Method and device for predicting wind-solar combined power and electronic equipment

Abstract

The invention discloses a wind-solar combined power prediction method and device and electronic equipment. The method comprises the following steps: acquiring wind power data corresponding to a wind power plant group, and acquiring photovoltaic data corresponding to a photovoltaic field group; and predicting the wind power data and the photovoltaic data by adopting the wind-solar combined power prediction model to obtain a wind power prediction result corresponding to the wind power plant group and a photovoltaic power prediction result corresponding to the photovoltaic power plant group. According to the method and the device, the technical problem of low power prediction accuracy caused by neglect of mutual influence between a wind power plant and a photovoltaic field during wind-solar power prediction in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of electric power technology, and in particular to a method, device and electronic equipment for predicting wind-solar combined power. Background Art

[0002] In the power system, wind power and photovoltaic power, as the main renewable energy, are indispensable components of the new power system. However, the output of wind power and photovoltaic power has the characteristics of randomness, volatility, and intermittency. The relevant technology ignores the mutual influence between wind farms and photovoltaic farms when predicting wind and solar power, resulting in low power prediction accuracy.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a method, device and electronic device for predicting combined wind and solar power, so as to at least solve the technical problem that the related art ignores the mutual influence between wind farms and photovoltaic fields when predicting wind and solar power, resulting in low power prediction accuracy.

[0005] According to one aspect of an embodiment of the present application, a method for predicting combined wind-solar power is provided, including: obtaining wind power data corresponding to a wind farm group, and obtaining photovoltaic data corresponding to a photovoltaic group; using a combined wind-solar power prediction model to predict the wind power data and the photovoltaic data, to obtain a wind power prediction result corresponding to the wind farm group and a photovoltaic power prediction result corresponding to the photovoltaic group, wherein the combined wind-solar power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module, the wind farm group prediction module is used to predict the power of all wind farm groups, the photovoltaic group prediction module is used to predict the power of all photovoltaic groups, and the combined wind-solar power prediction model is obtained by jointly training the historical wind power data of the wind farm group and the historical photovoltaic data corresponding to the photovoltaic group.

[0006] In some embodiments of the present application, the wind-solar combined power prediction model is trained in the following manner: obtaining historical wind power data of a wind farm group, and obtaining historical photovoltaic data corresponding to a photovoltaic farm group, wherein the historical wind power data includes wind power meteorological data, first historical power data of the wind farm group and first numerical weather forecast data, and the historical photovoltaic data includes photovoltaic meteorological data, second historical power data of the photovoltaic farm group and second numerical weather forecast data; using the original wind-solar combined power prediction model to predict the historical wind power data and the historical photovoltaic data to obtain a first historical prediction result and a second historical prediction result, wherein the first historical prediction result is used to represent the result obtained by predicting the historical wind power data, and the second historical prediction result is used to represent the result obtained by predicting the historical photovoltaic data; when the first historical prediction result and the second historical prediction result do not meet the preset conditions, adjusting the parameters of the original wind-solar combined power prediction model, and updating the first historical prediction result and the second historical prediction result to obtain an updated first historical prediction result and an updated second historical prediction result, until the updated first historical prediction result and the updated second historical prediction result meet the preset conditions, stopping updating the parameters to obtain the wind-solar combined power prediction model.

[0007] In some embodiments of the present application, the original wind-solar combined power prediction model is used to predict historical wind power data and historical photovoltaic data to obtain a first historical prediction result and a second historical prediction result, including: determining the historical wind power data and the historical photovoltaic data as input data, and using the original wind-solar combined power prediction model to determine the input feature vector of the input data; time-coding the input feature vector to obtain a time coding vector; position-coding the input feature vector to obtain a position coding vector; weighted similarity coding the input feature vector to obtain a similarity coding vector; fusing the time coding vector, the position coding vector and the similarity coding vector to obtain a fusion vector; and determining the first historical prediction result and the second historical prediction result based on the fusion vector.

[0008] In some embodiments of the present application, an input feature vector is time-encoded to obtain a time-encoded vector, including: obtaining a preset number of time dimensions, and obtaining a value range of each time dimension; determining an embedding matrix based on the time dimension and the value range, wherein the number of rows of the embedding matrix corresponds to the value range of each time dimension, the number of columns of the embedding matrix corresponds to the preset time dimension, and the row vector of the embedding matrix represents a feature representation of the time dimension value; obtaining the time information of the input feature vector, and obtaining an embedding vector corresponding to the time information from the embedding matrix, wherein the embedding vector is the sum of vectors of different time dimensions corresponding to the time information; determining the time-encoded vector based on the embedding vector and the input feature vector.

[0009] In some embodiments of the present application, position encoding is performed on an input feature vector to obtain a position encoding vector, including: obtaining position information corresponding to each wind farm group and each photovoltaic farm group; determining the total dimension of the position encoding vector; converting the position information into a position encoding value in the position encoding vector based on the total dimension, wherein when the dimension of the position encoding vector is an even number, a first position encoding value is generated based on a sine function, the position information and the total dimension, and when the dimension of the position encoding vector is an odd number, a second position encoding value is generated based on a cosine function, the position information and the total dimension; and the position encoding vector is determined based on the position encoding value.

[0010] In some embodiments of the present application, weighted similarity encoding is performed on the input feature vector to obtain a similarity encoding vector, including: obtaining the longitude and latitude coordinates of the wind farm group and the photovoltaic farm group; determining the distance similarity between any two farm groups based on the longitude and latitude coordinates, and determining the distance similarity matrix based on the distance similarity, wherein the any two farm groups are any two farm groups among all wind farm groups and all photovoltaic farm groups; extracting meteorological features from the input feature vector, wherein the meteorological features include features corresponding to the wind power meteorological data, features corresponding to the first numerical weather forecast data, features corresponding to the photovoltaic meteorological data, and features corresponding to the second numerical weather forecast data. corresponding features; determining the meteorological similarity between any two field groups based on the meteorological features, and determining the meteorological similarity matrix based on the meteorological similarity; extracting power features from the input feature vector, wherein the power features include features corresponding to the first historical power data and features corresponding to the second historical power data; determining the power similarity between any two field groups based on the power features, and determining the power similarity matrix based on the power similarity; performing weighted summation on the distance similarity matrix, the meteorological similarity matrix and the power similarity matrix to determine the weighted similarity matrix; fusing the weighted similarity matrix and the input feature vector to obtain a similarity coding vector.

[0011] In some embodiments of the present application, a first historical prediction result and a second historical prediction result are determined based on a fusion vector, including: encoding the fusion vector through the encoding layer in the original wind-solar combined power prediction model to obtain a first eigenvector; performing feature processing on the first eigenvector through the fully connected layer in the original wind-solar combined power prediction model to obtain a second eigenvector; processing the second eigenvector through the output layer in the original wind-solar combined power prediction model to obtain the first historical prediction result and the second historical prediction result.

[0012] According to another aspect of the embodiment of the present application, a wind-solar combined power prediction device is also provided, including: an acquisition module, used to obtain wind power data corresponding to a wind farm group, and to obtain photovoltaic data corresponding to a photovoltaic group; a prediction module, used to predict wind power data and photovoltaic data using a wind-solar combined power prediction model, to obtain wind power prediction results corresponding to the wind farm group and photovoltaic power prediction results corresponding to the photovoltaic group, wherein the wind-solar combined power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module, the wind farm group prediction module is used to predict the power of all wind farm groups, the photovoltaic group prediction module is used to predict the power of all photovoltaic groups, and the wind-solar combined power prediction model is obtained by jointly training the historical wind power data of the wind farm group and the historical photovoltaic data corresponding to the photovoltaic group.

[0013] According to another aspect of the embodiments of the present application, there is also provided an electronic device, including: a memory and a processor, the memory being used to store program instructions; the processor being connected to the memory and being used to execute the above-mentioned method for predicting the combined wind and solar power.

[0014] 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 wind-solar combined power prediction method by running the computer program.

[0015] 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 wind-solar combined power prediction method when executed by a processor.

[0016] In the embodiment of the present application, a wind-solar combined power prediction model is constructed and used to predict wind power data and photovoltaic data, thereby obtaining wind power prediction results corresponding to the wind farm group and photovoltaic power prediction results corresponding to the photovoltaic field group. This achieves the purpose of effectively capturing the mutual influence between the wind farm group and the photovoltaic field group and improving the accuracy of power prediction, thereby achieving the technical effect of improving the efficiency of power grid dispatch and reducing the operating cost of the power system, and further solves the technical problem that the related technology ignores the mutual influence between wind farms and photovoltaic fields when performing wind-solar power prediction, resulting in low power prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] 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:

[0018] Figure 1It is a hardware structure block diagram of a computer terminal of a method for predicting wind-solar combined power according to an embodiment of the present application;

[0019] Figure 2 is a flow chart of a method for predicting wind-solar combined power according to an embodiment of the present application;

[0020] Figure 3 It is an overall framework diagram of a method for predicting wind-solar combined power according to an embodiment of the present application;

[0021] Figure 4 It is a schematic diagram of the structure of a wind-solar combined power prediction model of a wind-solar combined power prediction method according to an embodiment of the present application;

[0022] Figure 5 It is a structural schematic diagram of a wind-solar combined power prediction device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] 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.

[0024] 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.

[0025] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0026] Numerical Weather Prediction (NWP) is a method of predicting future atmospheric motion and weather phenomena based on actual atmospheric conditions by using large computers to solve the fluid mechanics and thermodynamics equations that describe the weather evolution process.

[0027] Self-Attention Mechanism: The self-attention mechanism is a core component of the Transformer architecture. It allows the model to focus on elements at different positions in the sequence when processing sequence data, thereby capturing long-distance dependency features.

[0028] Wind and solar power forecasting is an important part of renewable energy power generation management and optimization, but related technologies often fail to fully consider the mutual influence between wind farms and photovoltaic stations when making forecasts. First, wind farms and photovoltaic stations are usually located in the same geographical area, which means that they may be affected by similar meteorological conditions. For example, a sudden storm may affect the power generation of multiple nearby wind farms and photovoltaic stations at the same time. However, related forecasting technologies often treat wind farms and photovoltaic stations independently, failing to fully utilize the possible geographical correlation between them, resulting in the forecast results failing to reflect this key factor in the actual scenario.

[0029] Secondly, although wind power and photovoltaic power generation are directly affected by weather conditions, there are significant differences in their responses to different weather conditions. For example, strong winds have a significant impact on the power generation of wind farms, while clear and cloudless weather has a positive impact on the power generation of photovoltaic stations. Since related technologies fail to combine the complementary characteristics of wind power and photovoltaics, the prediction model may overly rely on the historical data of a single energy station and fail to effectively capture the comprehensive impact of weather changes on the combined wind and solar power generation, resulting in a decrease in prediction accuracy.

[0030] In addition, the power output of wind power and photovoltaic power generation is not isolated, and there are direct or indirect interactions between them. For example, when the wind weakens, photovoltaic power may become the main source of electricity, and as the sun sets, photovoltaic power generation decreases, and wind power may become dominant. This mutual complementation and balance of power output is the key to the stable operation of the wind-solar combined power generation system. The relevant technology fails to consider their synergistic effect as a whole, resulting in a large deviation between the prediction results and the actual operation status of the wind-solar system.

[0031] In order to solve the above technical problems, the embodiments of the present application provide corresponding solutions, which are described in detail below.

[0032] The wind-solar 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 method for predicting wind-solar combined power is shown in FIG. 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 are shown.

[0033] 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).

[0034] 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 wind-solar 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 wind-solar 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 a combination thereof.

[0035] 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.

[0036] 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 .

[0037] 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.

[0038] Under the above-mentioned operating environment, an embodiment of the present application provides an embodiment of a method for predicting combined wind and solar 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.

[0039] Figure 2 is a flow chart of a method for predicting wind-solar combined power according to an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0040] Step S202, obtaining wind power data corresponding to the wind farm group, and obtaining photovoltaic data corresponding to the photovoltaic farm group.

[0041] In the above step S202, a wind farm group refers to a set of wind farms consisting of multiple wind farms that are geographically adjacent or have common meteorological conditions. The wind power data corresponding to the wind farm group is used to provide the power output characteristics of the wind farm at different time points and the meteorological condition information related thereto, which may include the historical power output records of the wind farm, the real-time meteorological data, and the numerical weather forecast data (NWP). In some embodiments of the present application, the historical power data of all wind farms in the wind farm group (including but not limited to the daily and hourly power generation records) can be collected, the real-time meteorological data of each wind farm (such as wind speed, temperature, humidity, etc.) can be obtained, and the NWP data (covering future weather forecasts, such as wind speed, cloud cover, precipitation probability, etc.) can be obtained.

[0042] A photovoltaic field group refers to a collection of photovoltaic fields consisting of multiple photovoltaic fields that are geographically adjacent or have common meteorological conditions. The photovoltaic data corresponding to the photovoltaic field group may include power output records of photovoltaic stations at past time points, real-time meteorological data, and numerical weather forecast data. Similar to wind power data, photovoltaic data provides a characteristic description of the power output of photovoltaic stations and related meteorological conditions for the wind-solar combined power prediction model. In some embodiments of the present application, historical power data can be collected from all photovoltaic stations in the photovoltaic field group, real-time meteorological data of the stations can be obtained (such as radiation intensity, temperature, humidity and other factors directly related to photovoltaic power generation), and NWP data can be obtained (such as predicting future radiation intensity, cloud cover and other key meteorological parameters that affect photovoltaic power output).

[0043] Step S204, use the wind-solar combined power prediction model to predict the wind power data and photovoltaic data to obtain the wind power prediction results corresponding to the wind farm group and the photovoltaic power prediction results corresponding to the photovoltaic group, wherein the wind-solar combined power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module, the wind farm group prediction module is used to predict the power of all wind farm groups, and the photovoltaic group prediction module is used to predict the power of all photovoltaic groups, and the wind-solar combined power prediction model is obtained by joint training of the historical wind power data of the wind farm group and the historical photovoltaic data corresponding to the photovoltaic group.

[0044] In the above step S204, the wind-solar combined power prediction model is used to simultaneously predict the power output of multiple wind farms and photovoltaic stations. The wind-solar combined power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module, which aims to utilize the complementarity and correlation of wind power and photovoltaic data, consider the mutual influence of wind power and photovoltaic data at different time points and meteorological conditions, learn and predict the power of wind power and photovoltaic, and improve the prediction accuracy.

[0045] The wind farm group prediction module is a component of the wind-solar combined power prediction model. It is used to process and predict the power data of the wind farm group. By analyzing the historical power output and related meteorological data of the wind farm group, the wind farm group prediction module can capture the intrinsic connection between wind farms and the law of how wind power is affected by weather changes.

[0046] The photovoltaic cluster prediction module is similar to the wind farm cluster prediction module and is used for power prediction of photovoltaic clusters. The photovoltaic cluster prediction module uses historical photovoltaic power data and meteorological data, especially indicators directly related to photovoltaic power such as radiation intensity, to learn the spatiotemporal correlation between photovoltaic stations and the response of power output to meteorological conditions.

[0047] In order to improve the generalization ability and prediction accuracy of the prediction model so as to be applicable to various weather conditions and seasonal changes, the wind-solar combined power prediction model can be trained in the following manner: obtaining historical wind power data of the wind farm group, and obtaining historical photovoltaic data corresponding to the photovoltaic farm group, wherein the historical wind power data includes wind power meteorological data, first historical power data of the wind farm group and first numerical weather forecast data, and the historical photovoltaic data includes photovoltaic meteorological data, second historical power data of the photovoltaic farm group and second numerical weather forecast data; using the original wind-solar combined power prediction model to predict the historical wind power data and the historical photovoltaic data to obtain the first historical prediction result and the second historical prediction result, wherein the first historical prediction result is used to represent the result obtained by predicting the historical wind power data, and the second historical prediction result is used to represent the result obtained by predicting the historical photovoltaic data; when the first historical prediction result and the second historical prediction result do not meet the preset conditions, adjusting the parameters of the original wind-solar combined power prediction model, and updating the first historical prediction result and the second historical prediction result to obtain the updated first historical prediction result and the updated second historical prediction result, until the updated first historical prediction result and the updated second historical prediction result meet the preset conditions, stopping updating the parameters to obtain the wind-solar combined power prediction model.

[0048] Historical wind power data and historical photovoltaic data can respectively include the power generation records of wind farm groups and photovoltaic farm groups at historical time points, actual meteorological conditions and numerical weather forecast information, which serve as the basic data for training the wind-solar combined power prediction model, so that the model can learn the generation laws of wind power and photovoltaic power and their response to meteorological conditions.

[0049] In some embodiments of the present application, historical wind power data include historical power data of wind farms (i.e., first historical power data), actual meteorological data of wind farms (i.e., wind power meteorological data), and numerical weather forecast data (i.e., first numerical weather forecast data), which are used to provide power output conditions of wind farm groups at different time points and meteorological conditions affecting power output, and are the direct basis for model learning of wind power prediction; historical photovoltaic data include historical power data of photovoltaic stations (i.e., second historical power data), actual meteorological data of photovoltaic stations (i.e., photovoltaic meteorological data), and numerical weather forecast data (i.e., second numerical weather forecast data), and photovoltaic data focuses on factors that directly affect photovoltaic power, such as radiation intensity and temperature, to help the model understand the dynamic changes of photovoltaic power.

[0050] In order to enable the prediction model to more comprehensively understand the characteristics of the input data, including changes in time series, the influence of geographical location and the interaction between different field groups, the input data can be encoded in different dimensions. Specifically: historical wind power data and historical photovoltaic data are determined as input data, and the original wind-solar combined power prediction model is used to determine the input feature vector of the input data; the input feature vector is time-encoded to obtain a time encoding vector; the input feature vector is position-encoded to obtain a position encoding vector; the input feature vector is weighted similarity-encoded to obtain a similarity encoding vector; the time encoding vector, the position encoding vector and the similarity encoding vector are fused to obtain a fused vector; based on the fused vector, the first historical prediction result and the second historical prediction result are determined.

[0051] In the training phase of the wind-solar combined power prediction model, the input feature vector refers to the conversion of historical wind power data and historical photovoltaic data into a numerical vector representation that the model can process. The input feature vector includes the historical power output records of each wind farm and photovoltaic station, the actual meteorological conditions (such as wind speed, radiation intensity, temperature, etc.) and numerical weather forecast (NWP) data, which provides the model with basic information for predicting wind power and photovoltaic power output.

[0052] Time coding is the process of encoding the time information in the input feature vector to enhance the model's understanding of time dependency. The time coding vector can comprehensively consider time dimensions such as minutes, hours, working days, days, and months, and add time features to the data at each time point. Through time coding, the model can capture the periodic changes and seasonal laws of power output and improve prediction accuracy. Especially when processing sequence data, time coding is the key to capturing long-term dependencies in the sequence. In some embodiments of the present application, a fixed embedding method can be used to construct an embedding layer of hours, working days, days, months, and minutes to encode the time information in the input feature vector, and each time dimension corresponds to a vector with a dimension of d_model (the feature dimension of the model). The time information can also be converted to a long integer, and then the coding vector of the corresponding time dimension is calculated. The coding vectors of all time dimensions are added to obtain the final time coding vector to ensure that each data point contains rich time information.

[0053] In order to enable the model to capture the periodic characteristics of time series data, the input feature vector can be time-encoded through the following steps: obtain a preset number of time dimensions and obtain the value range of each time dimension; determine the embedding matrix based on the time dimension and the value range, wherein the number of rows of the embedding matrix corresponds to the value range of each time dimension, the number of columns of the embedding matrix corresponds to the preset time dimension, and the row vector of the embedding matrix represents the feature representation of the time dimension value; obtain the time information of the input feature vector, and obtain the embedding vector corresponding to the time information from the embedding matrix, wherein the embedding vector is the sum of the vectors of different time dimensions corresponding to the time information; determine the time encoding vector based on the embedding vector and the input feature vector.

[0054] The time dimension refers to multiple time angles used to describe and analyze time series data, such as minutes, hours, weekdays, days, months, etc. In the wind and solar power prediction model, the introduction of multiple time dimensions enables the model to understand the characteristics of power output at different time scales, including short-term fluctuations and long-term trends, which helps the model to capture the time dependence of power more accurately. In some embodiments of the present application, according to the needs of wind and solar combined power prediction, the preset time dimensions may include: minutes, hours, weekdays, days and months, and the value ranges of each time dimension are: minutes 0-59, hours 0-23, weekdays 0-6 (0 means Sunday, and so on), days 1-31, and months 1-12. You can also use loops or list derivations to enumerate all possible values ​​of each time dimension and build a complete list of value ranges. These value ranges will be used to determine the number of rows when constructing the embedding matrix later.

[0055] The embedding matrix refers to a matrix used to convert categorical variables into continuous feature vectors, which converts discrete time information into continuous feature vectors so that the model can process this time information and learn the representation of time features from it. In the scenario of time coding, each row of the embedding matrix corresponds to a specific value of a time dimension, and each column represents a preset time dimension (such as minutes, hours, etc.). In some embodiments of the present application, a matrix with a dimension of ("total number of time dimension values", "preset number of time dimensions") can be initialized, where the value range of each time dimension determines the number of rows of the matrix, and the number of time dimensions (such as 5) determines the number of columns of the matrix. Random initialization or pre-training can also be used to assign values ​​to the embedding matrix. Random initialization is suitable for the initial stage of model training.

[0056] Time information refers to a specific time point or time range in the input feature vector, such as a record of a certain day and time of a certain wind farm or photovoltaic station. The embedded vector is a feature vector obtained by encoding the time information through the embedding matrix. In some embodiments of the present application, the time information in the input feature vector (such as 10:30 on April 15, 2023) can be mapped to the rows of the embedding matrix to obtain the embedded vectors related to minutes, hours, working days, days, and months, and then these vectors are summed to obtain the final time coding vector. It is also possible to use the enumeration mapping relationship of time information to convert time information into index values, so as to directly search and obtain the corresponding embedded vector from the embedding matrix, and the time coding vector is obtained by summing the vectors of each time dimension to ensure the multi-dimensional representation of time information.

[0057] The time coding vector is a vector obtained by summing the embedding vectors of multiple time dimensions, which integrates information of multiple time dimensions such as minutes, hours, weekdays, days, months, etc. In some embodiments of the present application, the time coding vector is obtained by summing the embedding vectors of multiple time dimensions obtained from the embedding matrix. Before summing, the embedding vector of each time dimension can be weighted, and the weight can be set based on experience or learned through a training process to reflect the difference in importance of different time dimensions in power prediction.

[0058] Position coding is used to provide location information to the model without destroying the input sequence, so that the model can learn the spatial correlation between wind farms and photovoltaic stations and understand the impact of different geographical locations on power output. In the process of wind and solar power prediction, the position coding vector is generated according to the geographical location information (such as longitude and latitude) of the wind farm and photovoltaic station to ensure that the model can understand the spatial distribution characteristics of the data. In some embodiments of the present application, the position coding vector can be generated according to the longitude and latitude of the station by sine and cosine functions, and the dimension of each position coding vector is also d_model (the characteristic dimension of the model).

[0059] In order to enable the model to understand the geographical relationship between different field groups, the input feature vector can be position-encoded through the following steps: obtain the location information corresponding to each wind farm group and each photovoltaic field group; determine the total dimension of the position coding vector; convert the location information into a position coding value in the position coding vector based on the total dimension, wherein when the dimension of the position coding vector is an even number, a first position coding value is generated based on the sine function, the location information and the total dimension; when the dimension of the position coding vector is an odd number, a second position coding value is generated based on the cosine function, the location information and the total dimension; determine the position coding vector based on the position coding value.

[0060] The location information refers to the geographical location information of each wind farm group (one wind farm group corresponds to one wind farm station) and photovoltaic group (one photovoltaic group corresponds to one photovoltaic station), which can be expressed in longitude and latitude coordinates. Through the location information, the model can consider the spatial distance and geographical location characteristics between different stations, such as altitude, terrain, etc., which is crucial to understanding how the geographical location of wind farms and photovoltaic stations affects their power output. In some embodiments of the present application, the longitude and latitude information of each site can be extracted from the database of wind farm groups and photovoltaic groups. In order to further enrich the amount of information encoded in the location, additional geographical location data, such as altitude, terrain type, etc., can also be collected.

[0061] The total dimension refers to the dimensional size of the position encoding vector. The total dimension determines the number of location features contained in the vector. The determination of the total dimension affects the degree of refinement of the geographic location information that the model can capture. In some embodiments of the present application, a specific total dimension is selected according to the experimental settings or model architecture requirements, such as d_model=512, which determines the length of the position encoding vector, that is, each location information will be converted into a 512-dimensional vector. The selection of the total dimension can also be based on the input vector dimension of the model to ensure that the position encoding vector can be consistent with the input feature vector in dimension, which is convenient for subsequent model processing.

[0062] The first position coding value refers to the position coding value calculated by the sine function when the dimension of the position coding vector is an even number. The first position coding value is combined with the position information (such as longitude and latitude) through the sine function to provide a position feature representation for the even positions in the position coding vector. The second position coding value refers to the position coding value calculated by the cosine function when the dimension of the position coding vector is an odd number. The second position coding value and the first position coding value generated by the sine function complement each other and together constitute a complete representation of the position coding vector.

[0063] In some embodiments of the present application, for each even dimension (i.e., the 2nd, 4th, 6th, ... dimension) in the position encoding vector, a sine function is used to calculate the position encoding value. The calculation formula is:

[0064]

[0065] Among them, pos represents the index position (starting from 0), i represents the dimension index (starting from 0), and d model Represents the total dimension of the position encoding vector.

[0066] For each odd dimension in the position encoding vector (i.e., the 1st, 3rd, 5th, ... dimensions), the cosine function is used to calculate the position encoding value. The calculation formula is:

[0067]

[0068] Among them, pos represents the index position (starting from 0), i represents the dimension index (starting from 0), and d model Represents the total dimension of the position encoding vector.

[0069] The position coding vector is a feature vector that encodes the location information of a specific wind farm or photovoltaic station. The position coding vector carries the feature representation of the geographic location information, which enables the model to fully consider the geographic location characteristics of each station when processing the wind-solar combined power prediction task. In some embodiments of the present application, the generated first position coding value and the second position coding value can be spliced ​​in dimensional order to form a complete position coding vector for representing the location characteristics of a specific wind farm or photovoltaic station.

[0070] Weighted similarity coding is a method that encodes the similarities between wind farms (such as latitude and longitude distance, meteorological conditions, power output) and photovoltaic farms (such as radiation intensity, temperature, power output) and the similarities between wind farms and photovoltaic farms into input features. Each similarity is adjusted by a corresponding weight to reflect its importance in the prediction process. By integrating the similarity information between wind farms and photovoltaic stations, the model can capture the mutual influence between different stations, thereby improving the accuracy of prediction in joint prediction tasks. Especially when cross-station data is involved, weighted similarity coding enables the model to use this information more effectively.

[0071] In the case where there are significant geographical and meteorological differences between different farm groups, in order to improve the accuracy of the prediction, the input feature vector can be weighted similarity encoded, specifically: the longitude and latitude coordinates of the wind farm group and the photovoltaic farm group are obtained; based on the longitude and latitude coordinates, the distance similarity between any two farm groups is determined, and the distance similarity matrix is ​​determined based on the distance similarity, wherein any two farm groups are any two farm groups among all wind farm groups and all photovoltaic farm groups; meteorological features are extracted from the input feature vector, wherein the meteorological features include features corresponding to the wind power meteorological data, features corresponding to the first numerical weather forecast data, features corresponding to the photovoltaic meteorological data, and features corresponding to the second numerical weather forecast data. The invention relates to a method for extracting power features from an input feature vector, wherein the power features include features corresponding to the first historical power data and features corresponding to the second historical power data; the power similarity between any two field groups is determined based on the power features, and a power similarity matrix is ​​determined based on the power similarity; the distance similarity matrix, the meteorological similarity matrix and the power similarity matrix are weightedly summed to determine a weighted similarity matrix; the weighted similarity matrix and the input feature vector are fused to obtain a similarity coding vector.

[0072] Distance similarity is the spatial distance calculated based on the longitude and latitude coordinates of any two stations in the wind farm group and the photovoltaic farm group, and then converted into a similarity measure through a formula. Distance similarity is used to reflect the geographical proximity between stations. Considering that stations in close geographical locations may be affected by similar meteorological conditions, distance similarity is an important indicator for evaluating the interaction between stations. In some embodiments of the present application, the great circle distance between any two stations can be calculated, and the distance value can be converted into a similarity measure using a geographic distance calculation method such as the Haversine formula. The smaller the distance, the greater the similarity measure. It is also possible to construct an N×N distance matrix (N is the total number of stations), record the distances between each pair of stations in the matrix, and then convert the distance matrix into a distance similarity matrix through a specific similarity conversion function (such as a Gaussian kernel function).

[0073] Meteorological similarity is a similarity measure obtained by comparing the meteorological characteristics of any two stations in the wind farm group and the photovoltaic farm group, and is used to construct a meteorological similarity matrix to reveal the similarities and differences in meteorological conditions at different stations. In some embodiments of the present application, the meteorological similarity between two stations can be calculated based on meteorological data such as wind speed, temperature, radiation intensity, etc., using cosine similarity or Euclidean distance to construct a meteorological similarity matrix. It is also possible to normalize the meteorological data vector, and use the vector inner product or distance measurement method to calculate the similarity of meteorological data between stations to obtain a meteorological similarity matrix.

[0074] Power similarity is the similarity of power output between any two stations in a wind farm group and a photovoltaic farm group calculated by analyzing power characteristics, and is used to construct a power similarity matrix to reveal the relationship between power outputs of different stations. In some embodiments of the present application, the historical power data of each station can be analyzed, the correlation coefficient or similarity index of the power sequence can be calculated, and a power similarity matrix can be constructed. Statistical methods (such as the Pearson correlation coefficient) or machine learning methods (such as similarity metrics) can also be used to evaluate the similarity of power output sequences between two stations, and a power similarity matrix can be constructed based on the results.

[0075] The weighted similarity matrix is ​​the result of fusing the distance similarity matrix, the meteorological similarity matrix, and the power similarity matrix according to specific weights. The weighted similarity matrix is ​​used to provide the model with rich inter-station relationship information, which helps the model to more comprehensively understand the inherent relationship of wind and solar power generation. In some embodiments of the present application, distance weights, meteorological weights, and power weights can be set, multiplied by the distance similarity matrix, the meteorological similarity matrix, and the power similarity matrix, respectively, and then the three matrices are summed to obtain a weighted similarity matrix. In order to make the set weights more reasonable, the weights can also be dynamically adjusted through cross-validation, grid search, or adaptive learning algorithms to find the best weighting scheme.

[0076] The similarity coding vector is the result of fusing the weighted similarity matrix with the input feature vector. It carries the coded information about the interaction between sites, enhances the information dimension of the input feature vector, and enables the model to more effectively use cross-site information for power prediction. In some embodiments of the present application, the weighted similarity matrix can be converted into a vector form that matches the dimension of the input feature vector, and then multiplied or added element by element with the input feature vector to obtain the similarity coding vector. An additional fusion layer can also be added before the input layer or encoding layer of the Transformer model to process the fusion of the weighted similarity matrix and the input feature vector. The attention mechanism in deep learning can be used for weighted fusion to ensure that the model can effectively adjust the input features according to the similarity information.

[0077] After obtaining the time encoding vector, the position encoding vector, and the similarity encoding vector, these vectors can be fused to obtain a fused vector. For example, the three vectors can be directly concatenated to form a longer vector, namely the fused vector. This concatenation operation allows the time, position, and similarity information to be stored in parallel in the vector, making it easier for the model to process this information simultaneously. You can also use a fully connected layer or a specific fusion layer (such as an attention mechanism) to perform weighted fusion on the three vectors, and construct a fused vector by learning the importance of each vector.

[0078] When dealing with large-scale, high-dimensional wind-solar combined power prediction problems, a multi-layer processing method can be used to extract deep features from the fused vector. Specifically: the fused vector is encoded through the encoding layer in the original wind-solar combined power prediction model to obtain the first feature vector; the first feature vector is feature processed through the fully connected layer in the original wind-solar combined power prediction model to obtain the second feature vector; the second feature vector is processed through the output layer in the original wind-solar combined power prediction model to obtain the first historical prediction result and the second historical prediction result.

[0079] In the wind-solar joint power prediction model, the encoding layer is responsible for deep feature extraction of the fusion vector, learning the complex interrelationships between input features through the multi-head self-attention mechanism, and further refining the features through the feedforward network to obtain the first feature vector. For example, through the self-attention mechanism, each element in the fusion vector is weighted and summed, and the weight coefficient is determined according to the correlation between the elements.

[0080] In the wind-solar joint power prediction model, the fully connected layer is responsible for further feature processing of the first eigenvector output by the coding layer to extract abstract features with more predictive value, namely the second eigenvector. In some embodiments of the present application, the fully connected layer performs a linear transformation through a weight matrix and a bias vector, and then introduces nonlinearity through an activation function (such as ReLU) to convert the first eigenvector into a second eigenvector. The number of neurons in the fully connected layer can be adjusted according to task requirements and data characteristics.

[0081] In the wind-solar combined power prediction model, the output layer is responsible for converting the second eigenvector into a specific first historical prediction result (power prediction of the wind farm group) and a second historical prediction result (power prediction of the photovoltaic group). In some embodiments of the present application, linear activation can be used to directly convert the second eigenvector into a power prediction result, which is suitable for regression prediction problems. The number of output nodes of the output layer matches the number of predicted wind farms and photovoltaic stations, and each node corresponds to the power prediction of a specific station. By adjusting the weight and bias of each node, the model can learn how to map the second eigenvector to the power prediction value of each station.

[0082] Preset conditions refer to specific standards or conditions used to evaluate the performance of the prediction model during the training process, such as the range of prediction errors, the stability of prediction results, etc. The preset conditions serve as the goal of model training to ensure that the model stops updating parameters when these conditions are met, avoid overfitting or underfitting, and obtain a prediction model with stable performance. In some embodiments of the present application, the prediction error is calculated by comparing the first historical prediction results and the second historical prediction results obtained by the model with the actual power data, and evaluating whether the model performance meets the preset conditions (for example, the root mean square error (RMSE) is less than a specific threshold).

[0083] If the prediction result does not meet the preset conditions, the model parameters are adjusted through optimization algorithms (such as Adam, SGD, etc.) in order to reduce the error. The preset conditions can be specific prediction error standards (such as RMSE, MAE, etc.), or thresholds for model parameter changes. The updated model re-predicts the historical data to obtain an updated first historical prediction result and an updated second historical prediction result. Repeat the evaluation and parameter adjustment process, and continuously update the prediction results until the model converges, that is, the updated prediction result is very small compared to the previous round of prediction results, or the prediction error is stable within the preset conditions. When the updated first historical prediction result and the updated second historical prediction result meet the preset conditions, stop parameter updating, and the model at this time is the wind-solar combined power prediction model.

[0084] Through the above steps S202 and S204, by constructing a wind-solar combined power prediction model, the wind power data and photovoltaic data are predicted using the wind-solar combined power prediction model, and the wind power prediction results corresponding to the wind farm group and the photovoltaic power prediction results corresponding to the photovoltaic field group are obtained, thereby achieving the purpose of effectively capturing the mutual influence between the wind farm group and the photovoltaic field group and improving the accuracy of power prediction, thereby achieving the technical effect of improving the efficiency of power grid dispatching and reducing the operating cost of the power system, and further solving the technical problem that the related technology ignores the mutual influence between wind farms and photovoltaic fields when predicting wind and solar power, resulting in low power prediction accuracy.

[0085] Figure 3 is an overall framework diagram of a method for predicting wind-solar combined power according to an embodiment of the present application. In some embodiments of the present application, the prediction of wind-solar combined power can perform the following steps, such as Figure 3 As shown:

[0086] Step S302: Start new energy power prediction. This step marks the official start of the wind-solar combined power prediction task. At this stage, the model and prediction system are initialized and ready to receive and process input data.

[0087] Step S304: The system collects the required data from the wind farm group and the photovoltaic farm group respectively, and performs preprocessing to ensure the consistency and availability of the data. The preprocessing step may include data cleaning, removing outliers, filling missing values, etc., to provide high-quality input data for subsequent model prediction. Among them, step S204 includes S3041 and S3042. S3041 corresponds to obtaining wind power data corresponding to the wind farm group, including wind power real-time meteorological data (such as wind speed, wind direction, temperature and other real-time meteorological information), station historical power data (recording the power output of the wind farm station in the past period of time), and numerical weather forecast data NWP (predicting meteorological information within a certain time range in the future, used for auxiliary power prediction); S3042 corresponds to obtaining photovoltaic data corresponding to the photovoltaic farm group, including photovoltaic real-time meteorological data (such as light intensity, temperature, humidity and other real-time meteorological information), station historical power data (past power output records of photovoltaic stations), and NWP data (same as wind farms, providing future meteorological forecasts for photovoltaic power prediction).

[0088] Step S306: Prediction based on the Transformer wind-solar joint power prediction model with multi-dimensional coding fusion. Step S306 is the core of the entire wind-solar joint power prediction process. At this stage, the wind power and photovoltaic data collected and preprocessed in S304 are input into the Transformer model based on multi-dimensional coding fusion for joint prediction. The specific implementation process includes:

[0089] The real-time meteorological data, historical power data, and NWP data of wind power and photovoltaic power are input into the model; the data is pre-encoded through time coding, location coding, and weighted similarity coding to enhance the time, location, and similarity information between stations of the data; the encoded data enters the Transformer encoding layer, and the self-attention mechanism and multi-head attention mechanism are used to capture the complex dependencies between data; the features of the encoding layer enter the fully connected layer, and are further refined and integrated to generate more abstract feature representations; finally, through the multi-task output layer, the power prediction values ​​of the wind farm group and the photovoltaic farm group are generated respectively. In this process, the model gradually extracts and reconstructs deep features from the original data through multi-dimensional feature coding and multi-layer processing of deep learning algorithms (such as Transformer), providing a solid data foundation for the final power prediction.

[0090] Step S308: Based on the results of the previous multi-dimensional coding fusion Transformer model processing, the system outputs specific wind power and photovoltaic power forecast values. Among them, step S308 includes S3081 and S3082. S3081 corresponds to the output of wind power forecast results, and obtains the power forecast value of the wind farm group within the forecast period; S3082 corresponds to the output of photovoltaic forecast results, and obtains the power forecast value of the photovoltaic farm group within the same forecast period. These forecast values ​​will be used in scenarios such as power system scheduling, energy management, and market transactions to help decision makers better understand and respond to the volatility of wind and solar power generation.

[0091] The entire process, from starting power prediction (S302) to data collection (S304), to feature encoding and model prediction (S306), and finally outputting the prediction results (S308), is a complete processing chain from shallow to deep, from data to prediction. Each step is closely linked, and together constitutes the Transformer wind-solar joint power prediction method based on multi-dimensional coding fusion, which aims to improve the accuracy and reliability of wind-solar power prediction.

[0092] Figure 4 is a schematic diagram of a wind-solar combined power prediction model structure of a wind-solar combined power prediction method according to an embodiment of the present application, such as Figure 4 As shown, in some embodiments of the present application, the wind-solar combined power prediction model structure may include the following modules:

[0093] Module S402: Input layer. The input layer is responsible for receiving data from S304 ( Figure 3 The input layer can also perform preliminary preprocessing on the data, such as data normalization and data format conversion, to ensure that the data can be correctly interpreted and processed by the model.

[0094] Module S404: Embedding layer. The embedding layer is a key step in processing sequence data. It converts the high-dimensional feature vector received in the input layer into a low-dimensional dense vector representation, namely the input feature vector. This step is crucial to improving the efficiency and effectiveness of the model in processing sequence data. For example, mapping discrete category features (such as station ID) to continuous vector representations facilitates the model to learn the potential relationship between these features. It achieves dimensionality reduction processing of input data, reduces the amount of calculation and improves the efficiency of model training.

[0095] Module S406: Feature coding layer. The feature coding layer further pre-encodes the data to enhance the model's understanding of the data features. It consists of three sub-modules: S4061 weighted similarity coding, which calculates the similarity between each station, including longitude and latitude distance similarity, meteorological similarity, and power similarity. Through weighted fusion, these similarity information are encoded into the feature vectors of the corresponding stations to form a vector representation that contains the interaction between stations. For example, distance formulas (such as the Haversine formula) can be used to calculate the geographical distance between stations, as well as to measure the similarity of meteorological and power data (such as cosine similarity); S4062 time coding, time coding adds time information to the data at each time point, including minutes, hours, weekdays, days, and months. Through time embedding, the model can better learn the law of power changes over time. For example, a fixed time embedding function can be used to convert time information into part of the feature vector; S4063 position coding, position coding uses sine and cosine functions to generate a position coding vector for each station, and integrates geographic location information into the feature vector to ensure that the model can distinguish the locations of different stations and understand the impact of location on power output.

[0096] Module S408: Similarity calculation, this module specifically decomposes the similarity calculation steps in S4061 weighted similarity coding, including: S4081 latitude and longitude distance similarity, that is, by calculating the geographical distance between stations and converting these distances into similarity indicators. For example, after using a distance formula such as the Haversine formula to calculate the distance, it is converted into a similarity value through a Gaussian kernel function; S4082 meteorological similarity, that is, analyzing the similarity of meteorological conditions at different stations, similarity metrics (such as cosine similarity or Euclidean distance) can be used to calculate the similarity of actual meteorological and NWP data, and construct a meteorological similarity matrix; S4083 power similarity, that is, by analyzing historical power data, calculating the correlation or similarity metric between power sequences, and forming a power similarity matrix.

[0097] Module S410: Feature fusion. The feature fusion module fuses the feature vectors obtained by S4061, S4062 and S4063 to form a more comprehensive and richer feature representation. For example, different encoding vectors can be fused by concatenation, multiplication or addition, or a specific fusion layer (such as an attention mechanism) can be used to learn the importance of different encoding vectors, so as to combine this information more effectively.

[0098] Module S412: Transformer encoding layer. The Transformer encoding layer uses self-attention mechanism and multi-head attention mechanism to process the fused feature vector in S410, capture the long-distance dependencies in the data, and generate deeper feature representations. For example, the input sequence can be weighted summed first through the self-attention mechanism, and then the features can be further refined through the feedforward neural network, and finally residual connection and layer normalization can be added to stabilize the model training.

[0099] Module S414: Fully connected layer. The fully connected layer receives the features output by the Transformer encoding layer, and further refines and integrates the features through the fully connected operation of the neural network to make them more in line with the feature representation requirements of the power prediction task. For example, the fully connected layer performs linear transformation and nonlinear activation (such as ReLU activation function) to reduce redundant information and focus on key features.

[0100] Module S416: Multi-station power output layer. The multi-station power output layer generates power prediction results for wind power and photovoltaic stations based on the features processed by the fully connected layer S414. For example, it can include multiple output nodes, each node corresponds to the power prediction value of a specific station. By adjusting the weights and biases, the model learns how to map features to power prediction values, and finally outputs the joint power prediction results of the wind farm group and the photovoltaic farm group.

[0101] The design and implementation of the entire model, from data input to feature encoding, to deep feature extraction (Transformer encoding layer), and finally to power prediction output, follows the processing flow of the deep learning model and fully considers the multi-dimensional characteristics of wind and solar power generation to achieve more accurate and reliable wind-solar combined power prediction.

[0102] Figure 5 is a structural diagram of a wind-solar combined power prediction device according to an embodiment of the present application, such as Figure 5 As shown, the device comprises:

[0103] An acquisition module 502 is used to acquire wind power data corresponding to a wind farm group, and to acquire photovoltaic data corresponding to a photovoltaic farm group;

[0104] Prediction module 504 is used to predict wind power data and photovoltaic data using a wind-solar combined power prediction model to obtain wind power prediction results corresponding to the wind farm group and photovoltaic power prediction results corresponding to the photovoltaic group. The wind-solar combined power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module. The wind farm group prediction module is used to predict the power of all wind farm groups, and the photovoltaic group prediction module is used to predict the power of all photovoltaic groups. The wind-solar combined power prediction model is obtained by jointly training the historical wind power data of the wind farm group and the historical photovoltaic data corresponding to the photovoltaic group.

[0105] In the above-mentioned wind-solar combined power prediction device, the prediction module is also used to obtain historical wind power data of the wind farm group, and to obtain historical photovoltaic data corresponding to the photovoltaic farm group, wherein the historical wind power data includes wind power meteorological data, first historical power data of the wind farm group and first numerical weather forecast data, and the historical photovoltaic data includes photovoltaic meteorological data, second historical power data of the photovoltaic farm group and second numerical weather forecast data; the original wind-solar combined power prediction model is used to predict the historical wind power data and the historical photovoltaic data to obtain the first historical prediction result and the second historical prediction result, wherein the first historical prediction result is used to represent the result obtained by predicting the historical wind power data, and the second historical prediction result is used to represent the result obtained by predicting the historical photovoltaic data; when the first historical prediction result and the second historical prediction result do not meet the preset conditions, the parameters of the original wind-solar combined power prediction model are adjusted, and the first historical prediction result and the second historical prediction result are updated to obtain the updated first historical prediction result and the updated second historical prediction result, until the updated first historical prediction result and the updated second historical prediction result meet the preset conditions, the parameter updating is stopped to obtain the wind-solar combined power prediction model.

[0106] In the above-mentioned wind-solar combined power prediction device, the prediction module is also used to determine historical wind power data and historical photovoltaic data as input data, and use the original wind-solar combined power prediction model to determine the input feature vector of the input data; time-encode the input feature vector to obtain a time-encoded vector; position-encode the input feature vector to obtain a position-encoded vector; weighted similarity-encode the input feature vector to obtain a similarity-encoded vector; fuse the time-encoded vector, the position-encoded vector and the similarity-encoded vector to obtain a fused vector; and determine the first historical prediction result and the second historical prediction result based on the fused vector.

[0107] In the above-mentioned wind-solar combined power prediction device, the prediction module is also used to obtain a preset number of time dimensions and obtain the value range of each time dimension; determine the embedding matrix based on the time dimension and the value range, wherein the number of rows of the embedding matrix corresponds to the value range of each time dimension, the number of columns of the embedding matrix corresponds to the preset time dimension, and the row vector of the embedding matrix represents the characteristic representation of the time dimension value; obtain the time information of the input feature vector, and obtain the embedding vector corresponding to the time information from the embedding matrix, wherein the embedding vector is the sum of the vectors of different time dimensions corresponding to the time information; determine the time coding vector based on the embedding vector and the input feature vector.

[0108] In the above-mentioned wind-solar combined power prediction device, the prediction module is also used to obtain the position information corresponding to each wind farm group and each photovoltaic farm group; determine the total dimension of the position coding vector; convert the position information into a position coding value in the position coding vector based on the total dimension, wherein when the dimension of the position coding vector is an even number, a first position coding value is generated based on the sine function, the position information and the total dimension, and when the dimension of the position coding vector is an odd number, a second position coding value is generated based on the cosine function, the position information and the total dimension; and determine the position coding vector based on the position coding value.

[0109] In the above-mentioned wind-solar combined power prediction device, the prediction module is also used to obtain the longitude and latitude coordinates of the wind farm group and the photovoltaic farm group; determine the distance similarity between any two farm groups based on the longitude and latitude coordinates, and determine the distance similarity matrix based on the distance similarity, wherein any two farm groups are any two farm groups among all wind farm groups and all photovoltaic farm groups; extract meteorological features from the input feature vector, wherein the meteorological features include features corresponding to wind power meteorological data, features corresponding to the first numerical weather forecast data, features corresponding to photovoltaic meteorological data, and features corresponding to the second numerical weather forecast data; determine the meteorological similarity between any two farm groups based on the meteorological features, and determine the meteorological similarity matrix based on the meteorological similarity; extract power features from the input feature vector, wherein the power features include features corresponding to the first historical power data and features corresponding to the second historical power data; determine the power similarity between any two farm groups based on the power features, and determine the power similarity matrix based on the power similarity; perform weighted summation on the distance similarity matrix, the meteorological similarity matrix, and the power similarity matrix to determine the weighted similarity matrix; fuse the weighted similarity matrix and the input feature vector to obtain a similarity coding vector.

[0110] In the above-mentioned wind-solar combined power prediction device, the prediction module is also used to encode the fusion vector through the encoding layer in the original wind-solar combined power prediction model to obtain a first eigenvector; perform feature processing on the first eigenvector through the fully connected layer in the original wind-solar combined power prediction model to obtain a second eigenvector; process the second eigenvector through the output layer in the original wind-solar combined power prediction model to obtain a first historical prediction result and a second historical prediction result.

[0111] It should be noted that Figure 5 The wind-solar combined power prediction device shown is used to perform Figure 2 The wind-solar combined power prediction method shown in the figure is Figure 2 The relevant explanations in the wind-solar combined power prediction method in the Figure 5 The wind-solar combined power prediction device shown will not be described in detail here.

[0112] An embodiment of the present application also provides an electronic device, which includes a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the steps of the wind-solar combined power prediction method in each embodiment of the present application.

[0113] For example, the processor performs the following functions by executing program instructions stored in the memory:

[0114] Obtain wind power data corresponding to the wind farm group, and obtain photovoltaic data corresponding to the photovoltaic group; use the wind-solar combined power prediction model to predict the wind power data and photovoltaic data, and obtain the wind power prediction results corresponding to the wind farm group and the photovoltaic power prediction results corresponding to the photovoltaic group, wherein the wind-solar combined power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module, the wind farm group prediction module is used to predict the power of all wind farm groups, and the photovoltaic group prediction module is used to predict the power of all photovoltaic groups, and the wind-solar combined power prediction model is obtained by jointly training the historical wind power data of the wind farm group and the historical photovoltaic data corresponding to the photovoltaic group.

[0115] 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 steps of the wind-solar combined power prediction method in each embodiment of the present application by running the computer program.

[0116] An 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 wind-solar combined power prediction method in each embodiment of the present application.

[0117] The embodiments of the present application also provide a computer program, which, when executed by a processor, implements the steps of the wind-solar combined power prediction method in each embodiment of the present application.

[0118] 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.

[0119] 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.

[0120] 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.

[0121] 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.

[0122] 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.

[0123] 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 computer 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.

[0124] 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 wind-solar combined power, characterized in that: include: Obtain wind power data corresponding to the wind farm group, and obtain photovoltaic data corresponding to the photovoltaic farm group; A wind-solar combined power prediction model is used to predict the wind power data and the photovoltaic data to obtain wind power prediction results corresponding to the wind farm group and photovoltaic power prediction results corresponding to the photovoltaic group, wherein the wind-solar combined power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module, the wind farm group prediction module is used to predict the power of all wind farm groups, and the photovoltaic group prediction module is used to predict the power of all photovoltaic groups, and the wind-solar combined power prediction model is obtained by joint training of historical wind power data of the wind farm group and historical photovoltaic data corresponding to the photovoltaic group.

2. The method according to claim 1, characterized in that The wind-solar combined power prediction model is trained in the following way: Acquire historical wind power data of the wind farm group, and acquire historical photovoltaic data corresponding to the photovoltaic farm group, wherein the historical wind power data includes wind power meteorological data, first historical power data of the wind farm group, and first numerical weather forecast data, and the historical photovoltaic data includes photovoltaic meteorological data, second historical power data of the photovoltaic farm group, and second numerical weather forecast data; The historical wind power data and the historical photovoltaic data are predicted using the original wind-solar combined power prediction model to obtain a first historical prediction result and a second historical prediction result, wherein the first historical prediction result is used to represent a result obtained by predicting the historical wind power data, and the second historical prediction result is used to represent a result obtained by predicting the historical photovoltaic data; When the first historical prediction result and the second historical prediction result do not meet the preset conditions, adjust the parameters of the original wind-solar combined power prediction model, and update the first historical prediction result and the second historical prediction result to obtain updated first historical prediction results and updated second historical prediction results. Stop updating parameters when the updated first historical prediction result and the updated second historical prediction result meet the preset conditions to obtain the wind-solar combined power prediction model.

3. The method according to claim 2, characterized in that The historical wind power data and the historical photovoltaic data are predicted using the original wind-solar combined power prediction model to obtain a first historical prediction result and a second historical prediction result, including: Determine the historical wind power data and the historical photovoltaic data as input data, and use the original wind-solar combined power prediction model to determine the input feature vector of the input data; Performing time coding on the input feature vector to obtain a time coding vector; Performing position encoding on the input feature vector to obtain a position encoding vector; Performing weighted similarity coding on the input feature vector to obtain a similarity coding vector; fusing the time coding vector, the position coding vector and the similarity coding vector to obtain a fused vector; The first historical prediction result and the second historical prediction result are determined based on the fusion vector.

4. The method according to claim 3, characterized in that The input feature vector is time-coded to obtain a time-coded vector, including: Get a preset number of time dimensions and the value range of each time dimension; Determine an embedding matrix according to the time dimension and the value range, wherein the number of rows of the embedding matrix corresponds to the value range of each time dimension, the number of columns of the embedding matrix corresponds to the preset time dimension, and the row vector of the embedding matrix represents a feature representation of the time dimension value; Acquire time information of the input feature vector, and acquire an embedding vector corresponding to the time information from the embedding matrix, wherein the embedding vector is the sum of vectors of different time dimensions corresponding to the time information; The temporal encoding vector is determined according to the embedding vector and the input feature vector.

5. The method according to claim 3, characterized in that: Position encoding is performed on the input feature vector to obtain a position encoding vector, including: Obtaining the location information corresponding to each wind farm group and each photovoltaic farm group; Determining the total dimension of the position encoding vector; Converting the position information into a position coding value in a position coding vector according to the total dimension, wherein when the dimension of the position coding vector is an even number, generating a first position coding value according to a sine function, the position information and the total dimension, and generating a second position coding value according to a cosine function, the position information and the total dimension when the dimension of the position coding vector is an odd number; The position encoding vector is determined according to the position encoding value.

6. The method according to claim 3, characterized in that The input feature vector is subjected to weighted similarity coding to obtain a similarity coding vector, including: Obtaining the longitude and latitude coordinates of the wind farm group and the photovoltaic farm group; Determine the distance similarity between any two farm groups according to the longitude and latitude coordinates, and determine a distance similarity matrix according to the distance similarity, wherein the any two farm groups are any two farm groups among all wind farm groups and all photovoltaic farm groups; Extracting meteorological features from the input feature vector, wherein the meteorological features include features corresponding to the wind power meteorological data, features corresponding to the first numerical weather forecast data, features corresponding to the photovoltaic meteorological data, and features corresponding to the second numerical weather forecast data; Determining the meteorological similarity between any two field groups based on the meteorological characteristics, and determining a meteorological similarity matrix based on the meteorological similarity; Extracting a power feature from the input feature vector, wherein the power feature includes a feature corresponding to the first historical power data and a feature corresponding to the second historical power data; Determining power similarity between any two field groups according to the power characteristics, and determining a power similarity matrix according to the power similarity; Performing weighted summation on the distance similarity matrix, the meteorological similarity matrix, and the power similarity matrix to determine a weighted similarity matrix; The weighted similarity matrix and the input feature vector are fused to obtain the similarity coding vector.

7. The method according to claim 3, characterized in that Determining the first historical prediction result and the second historical prediction result according to the fusion vector includes: The fusion vector is encoded by the encoding layer in the original wind-solar joint power prediction model to obtain a first eigenvector; Performing feature processing on the first feature vector through a fully connected layer in the original wind-solar combined power prediction model to obtain a second feature vector; The second feature vector is processed through the output layer in the original wind-solar combined power prediction model to obtain the first historical prediction result and the second historical prediction result.

8. A wind-solar combined power prediction device, characterized in that: include: An acquisition module is used to acquire wind power data corresponding to a wind farm group, and to acquire photovoltaic data corresponding to a photovoltaic farm group; A prediction module is used to predict the wind power data and the photovoltaic data using a wind-solar combined power prediction model to obtain wind power prediction results corresponding to the wind farm group and photovoltaic power prediction results corresponding to the photovoltaic group, wherein the wind-solar combined power prediction model includes a wind farm group prediction module and a photovoltaic group prediction module, the wind farm group prediction module is used to predict the power of all wind farm groups, and the photovoltaic group prediction module is used to predict the power of all photovoltaic groups, and the wind-solar combined power prediction model is obtained by jointly training the historical wind power data of the wind farm group and the historical photovoltaic data corresponding to the photovoltaic group.

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

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the wind-solar combined power prediction method as described in any one of claims 1 to 7 by running the computer program.

11. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by the processor, the method for predicting the wind-solar combined power as described in any one of claims 1 to 7 is implemented.

Images