Photovoltaic ultra-short-term power prediction method and system based on multi-source cloud picture data and deep learning

Abstract

The invention belongs to the technical field of photovoltaic power station power prediction, and discloses a photovoltaic ultra-short-term power prediction method and system based on multi-source cloud picture data and deep learning, and the method comprises the steps: collecting a photovoltaic field station foundation cloud picture, a photovoltaic field station satellite cloud picture and NWP data; preprocessing the photovoltaic station foundation cloud picture, and inputting the processed image into an AE cloud picture feature extraction model to obtain foundation cloud cluster features; preprocessing the photovoltaic station satellite cloud picture, and inputting the processed image into an AE cloud picture feature extraction model to obtain satellite cloud picture features; respectively inputting the ground-based cloud cluster features, the satellite cloud picture features and the NWP data into an LSTM layer to obtain a hidden vector A, a hidden vector B and a hidden vector C; and the hidden vector A, the hidden vector B and the hidden vector C are input into a full connection layer for fusion, and a photovoltaic ultra-short-term power prediction result is output, so that accurate, timely and refined cloud cluster information is provided for photovoltaic ultra-short-term power prediction, and the ultra-short-term power prediction precision is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power station power prediction, and relates to a photovoltaic ultra-short-term power prediction method and system based on multi-source cloud image data and deep learning. Background Art

[0002] Photovoltaic power generation, with its advantages of sufficient supply, green and clean, and wide distribution, is a key way to achieve sustainable energy development. Accurate ultra-short-term photovoltaic power forecast results can provide a reliable basis for grid optimization and dispatching, which is of great significance for ensuring the safety and stability of the grid. Environmental changes will affect the output power of photovoltaic power stations, among which the shading of photovoltaic power stations by randomly moving clouds is the main reason for the drastic changes in photovoltaic output. At present, there are three main technical means for ultra-short-term prediction of photovoltaic power generation at home and abroad.

[0003] The first is based on historical data mining, and commonly used methods include continuous prediction, autoregressive moving average, and neural network. Wavelet decomposition and other methods are used to decompose the irradiance data into sequences and learn sequence features, and then the deep learning algorithm is combined to construct a prediction model. Compared with traditional time series, this method has better prediction performance, but because it does not consider the influence of meteorological factors such as temperature and cloud cover, the model only realizes trend prediction and is difficult to apply to complex and sudden weather.

[0004] The second is to combine it with NWP, combining characteristic variables such as temperature and relative humidity in NWP with power data, and using physical models or statistical models to dig out the mapping relationship between various meteorological elements and power, so as to establish a power prediction model. The prediction accuracy and generalization ability of such models are easily affected by the robustness of the algorithm and the accuracy of NWP data. For cloudy and sudden process weather, the location, size, thickness, etc. of clouds are prone to significant changes in a short time and small range. However, it is difficult for the current NWP to capture it accurately, so the prediction results under this technical route still have a lot of room for improvement. Among them, NWP is a numerical weather forecast.

[0005] The third method is combined with cloud observation information. The first is power prediction based on all-sky imager data. The limitation of this method is mainly that the viewing angle of the all-sky imager is limited. When the high-altitude clouds move quickly, it is impossible to monitor the cloud information several hours later, which is not enough to support ultra-short-term predictions at the hourly level. Although the all-sky imager monitoring network can be used to overcome this problem, it will face the problems of high engineering application cost and difficult maintenance. The second is prediction based on satellite cloud observation data. Geostationary satellites can monitor cloud movement in real time over a large range, and the acquired cloud observation data can be applied to power prediction of multiple stations at the same time. However, since the model generally only models cloud maps at historical moments, the current cloud movement information is not fully utilized, and the main idea of ​​using cloud map information is to predict the cloud movement trajectory through cloud maps, and to specifically model the solar irradiance received by the photovoltaic power station in combination with information such as solar altitude angle and solar azimuth angle. The model is often complex and not easy to promote, and the actual prediction accuracy still needs to be improved. Summary of the invention

[0006] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a photovoltaic ultra-short-term power prediction method and system based on multi-source cloud image data and deep learning, to extract features of ground-based cloud images, and to construct low-dimensional abstract feature vectors suitable for photovoltaic power prediction models; to extract satellite cloud image features from a satellite cloud image feature extraction model, and to comprehensively extract satellite cloud image features; to fuse ground-based cloud image features, satellite cloud image features and NWP, and to perform ultra-short-term power prediction of photovoltaic stations based on an LSTM-MLP model.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: The present invention provides a photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning, comprising: collecting photovoltaic station ground-based cloud images, photovoltaic station satellite cloud images, and NWP data; preprocessing the photovoltaic station ground-based cloud images, and inputting the processed images into an AE cloud image feature extraction model to obtain ground-based cloud cluster features; preprocessing the photovoltaic station satellite cloud images, and inputting the processed images into an AE cloud image feature extraction model to obtain satellite cloud image features; inputting the ground-based cloud cluster features, satellite cloud image features, and NWP data into an LSTM layer to obtain hidden vectors A, hidden vectors B, and hidden vectors C respectively; and inputting the hidden vectors A, B, and C into a fully connected layer for fusion to output a photovoltaic ultra-short-term power prediction result.

[0008] Furthermore, the AE cloud image feature extraction model is composed of an encoder and a decoder, the encoder is composed of a convolutional layer and a pooling layer stacked in sequence, and the decoder is composed of a transposed convolutional layer and a convolutional layer stacked in sequence.

[0009] Furthermore, the photovoltaic station foundation cloud image is preprocessed, and the processed image is input into the AE cloud image feature extraction model to obtain the foundation cloud cluster features, including: image cropping, binarization, and grayscale processing of the photovoltaic station foundation cloud image, and the grayscale photovoltaic station foundation cloud image is input into the AE cloud image feature extraction model to obtain an output mapping image, which converts the high-dimensional complex image information into a low-dimensional abstract feature vector, and the output mapping image is the foundation cloud cluster features.

[0010] Furthermore, the encoder performs downsampling processing on the input feature map, and the decoder performs upsampling processing on the input feature map.

[0011] Furthermore, the satellite cloud image of the photovoltaic station is preprocessed, and the processed image is input into an AE cloud image feature extraction model to obtain satellite cloud image features, including: the AE cloud image feature extraction model extracts the CDF time series of the satellite cloud image, and the CDF time series of the satellite cloud image is integrated with the photovoltaic power generation power time series to obtain the satellite cloud image features.

[0012] Furthermore, the photovoltaic station foundation cloud map also includes collecting the irradiation data and power generation data of the photovoltaic station before preprocessing.

[0013] Furthermore, the output map is:

[0014] in, For the Output map, is the activation function, For the Layer convolution kernels, For the The bias value of the layer, for The corresponding input.

[0015] The present invention also provides a photovoltaic ultra-short-term power prediction method system based on multi-source cloud image data and deep learning, including: a collection module, used to collect photovoltaic station ground-based cloud images, photovoltaic station satellite cloud images, and NWP data; a first preprocessing module, used to preprocess the photovoltaic station ground-based cloud images, and the processed images are input into an AE cloud image feature extraction model to obtain ground-based cloud cluster features; a second preprocessing module, used to preprocess the photovoltaic station satellite cloud images, and the processed images are input into an AE cloud image feature extraction model to obtain satellite cloud image features; an operation module, used to input the ground-based cloud cluster features, satellite cloud image features, and NWP data into an LSTM layer respectively to obtain hidden vectors A, hidden vectors B, and hidden vectors C; an output module: used to input the hidden vectors A, B, and C into a fully connected layer for fusion and output of photovoltaic ultra-short-term power prediction results.

[0016] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned photovoltaic ultra-short-term power prediction method based on multi-source cloud map data and deep learning are implemented.

[0017] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned photovoltaic ultra-short-term power prediction method based on multi-source cloud map data and deep learning.

[0018] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention discloses a photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning. The AE cloud features are used to extract features of ground-based cloud images and satellite cloud images. While accurately and efficiently extracting rich spatiotemporal features of the cloud layer, the training time is effectively reduced.

[0019] The present invention provides a photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning, which integrates ground-based cloud image features, satellite cloud image features and NWP data, thereby providing accurate, timely and refined cloud information for photovoltaic ultra-short-term power prediction, effectively improving the accuracy of ultra-short-term power prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning in the present invention; Figure 2 This is a structural diagram of the AE cloud image feature extraction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0022] Example 1 A photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning, such as Figure 1 As shown, the following steps are included: collecting photovoltaic station ground cloud images, photovoltaic station satellite cloud images, and NWP data; preprocessing the photovoltaic station ground cloud images, and inputting the processed images into the AE cloud image feature extraction model to obtain ground cloud cluster features; preprocessing the photovoltaic station satellite cloud images, and inputting the processed images into the AE cloud image feature extraction model to obtain satellite cloud image features; inputting the ground cloud cluster features, satellite cloud image features, and NWP data into the LSTM layer to obtain hidden vectors A, B, and C; and inputting the hidden vectors A, B, and C into the fully connected layer for fusion and outputting photovoltaic ultra-short-term power prediction results. Wherein, AE is an autoencoder.

[0023] Specifically, we first collect the irradiation data, ground cloud map and power generation of the photovoltaic station. The irradiation data contains information such as the intensity, direction and changes of solar radiation at different times. For photovoltaic stations, solar irradiation is one of the most critical factors affecting photovoltaic power generation. It directly determines how much solar energy the photovoltaic panel can receive, which in turn affects the power generation. The ground cloud map can intuitively present the state of the cloud, and for photovoltaic stations, clouds are an important factor affecting the solar radiation reaching the photovoltaic panel. Large areas of stratus clouds may greatly reduce the direct light reaching the photovoltaic panel, while thin cirrus clouds may have a relatively small effect on the amount of radiation. Power generation is a direct indicator to measure the actual power generation effect of photovoltaic stations. By collecting and analyzing the power generation data, combining the power generation with irradiation data, ground cloud maps, etc., we can deeply explore the influence of various factors on power generation.

[0024] Then, we collect satellite cloud map data of the area where the photovoltaic station is located, and finally collect NWP data of the photovoltaic station, where NWP is numerical weather forecast. NWP data contains weather conditions information for a period of time in the future, such as wind speed, temperature, humidity, sunshine duration, etc., which are key factors affecting photovoltaic power generation and can more accurately estimate the power generation of photovoltaic stations at different times in the future.

[0025] The photovoltaic station foundation cloud image is first cropped, binarized, and grayed, and then input into the AE cloud image feature extraction model to extract the cloud distribution features and convert the high-latitude complex image information into a low-dimensional abstract feature vector. Figure 2 As shown in the figure, the AE cloud image feature extraction model consists of an encoder and a decoder. The encoder is composed of a convolution layer and a pooling layer stacked in sequence, and the decoder is composed of a transposed convolution layer and a convolution layer stacked in sequence. When the ground-based cloud image is input into the AE cloud image feature extraction model, the encoder compresses the high-dimensional image information into a low-dimensional CDF, which is the cloud cluster distribution feature. The decoder uses CDF to reconstruct the image information and outputs the CDF time series, which is fused with the irradiation time series to obtain the ground-based cloud cluster feature, that is, the output mapping diagram. The output map is:

[0026] in, For the Output map, is the activation function, For the Layer convolution kernels, For the The bias value of the layer, for The corresponding input.

[0027] In this embodiment, the convolution layer is a three-dimensional convolution layer, and the 2D convolution kernel has two moving directions. After performing a 2D convolution operation on multiple frames, a 2D feature map will be generated; while the 3D convolution kernel has three moving directions, and after performing a 3D convolution operation on multiple frames, a 3D feature map will be generated. Under the same conditions, compared with 2D convolution, the operation of 3D convolution is more sophisticated, more complex, and the corresponding operation time is longer.

[0028] The satellite cloud images of photovoltaic stations are preprocessed, and the processed images are input into the AE cloud image feature extraction model to obtain satellite cloud image features. Specifically, the satellite cloud images are downsampled and compressed, and the downsampled satellite cloud images are input into the AE cloud image feature extraction model to extract the CDF time series of the satellite cloud images, and the CDF time series of the satellite cloud images is fused with the photovoltaic power generation time series to obtain the satellite cloud cluster features.

[0029] The ground-based cloud cluster features are input into the LSTM layer to obtain the hidden layer vector A, the satellite cloud cluster features are input into the LSTM layer to obtain the hidden layer vector B, the NWP data is input into the LSTM layer to obtain the hidden layer vector C, the hidden layer vector A, the hidden layer vector B, and the hidden layer vector C are fused and input into the fully connected layer to output the photovoltaic ultra-short-term power prediction results. LSTM is long short-term memory.

[0030] Among them, the ground-based cloud cluster features are input into the LSTM layer to obtain the hidden layer vector A, which includes the following steps: feature extraction of ground-based clouds. These features may include the shape of the cloud (such as block, layer or flocculent, etc.), size (which can be measured by indicators such as area, length, width, etc.), height (approximate height range from the ground), color (grayscale value or specific color features in color images), texture (uniform or complex texture changes, etc.) and dynamic features such as the movement speed and direction of the cloud. These multi-dimensional features are quantified and organized into a format suitable for input into the neural network, which can usually be expressed in vector form. For example, each set of cloud cluster features corresponds to a feature vector ,in represents a specific eigenvalue, is the number of features The value ranges of different features may vary greatly. The area of ​​a cloud may be hundreds of square meters, while the grayscale value of a color is between 0 and 255. To prevent certain features from having a dominant or adverse effect on training due to excessively large or small values, the data is generally normalized. Common normalization methods include minimum-maximum normalization, which maps feature values ​​to The interval is ,in is the original eigenvalue, and are the minimum and maximum values ​​of the feature respectively. As all cloud feature vectors are input in sequence and processed by the LSTM layer, the hidden state output at the last time step or the last LSTM layer (if it is a multi-layer structure) is hidden vector A. Hidden vector B is similar to hidden vector A.

[0031] Example 2 The present invention discloses a photovoltaic ultra-short-term power prediction method system based on multi-source cloud image data and deep learning, comprising a collection module, a first preprocessing module, a second preprocessing module, an operation module and an output module.

[0032] A collection module is used to collect ground-based cloud images of photovoltaic stations, satellite cloud images of photovoltaic stations, and NWP data; a first preprocessing module is used to preprocess the ground-based cloud images of photovoltaic stations, and the processed images are input into the AE cloud image feature extraction model to obtain ground-based cloud cluster features; a second preprocessing module is used to preprocess the satellite cloud images of photovoltaic stations, and the processed images are input into the AE cloud image feature extraction model to obtain satellite cloud image features; an operation module is used to input the ground-based cloud cluster features, satellite cloud image features, and NWP data into the LSTM layer respectively to obtain hidden vectors A, B, and C; an output module: used to input the hidden vectors A, B, and C into the fully connected layer for fusion and output of photovoltaic ultra-short-term power prediction results.

[0033] Example 3 The present invention provides a terminal device, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the photovoltaic ultra-short-term power prediction method based on multi-source cloud map data and deep learning.

[0034] Example 4 The present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0035] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning in the above embodiment.

[0036] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0037] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0038] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0039] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning, characterized in that: The following steps are involved: Collect PV station foundation cloud images, PV station satellite cloud images, and NWP data; The photovoltaic station foundation cloud image is preprocessed, and the processed image is input into the AE cloud image feature extraction model to obtain the foundation cloud cluster features; The photovoltaic station satellite cloud image is preprocessed, and the processed image is input into the AE cloud image feature extraction model to obtain the satellite cloud image features; The ground-based cloud cluster features, satellite cloud image features, and NWP data are respectively input into the LSTM layer to obtain hidden vector A, hidden vector B, and hidden vector C; The hidden vector A, the hidden vector B, and the hidden vector C are input into the fully connected layer for fusion and output as the photovoltaic ultra-short-term power prediction result.

2. The photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning according to claim 1 is characterized in that: The AE cloud image feature extraction model consists of an encoder and a decoder, the encoder is composed of a convolution layer and a pooling layer stacked in sequence, and the decoder is composed of a transposed convolution layer and a convolution layer stacked in sequence.

3. The photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning according to claim 1 is characterized in that: The photovoltaic station foundation cloud image is preprocessed, and the processed image is input into the AE cloud image feature extraction model to obtain the foundation cloud cluster features, including: The photovoltaic station foundation cloud image is subjected to image cropping, binarization and grayscale processing. The grayscale photovoltaic station foundation cloud image is input into the AE cloud image feature extraction model to obtain an output mapping image, which converts the high-dimensional complex image information into a low-dimensional abstract feature vector. The output mapping image is the foundation cloud cluster feature.

4. The photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning according to claim 2 is characterized in that: The encoder performs downsampling processing on the input feature map, and the decoder performs upsampling processing on the input feature map.

5. The photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning according to claim 1 is characterized in that: The satellite cloud image of the photovoltaic station is preprocessed, and the processed image is input into the AE cloud image feature extraction model to obtain the satellite cloud image features, including: The AE cloud image feature extraction model extracts the CDF time series of the satellite cloud image, and the CDF time series of the satellite cloud image is fused with the photovoltaic power generation time series to obtain the satellite cloud image features.

6. The photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning according to claim 1 is characterized in that: The photovoltaic station foundation cloud map also includes collecting the irradiation data and power generation data of the photovoltaic station before preprocessing.

7. The photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning according to claim 3 is characterized by: The output map is: in, For the Output map, is the activation function, For the Layer convolution kernels, For the The bias value of the layer, for The corresponding input.

8. A photovoltaic ultra-short-term power prediction method system based on multi-source cloud image data and deep learning, characterized in that: include: The collection module is used to collect photovoltaic station foundation cloud images, photovoltaic station satellite cloud images, and NWP data; The first preprocessing module is used to preprocess the photovoltaic station foundation cloud image, and the processed image is input into the AE cloud image feature extraction model to obtain the foundation cloud cluster features; The second preprocessing module is used to preprocess the satellite cloud image of the photovoltaic station, and the processed image is input into the AE cloud image feature extraction model to obtain the satellite cloud image features; An operation module, used for inputting the ground-based cloud cluster features, satellite cloud image features, and NWP data into the LSTM layer to obtain hidden vector A, hidden vector B, and hidden vector C respectively; Output module: used to input the hidden vector A, hidden vector B, and hidden vector C into the fully connected layer for fusion and output of photovoltaic ultra-short-term power prediction results.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the photovoltaic ultra-short-term power prediction method based on multi-source cloud image data and deep learning are implemented as described in any one of claims 1 to 7.

Images