent the degree of similarity between multiple photovoltaic stations; The prediction module is used to analyze the position coding vector, the time coding vector and the similarity coding vector by using a power prediction model to obtain a prediction result, wherein the prediction result represents the power of the multiple photovoltaic stations in the prediction period.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and is used to execute the photovoltaic combined power prediction method as described in any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the photovoltaic combined power prediction method according to any one of claims 1 to 7 is implemented.

Images