Photovoltaic power generation prediction method and device based on multi-input model

Abstract

The present invention relates to the technical field of new energy power generation prediction, and discloses a photovoltaic power generation prediction method and device based on a multi-input model. The method pre-processes the cloud map data and meteorological data corresponding to photovoltaics; the multi-input model is constructed, and the pre-processed cloud map data and meteorological data are respectively input into two different feature processing branch networks to obtain corresponding features, and the features are fused through a feature fusion layer, and the fused features are processed into new features through a shared hidden layer, and the input output layer is used to predict photovoltaic power generation; the multi-input model is trained to obtain a trained model as a prediction model; the cloud map data and meteorological data corresponding to the photovoltaics to be predicted are input into the prediction model after pre-processing to obtain a prediction result. Compared with the traditional method, the present invention has better generalization performance and prediction accuracy, can more accurately predict photovoltaic power generation, and improve the practicality of the prediction.

Description

Technical Field

[0001] The present invention relates to the field of new energy power generation prediction, and in particular to a photovoltaic power generation power prediction method and device based on a multi-input model. Background Art

[0002] Photovoltaic power prediction method refers to predicting the size of photovoltaic power generation in the future by analyzing and modeling the historical data of photovoltaic power stations, meteorological data, etc. At present, the main photovoltaic power prediction methods include the following:

[0003] 1. Machine learning-based methods: By collecting a large amount of historical data and weather data, a prediction model is constructed using machine learning algorithms, such as support vector machine (SVM), artificial neural network (ANN) and decision tree algorithms to predict photovoltaic power generation. Machine learning algorithms can effectively improve the accuracy of photovoltaic power generation prediction.

[0004] 2. Methods based on deep learning: In recent years, deep learning algorithms have been widely used in photovoltaic power generation prediction. Researchers build deep neural network models to learn the rules in historical data, thereby predicting future photovoltaic power generation. For example, models such as convolutional neural networks (CNN), recurrent neural networks (RNN) and long short-term memory networks (LSTM) are used to predict photovoltaic power generation. Deep learning algorithms can effectively improve the accuracy and stability of photovoltaic power generation prediction.

[0005] 3. Methods based on statistical models: Methods based on statistical models have always been one of the main methods for photovoltaic power generation prediction. Researchers analyze and process historical data, use regression analysis, time series analysis and other methods to establish a prediction model, and use the model to predict future photovoltaic power generation. Statistical models can improve the accuracy of photovoltaic power generation prediction to a certain extent.

[0006] 4. Methods based on hybrid models: In recent years, researchers have also begun to explore photovoltaic power prediction methods based on hybrid models. This method mainly combines multiple prediction models and uses their advantages to improve prediction accuracy. For example, the prediction results based on statistical models and machine learning models are weighted combined, or the prediction results based on deep learning and machine learning are fused. Hybrid models can further improve the accuracy and robustness of photovoltaic power prediction.

[0007] The existing technologies still have the following problems or shortcomings in photovoltaic power generation prediction:

[0008] 1. The method based on meteorological monitoring data only uses meteorological parameters to predict the accuracy of photovoltaic power generation, which is greatly affected by weather factors such as cloud cover, and the prediction accuracy is not accurate enough;

[0009] 2. The image recognition-based method requires high-level image analysis technology and computing resources. In practical applications, it is difficult to obtain and process the monitored image data.

[0010] 3. Most of the existing prediction models use statistical methods or machine learning algorithms, and lack the understanding and model building of the intrinsic physical mechanism of photovoltaic power generation systems, resulting in weak interpretability of the models. At the same time, in new data scenarios, the models need to be retrained, resulting in poor model robustness. Summary of the invention

[0011] In view of this, the present invention provides a photovoltaic power prediction method, device, computer equipment and storage medium based on a multi-input model to solve the problems of poor model robustness and low prediction accuracy in the prior art.

[0012] In a first aspect, the present invention provides a photovoltaic power generation prediction method based on a multi-input model, comprising:

[0013] Pre-process the cloud map data and meteorological data corresponding to photovoltaics respectively;

[0014] The constructed multi-input model includes: two input modules, two feature processing branch networks, a feature fusion layer, a shared hidden layer and an output layer. The pre-processed cloud image data and meteorological data are respectively input into two different feature processing branch networks through the two input modules for processing to obtain the features corresponding to the meteorological data and the cloud image data. The two types of features are fused through the feature fusion layer. The fused features are processed through several layers of neural network through the shared hidden layer to obtain new features, which are then input into the output layer for photovoltaic power generation prediction.

[0015] Train the constructed multi-input model and obtain the trained model as a prediction model;

[0016] The cloud map data and meteorological data corresponding to the photovoltaic power to be predicted are input for preprocessing and then input into the prediction model to obtain the corresponding photovoltaic power generation power prediction result.

[0017] The photovoltaic power generation prediction method based on a multi-input model provided in this embodiment can use two different types of data, cloud map data and meteorological data, to predict photovoltaic power generation. Compared with the traditional method based solely on meteorological monitoring data or cloud map data, it can more accurately predict photovoltaic power generation and improve the practicability of the prediction.

[0018] In an optional embodiment, the cloud map data is processed using a first CNN network branch, which includes: multiple convolution modules, a residual connection structure and a multi-scale enhancement module. The multi-scale enhancement module is parasitic between two independent convolution modules through a residual connection structure, and uses void convolution internally to enrich the multi-scale representation of features.

[0019] In an optional implementation, the parameters involved in the multi-scale enhancement module include: a dilation rate factor and the number of convolution layers, and different dilation convolutions are obtained by using different dilation rates for the input two-dimensional feature map.

[0020] The multi-scale enhancement module uses a 32-dimensional 3x3 convolution kernel to process the input cloud map data. The back end is connected to multiple hole convolution kernels for extracting multi-dimensional features. The multi-dimensional features extracted by the hole convolution are merged, and the 1x1 convolution kernel is used to perform dimensional transformation to output enhanced features, thereby obtaining a multi-channel two-dimensional feature matrix.

[0021] In an optional embodiment, the meteorological data is processed in a second CNN network branch, and the processing process includes: processing the meteorological data through multiple layers of one-dimensional convolution, then expanding the features into a one-dimensional vector, transforming the number of feature points through a fully connected layer, and finally converting it into a multi-channel two-dimensional matrix with the same scale as the output features of the first CNN network branch.

[0022] In an optional embodiment, a joint loss function is established to train the multi-input model, and a weighted average loss function is used to train the network.

[0023] In an optional implementation, different data sets and loss functions are used alternately for training during the training process, and the process includes:

[0024] The meteorological data and cloud image data are respectively input into the corresponding network branches for separate training to obtain the pre-trained network weights;

[0025] Data is input into the two networks at the same time, and a joint loss function is added for end-to-end joint training. The loss functions in the two networks are calculated separately during each training, and the parameters of the two networks are updated separately according to their weights.

[0026] In a second aspect, the present invention provides a photovoltaic power generation prediction device based on a multi-input model, the device comprising:

[0027] The data preprocessing module is used to preprocess the cloud map data and meteorological data corresponding to photovoltaics respectively;

[0028] A multi-input model building module is used to build a multi-input model, which includes: two input modules, two feature processing branch networks, a feature fusion layer, a shared hidden layer and an output layer. The pre-processed cloud image data and meteorological data are respectively input into two different feature processing branch networks through the two input modules for processing to obtain the features corresponding to the meteorological data and the cloud image data, and the two types of features are fused through the feature fusion layer. The fused features are processed by several layers of neural network through the shared hidden layer to obtain new features, and the new features are input into the output layer for photovoltaic power generation prediction;

[0029] The model training module is used to train the constructed multi-input model and obtain the trained model as a prediction model;

[0030] The power prediction module is used to input the cloud map data and meteorological data corresponding to the photovoltaic power to be predicted, and after preprocessing, input them into the prediction model to obtain the corresponding photovoltaic power prediction result.

[0031] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the photovoltaic power prediction method based on a multi-input model of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the photovoltaic power prediction method based on a multi-input model of the above-mentioned first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 is a schematic flow chart of a photovoltaic power generation prediction method based on a multi-input model according to an embodiment of the present invention;

[0035] Figure 2 is a schematic diagram of a multi-input model architecture provided by an embodiment of the present invention;

[0036] Figure 3 is a schematic diagram of a network structure of a first CNN network branch according to an embodiment of the present invention;

[0037] Figure 4 is a structural block diagram of a photovoltaic power prediction device based on a multi-input model according to an embodiment of the present invention;

[0038] Figure 5 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0040] According to an embodiment of the present invention, an embodiment of a photovoltaic power generation power prediction method based on a multi-input model is provided. 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 a 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.

[0041] In this embodiment, a photovoltaic power generation prediction method based on a multi-input model is provided, which can be used for a computer device terminal. Figure 1 is a flow chart of a photovoltaic power prediction method based on a multi-input model according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0042] Step S101, pre-processing the cloud map data and meteorological data corresponding to photovoltaics respectively.

[0043] For each input data type, corresponding preprocessing is performed, including data cleaning, missing value processing, feature engineering, data standardization, etc. Then the original data set is divided into a training set and a test set. The details are as follows:

[0044] 1. Data cleaning: First, the collected data needs to be cleaned to remove duplicate values, invalid values, and error values ​​to ensure the quality and accuracy of the data.

[0045] 2. Missing value processing: Since missing values ​​may appear in the process of data collection, it is necessary to process missing values. Processing methods include: deleting samples containing missing values, filling missing values ​​with statistics such as mean or median, and using interpolation methods. Specifically: for a feature x, if there are n missing values, the formula for filling using the mean filling method is:

[0046]

[0047] 3. Feature engineering: By performing feature engineering on the collected meteorological data, characteristic parameters related to photovoltaic power generation are extracted, including irradiance, temperature, humidity, wind speed, wind direction, etc., and processed into feature vectors.

[0048] 4. Data standardization: Since different features have different units and magnitudes, it will affect the learning effect of the model. Therefore, the data needs to be standardized so that the values ​​of different features are in the same magnitude. The standardization formula is: where x i is the i-th feature, u is the sample mean, and σ is the sample standard deviation.

[0049] 5. Data division: Before model training, the collected data set needs to be divided into training set, validation set and test set. The training set is used for model learning and parameter adjustment, the validation set is used for model selection and adjustment, and the test set is used to evaluate the generalization ability and prediction effect of the model.

[0050] Training set = {x 1 ,x 2 ,...,x m}

[0051] Validation set = {x m+1 ,x m+2 ,...,x m+n}

[0052] Test set = {x m+n+1 ,x m+n+2 ,...,x m+n+p}

[0053] Among them, m is the number of samples in the training set, n is the number of samples in the validation set, and p is the number of samples in the test set.

[0054] In the embodiment of the present invention, the data set division is to divide the generated data set into a training set, a validation set and a test set. The training set accounts for 80% of the total data set, the validation set accounts for 10% of the total data set, and the test set accounts for 10% of the total data set, which is only an example and is not limited to this.

[0055] Step S102, constructing a multi-input model, which includes: two input modules, two feature processing branch networks, a feature fusion layer, a shared hidden layer and an output layer, the preprocessed cloud image data and meteorological data are respectively input into two different feature processing branch networks through the two input modules for processing, to obtain the features corresponding to the meteorological data and the cloud image data, and the two types of features are fused through the feature fusion layer, the fused features are processed through several layers of neural network through the shared hidden layer to obtain new features, and the new features are input into the output layer for photovoltaic power generation prediction.

[0056] The input of the multi-input model constructed in the embodiment of the present invention consists of two parts, namely meteorological data (including numerical information such as irradiance, temperature, wind speed, humidity, air pressure, etc.) and cloud image data (image data taken by satellite). In the specific implementation, the meteorological data and cloud image data are respectively input into two different neural network branches for processing to obtain their feature representations respectively; then the features of the two aspects are fused, and the shared hidden layer (the embodiment of the present invention uses ResNet as a shared network model) is responsible for processing the fused features through several layers of neural networks to obtain a more comprehensive feature representation, and send it to the output layer for power prediction. Figure 2 Shown is a multi-input model architecture constructed according to an embodiment of the present invention.

[0057] In an optional embodiment, if Figure 3 As shown, the cloud map data is processed by the first CNN network branch (CNN branch 1), and the first CNN network branch includes: multiple convolution modules, residual connection structure and multi-scale enhancement module SEM, wherein each convolution module contains convolution, normalization, activation, and pooling steps; in order to make full use of the many local and global information contained in the cloud map data, the embodiment of the present invention establishes a residual connection structure and a multi-scale enhancement module for multi-scale processing and enhancement of features, promotes the forward propagation and fusion of features, and effectively improves the utilization rate of features. The multi-scale enhancement module parasitizes between two independent convolution modules through a residual connection structure, and uses a hole convolution inside to enrich the multi-scale representation of features.

[0058] Specifically, Figure 3 As shown in the figure, the multi-scale enhancement module uses a 32-dimensional 3x3 convolution kernel to process the input cloud image data. The back end is connected to multiple dilated convolution kernels to extract multi-dimensional features. The multi-dimensional features extracted by dilated convolution are merged, and the 1x1 convolution kernel is used to transform the dimension and output the enhanced features to obtain a multi-channel two-dimensional feature matrix. There are two parameters involved in the multi-scale enhancement module, namely the expansion rate factor r and the expansion rate factor r. 0And the number of convolution layers K, different expansion rates can be used for the input two-dimensional feature map to obtain different dilated convolutions. This processing method can extract features with different local receptive fields for the same feature, thereby improving feature utilization. After the original cloud map data is processed by CNN branch 1, a multi-channel two-dimensional feature matrix is ​​finally obtained.

[0059] Since cloud map data is two-dimensional image data, while meteorological data is not two-dimensional, feature fusion requires the unification of feature scales. In the second CNN network branch (CNN branch 2) in the embodiment of the present invention, numerical meteorological data is first processed through multiple layers of one-dimensional convolution, and then the features are expanded into one-dimensional vectors. The number of feature points is then transformed through a fully connected layer, and finally converted into a multi-channel two-dimensional matrix with the same scale as the output features of CNN branch 1.

[0060] The feature outputs of the two branches are merged together through the channel fusion layer (Concatenate) to form a new feature map, which contains the abstract features of the two input data at the same time. Next, the fused features will be passed to the shared ResNet model for feature extraction and fusion. The shared ResNet model is a convolutional neural network with a deep hierarchical structure and residual connections. It can help the model extract higher-level and abstract features and avoid training problems such as gradient disappearance. Finally, the features obtained after processing by the shared ResNet model will be input into the fully connected layer for prediction output. Each neuron in the fully connected layer is associated with all input features. The weights and biases obtained through training can map the input features to the prediction result space.

[0061] Step S103, training the constructed multi-input model to obtain a trained model as a prediction model.

[0062] The present invention establishes a joint loss function to train the model and adopts a weighted average loss function to train the network. Specifically, the model is composed of two inputs, x1 and x2, and the corresponding loss functions are L1 and L2. The form of the weighted average loss function is: L = w 1 L 1 +w 2 L 2 , w1 and w2 respectively represent the weights of the loss, and they should be set reasonably according to actual needs.

[0063] During the training process, an alternating training method is adopted, that is, different data sets and loss functions are used alternately for training to achieve the effect of joint training. Specifically, the meteorological data and cloud map data can be input into the corresponding network branches for separate training to obtain the pre-trained network weights. Then the two networks are input with data at the same time, and the joint loss function is added for end-to-end joint training. In each training, the loss functions in the two networks can be calculated separately, and the parameters of the two networks can be updated according to their weights.

[0064] In a specific embodiment, the multi-input model established in the above method is built using the open source framework Pytorch, a loss function is established, the preprocessed training set is input into the built multi-input model, the model is trained using the training set, and the model parameters are continuously adjusted through the back propagation algorithm to minimize the loss function value of the model. In the training stage, two different network branches are pre-trained by alternating training, and then the entire network is trained using the joint loss function. During the training process, the learning rate is continuously adjusted until the loss value is no longer significantly reduced. At the same time, the model prediction accuracy after each round of iteration is verified by using the validation set to prevent the model from overfitting.

[0065] After training, the trained model is evaluated using the test set to calculate the values ​​of various evaluation indicators. Based on the evaluation results, the model can be further adjusted and optimized. The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the performance indicators of the verification. The formulas are:

[0066]

[0067]

[0068] Where n is the number of samples, y pred is the predicted value of the model, y true is the true value of the sample. The smaller the difference, the better the prediction effect of the model.

[0069] Step S104, inputting the cloud map data and meteorological data corresponding to the photovoltaic power to be predicted into the prediction model after preprocessing, and obtaining the corresponding photovoltaic power generation power prediction result.

[0070] The method provided in an embodiment of the present invention constructs a multi-input deep learning model to input meteorological data and cloud map data into the same model for processing to obtain prediction results. The multi-input deep learning model is a neural network model that accepts multiple inputs. It can input different types of data into the same model for processing and fusion to obtain more comprehensive and accurate prediction results.

[0071] In this embodiment, a photovoltaic power generation prediction device based on a multi-input model is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0072] This embodiment provides a photovoltaic power prediction device based on a multi-input model, such as Figure 4 As shown, including:

[0073] The data preprocessing module is used to preprocess the cloud map data and meteorological data corresponding to photovoltaics respectively;

[0074] A multi-input model building module is used to build a multi-input model, which includes: two input modules, two feature processing branch networks, a feature fusion layer, a shared hidden layer and an output layer. The pre-processed cloud image data and meteorological data are respectively input into two different feature processing branch networks through the two input modules for processing to obtain the features corresponding to the meteorological data and the cloud image data, and the two types of features are fused through the feature fusion layer. The fused features are processed by several layers of neural network through the shared hidden layer to obtain new features, and the new features are input into the output layer for photovoltaic power generation prediction;

[0075] The model training module is used to train the constructed multi-input model and obtain the trained model as a prediction model;

[0076] The power prediction module is used to input the cloud map data and meteorological data corresponding to the photovoltaic power to be predicted, and after preprocessing, input them into the prediction model to obtain the corresponding photovoltaic power prediction result.

[0077] The photovoltaic power prediction device based on the multi-input model in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0078] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0079] The embodiment of the present invention also provides a computer device having the above Figure 5 The photovoltaic power generation prediction device based on the multi-input model is shown.

[0080] See also Figure 5 , Figure 5is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 5 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 5 A processor 10 is taken as an example.

[0081] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0082] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0083] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of a computer device based on the presentation of a small program landing page, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0084] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0085] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0086] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0087] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A photovoltaic power generation prediction method based on a multi-input model, characterized in that: The method comprises: Pre-process the cloud map data and meteorological data corresponding to photovoltaics respectively; The multi-input model constructed includes: two input modules, two feature processing branch networks, a feature fusion layer, a shared hidden layer and an output layer. The pre-processed cloud image data and meteorological data are respectively input into two different feature processing branch networks through the two input modules for processing to obtain the features corresponding to the meteorological data and the cloud image data, and the two types of features are fused through the feature fusion layer. The fused features are processed by several layers of neural network through the shared hidden layer to obtain new features, and the new features are input into the output layer for photovoltaic power generation prediction; wherein the cloud image data is processed by a first CNN network branch, and the first CNN network branch includes: multiple convolution modules, a residual connection structure and a multi-scale enhancement module, and the multi-scale enhancement module is connected to the CNN network. The residual connection structure is parasitic between two independent convolution modules, and the dilated convolution is used internally to enrich the multi-scale representation of features. The multi-scale enhancement module uses a 32-dimensional 3x3 convolution kernel to process the input cloud image data. The back end is connected to multiple dilated convolution kernels to extract multi-dimensional features. The multi-dimensional features extracted by dilated convolution are merged, and the 1x1 convolution kernel is used to transform the dimension and output the enhanced features to obtain a multi-channel two-dimensional feature matrix. The meteorological data is processed in the second CNN network branch. The processing process includes: processing the meteorological data through multiple layers of one-dimensional convolution, then expanding the features into one-dimensional vectors, transforming the number of feature points through a fully connected layer, and finally converting it into a multi-channel two-dimensional matrix with the same scale as the output features of the first CNN network branch. A joint loss function is established to train the multi-input model, and a weighted average loss function is used to train the network. The trained model is used as a prediction model. Different data sets and loss functions are used alternately during the training process. The process includes: The meteorological data and cloud image data are respectively input into the corresponding network branches for separate training to obtain the pre-trained network weights; The two networks are fed with data at the same time, and a joint loss function is added for end-to-end joint training. The loss functions of the two networks are calculated separately during each training, and the parameters of the two networks are updated separately according to their weights. The cloud map data and meteorological data corresponding to the photovoltaic power to be predicted are input for preprocessing and then input into the prediction model to obtain the corresponding photovoltaic power generation power prediction result.

2. The method according to claim 1, characterized in that The parameters involved in the multi-scale enhancement module include: dilation rate factor and number of convolution layers, and different dilation convolutions are obtained by using different dilation rates for the input two-dimensional feature map.

3. A photovoltaic power generation prediction device based on a multi-input model, characterized in that: For executing the method according to claim 1, the device comprises: The data preprocessing module is used to preprocess the cloud map data and meteorological data corresponding to photovoltaics respectively; A multi-input model building module is used to build a multi-input model, which includes: two input modules, two feature processing branch networks, a feature fusion layer, a shared hidden layer and an output layer. The pre-processed cloud image data and meteorological data are respectively input into two different feature processing branch networks through the two input modules for processing to obtain the features corresponding to the meteorological data and the cloud image data, and the two types of features are fused through the feature fusion layer. The fused features are processed by several layers of neural network through the shared hidden layer to obtain new features, and the new features are input into the output layer for photovoltaic power generation prediction; The model training module is used to train the constructed multi-input model and obtain the trained model as a prediction model; The power prediction module is used to input the cloud map data and meteorological data corresponding to the photovoltaic power to be predicted, and after preprocessing, input them into the prediction model to obtain the corresponding photovoltaic power prediction result.

4. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the photovoltaic power generation power prediction method based on a multi-input model as described in any one of claims 1 to 2 by executing the computer instructions.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the photovoltaic power prediction method based on a multi-input model as described in any one of claims 1 to 2.

Images