Photovoltaic power generation prediction method, device and storage medium

Abstract

The application relates to the photovoltaic power generation prediction field, in particular to a photovoltaic power generation prediction method, equipment and a storage medium. The prediction method comprises the following steps: acquiring historical weather features, equipment identification information and historical actual power generation; inputting the equipment identification information into a feature coding model after being converted into a numerical serial number to obtain equipment coding features; inputting the historical weather features and the equipment coding features into a deep learning model after being fused to obtain photovoltaic predicted power generation; based on a preset physical estimation model, extracting part of the weather features and part of the coding features to calculate photovoltaic estimated power generation; finally, calculating fusion loss based on the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical actual power generation to perform model training, and predicting photovoltaic power generation based on the trained feature coding model and the deep learning model. According to the method, the model can effectively predict under the condition of less data or more noise, and the generalization ability and the robustness of the model are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power generation prediction, in particular to a photovoltaic power generation prediction method, device and storage medium. BACKGROUND

[0002] Photovoltaic power generation prediction is the core technical support for users to maximize economic performance, optimize operation efficiency and stabilize grid cooperation. Different photovoltaic devices have different generating capacities under the same weather conditions due to differences in orientation, inclination angle, photovoltaic panel area, shading, dust, etc.

[0003] In the prior art, photovoltaic power generation prediction is a machine learning method driven by historical weather feature data and historical photovoltaic power generation data. This method, on the one hand, relies largely on a large amount of high-quality data, and in the case of less data or more noise, the prediction accuracy is low; on the other hand, it cannot distinguish device personality and is prone to prediction bias, with poor model generalization ability. SUMMARY

[0004] Embodiments of the present application aim to provide a photovoltaic power generation prediction method, device and storage medium to solve the problem that the deep learning model in the prior art can only rely on historical weather feature data and historical photovoltaic power generation data to drive and cannot obtain accurate and reliable prediction results and cannot distinguish device personality, which is prone to prediction bias.

[0005] To solve the above technical problems, the embodiments of the present application provide the following technical solutions:

[0006] According to a first aspect of the present application, a photovoltaic power generation prediction method is provided, the method comprising:

[0007] obtaining historical weather features, device identification information and historical true power generation;

[0008] converting the device identification information into a numerical sequence number and inputting it into a feature encoding model to obtain device encoding features;

[0009] inputting the historical weather features and the device encoding features into a deep learning model after fusion to obtain photovoltaic predicted power generation;

[0010] based on a pre-set physical estimation model, extracting part of the weather features from the historical weather features and part of the encoding features from the device encoding features to calculate photovoltaic estimated power generation;

[0011] fusing loss is calculated based on the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical real power generation, and the deep learning model and the feature encoding model are trained based on the fusing loss to obtain a trained deep learning model and a trained feature encoding model;

[0012] The photovoltaic power generation is predicted based on the trained feature encoding model and the trained deep learning model.

[0013] Optionally, the feature encoding model comprises an embedding layer, a batch normalization layer, a full connection layer and a Sigmoid normalization layer.

[0014] Optionally, the device encoding features comprise conversion efficiency, photovoltaic panel area, temperature coefficient, photovoltaic panel inclination angle, photovoltaic panel pitch angle and a plurality of device characteristic features.

[0015] Optionally, the extracting part of weather features from the historical weather features and the extracting part of encoding features from the device encoding features based on the preset physical estimation model comprises:

[0016] The radiation and the temperature are extracted from the historical weather features and the conversion efficiency, the photovoltaic panel area and the temperature coefficient are extracted from the device encoding features based on the preset physical estimation model.

[0017] Optionally, the calculation formula of the physical estimation model is:

[0018]

[0019] wherein, is the photovoltaic estimated power generation, is the conversion efficiency, is the radiation, is the photovoltaic panel area, is the temperature coefficient, is the temperature, is a preset reference temperature.

[0020] Optionally, the deep learning model comprises a convolutional neural network module, a position encoding module, a long short-term memory network module, a Transformer module and a full connection layer.

[0021] Optionally, the input feature of the long short-term memory network module is a feature fused by adding the output feature of the convolutional neural network module and the output feature of the position encoding module.

[0022] Optionally, the calculating the fusing loss based on the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical real power generation comprises:

[0023] calculate a physical constraint loss based on the photovoltaic predicted power generation and the photovoltaic estimated power generation;

[0024] calculate a model prediction loss based on the photovoltaic predicted power generation and the historical real power generation;

[0025] calculate a fusion loss based on the physical constraint loss and the model prediction loss.

[0026] According to a second aspect of the present application, an electronic device is provided, comprising at least one processor and a memory connected to the at least one processor in communication, the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the photovoltaic power generation prediction method described above.

[0027] According to a third aspect of the present application, a computer storage medium is provided, the computer storage medium storing instructions or programs, when the instructions or programs are executed by at least one processor, the at least one processor performs the photovoltaic power generation prediction method described above.

[0028] The beneficial effects of the embodiments of the present application are: different from the prior art, in the embodiments of the present application, a photovoltaic power generation prediction method is provided, historical weather features, device identification information and historical real power generation are obtained first, and the device identification information is converted into a numerical sequence number and input into a feature coding model to obtain device coding features; then the historical weather features and the device coding features are fused and input into a deep learning model to obtain a photovoltaic predicted power generation, and based on a preset physical estimation model, part of the weather features is extracted from the historical weather features, part of the coding features is extracted from the device coding features, and a photovoltaic estimated power generation is calculated; finally, a fusion loss is calculated based on the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical real power generation, the deep learning model and the feature coding model are trained based on the fusion loss, a trained deep learning model and a trained feature coding model are obtained, and the photovoltaic power generation is predicted based on the trained feature coding model and the trained deep learning model. The method of the present application is different from the traditional data-driven neural network, not only the individualized features of the device are added in the learning process, but also the model is guided by the physical knowledge law in the learning process, so that the model can effectively predict in the case of less data or more noise, and the generalization ability and robustness of the model are improved. BRIEF DESCRIPTION OF DRAWINGS

[0029] One or more embodiments are illustrated by way of example in the drawings, which are for illustrative purposes only and are not intended to limit the embodiments, elements having the same reference numerals in the drawings represent similar elements, unless otherwise specified, the drawings do not constitute a proportional limit.

[0030] Figure 1 is an algorithm framework schematic diagram of a photovoltaic power generation prediction model provided by an embodiment of the present application;

[0031] Figure 2 is a structural diagram of a feature encoding model provided by an embodiment of the present application;

[0032] Figure 3 is a structural diagram of a deep learning model provided by an embodiment of the present application;

[0033] Figure 4 is a flowchart schematic diagram of a photovoltaic power generation prediction method provided by an embodiment of the present application;

[0034] Figure 5 is a structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0035] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in a clear and complete manner with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0036] In addition, the technical features involved in each of the embodiments of the present application described below can be combined with each other as long as there is no conflict.

[0037] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0038] The key terms in the present application are explained as follows:

[0039] 1. Data-driven: It is a method based on data analysis and machine learning, which trains the model through a large amount of data to make predictions or decisions.

[0040] 2. Physics-informed: It means to incorporate physical laws and prior knowledge into the machine learning model to improve the accuracy and interpretability of the model.

[0041] 3. Convolutional Neural Network (CNN): A deep learning model commonly used for image recognition and processing, which extracts local features through convolutional layers.

[0042] 4. Long Short-Term Memory (LSTM): A special type of recurrent neural network that can effectively handle long-term dependencies in time series data.

[0043] 5. Transformer: A deep learning model based on self-attention mechanisms, widely used in natural language processing and time series prediction, capable of capturing global dependencies.

[0044] 6. Positional Encoding: A method used in Transformer models to represent the position information of each element in the input sequence, helping the model understand the order of the sequence.

[0045] 7. Loss Function: A function used to measure the difference between the model's predicted values and the true values, optimizing model parameters by minimizing the loss function.

[0046] 8. Generalization Ability: The ability of a model to perform well on unseen data, meaning the model can adapt to new data and make accurate predictions.

[0047] 9. Embedding Layer: A neural network layer that converts discrete data (such as categorical variables) into continuous vectors, facilitating model processing and learning.

[0048] 10. Conversion Efficiency: The efficiency of a photovoltaic system in converting solar energy into electrical energy, usually expressed as a percentage.

[0049] 11. Photovoltaic Panel Area: The effective area of a photovoltaic module, with a larger area theoretically resulting in higher power generation.

[0050] 12. Temperature Coefficient: The ratio of the change in photovoltaic module output power with respect to temperature, usually negative, indicating that power decreases as temperature increases.

[0051] 13. Photovoltaic Panel Tilt Angle: This refers to the angle between the photovoltaic panel and the horizontal plane. A reasonable tilt angle can improve power generation efficiency.

[0052] 14. Photovoltaic Panel Pitch Angle: This refers to the angle of the photovoltaic panel in the vertical direction, which affects its radiation reception efficiency.

[0053] 15. SN code (Serial Number): A unique serial number for a device or component, used to identify and track product manufacturing, installation, and maintenance information.

[0054] 16. Feature Encoding: This is the process of converting non-numerical features into numerical forms so that they can be processed by machine learning models.

[0055] 17. Dense Vector: A vector consisting of continuous numerical values, typically used to represent features or embeddings.

[0056] 18. Photovoltaic Panel Dust: This refers to the dust accumulated on photovoltaic panels, which can affect their efficiency in receiving radiation.

[0057] 19. Occlusion: This refers to the photovoltaic panel being partially or completely covered by an object, resulting in a decrease in power generation efficiency.

[0058] 20. Generalization Ability: This refers to the model's ability to perform well on unseen data, meaning the model can adapt to new data and make accurate predictions.

[0059] 21. Robustness: refers to the ability of a model to maintain stable performance when faced with noise, outliers, or data changes.

[0060] 22. Adam (Adaptive Moment Estimation): This is a widely used deep learning optimization algorithm that dynamically adjusts the learning rate of each parameter by calculating the first moment (mean) and second moment (variance) of the gradient. AdamW is an improved version of Adam, which corrects the implementation of weight decay in the traditional Adam optimizer, providing better generalization performance in deep learning training.

[0061] The embodiment of the present application provides a photovoltaic power generation prediction method fusing physical information. The method uses physical laws to guide the training of a deep learning model, adds the individual characteristics of equipment in the training process, introduces physical knowledge into a loss function, so that the model can not only learn from data, but also comply with the physical constraints in the photovoltaic field, can effectively train in the case of data shortage and noise, and generate a prediction with high accuracy, thereby enhancing the robustness and generalization ability of the model.

[0062] Please refer to Figure 1 , Figure 1 is an algorithm framework schematic diagram of a photovoltaic power generation prediction model provided by the embodiment of the present application. As shown in Figure 1 , the photovoltaic power generation prediction model 100 comprises a first input module 10, a second input module 20, an identification value conversion module 30, a feature coding model 40, a deep learning model 50, a physical estimation module 60 and a loss calculation module 70.

[0063] The first input module 10 is used to acquire historical weather features, including radiation, temperature, air pressure, wind speed, cloud cover, solar angle and the like.

[0064] The second input module 20 is used to acquire equipment identification information. In an embodiment, the equipment identification information is an equipment SN code, which is usually a string containing letters and numbers.

[0065] The identification value conversion module 30 is used to convert the equipment identification information into a numerical sequence. Since the equipment identification information contains letters and numbers, it cannot be directly input into a neural network, so it needs to be converted into a numerical sequence such as 0, 1, 2,..., 9999.

[0066] The feature coding model 40 is used to map the discrete numerical sequence into a continuous vector representation, convert the high-dimensional vector representation into a low-dimensional dense vector, and obtain equipment coding features through normalization processing. Mapping the discrete numerical sequence into a continuous vector representation not only facilitates the processing of the deep learning model, but also captures the semantic information and similarity between categories. Converting the high-dimensional coefficient data into a low-dimensional dense vector effectively reduces the complexity and computational amount of the model, and at the same time, as the input layer of the model, provides a suitable data form for the subsequent deep learning model, so that the model can learn more rich feature representations, thereby improving the overall performance. In an embodiment, the equipment coding features include conversion efficiency, photovoltaic panel area, temperature coefficient, photovoltaic panel inclination angle, photovoltaic panel pitch angle and a plurality of equipment characteristic features; the equipment characteristic features include electronic characteristics, photovoltaic panel dust, shielding and the like.

[0067] The deep learning model 50 takes the historical weather features obtained by the first input module 10 and the features obtained by fusing the device encoding features output by the feature encoding model 40 as input features, and outputs the photovoltaic predicted power generation. The specific network structure of the deep learning model 50 is described below in relation to Figure 3

[0068] The physical estimation module 60 is configured to calculate the photovoltaic estimated power generation based on a preset physical estimation model, from the historical weather features and from the device encoding features. In an embodiment, the calculation formula of the physical estimation model is as follows:

[0069]

[0070] wherein, is the photovoltaic estimated power generation, is the conversion efficiency, is the radiation, is the photovoltaic panel area, is the temperature coefficient, is the temperature, is the preset reference temperature.

[0071] According to the parameter information required by the physical estimation model, the part of weather features extracted from the historical weather features and the part of encoding features extracted from the device encoding features include: the radiation and the temperature extracted from the historical weather features, and the conversion efficiency, the photovoltaic panel area and the temperature coefficient extracted from the device encoding features. It can be understood that those skilled in the art can also construct other physical estimation models and extract the required features from the historical weather features and the device encoding features for calculation.

[0072] The loss calculation module 70 is configured to calculate the fusion loss based on the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical real power generation. Specifically, the physical constraint loss is calculated based on the photovoltaic predicted power generation and the photovoltaic estimated power generation, the model prediction loss is calculated based on the photovoltaic predicted power generation and the historical real power generation, and the fusion loss is calculated based on the physical constraint loss and the model prediction loss. This module is mainly applied in the training process of the photovoltaic power generation prediction model, and is aimed at adjusting the related parameters of the deep learning model and the feature encoding model according to the fusion loss, to provide guidance for the learning direction of the model.

[0073] In an embodiment, the calculation formula of the fusion loss is as follows:

[0074]

[0075] wherein, is the fusion loss, is the physical constraint loss, is the model prediction loss,​ , is a weight coefficient, is a sample quantity, is a true value of the i-th sample, is an estimated value of the i-th sample by a physical estimation model, is a predicted value of the i-th sample by a deep learning model.

[0076] Please refer to Figure 2 , Figure 2 is a structural diagram of the feature encoding model 40 provided by the embodiment of the present application. As shown in Figure 2 , the feature encoding model 40 includes an embedding layer, a batch normalization layer, a full connection layer and a sigmoid normalization layer. The input of the feature encoding model 40 is a numerical serial number converted from a device SN code, the embedding layer maps the discrete numerical serial number to a continuous vector representation through semantic vectorization. The batch normalization layer performs batch normalization on the continuous vector representation, the full connection layer is used to extract core features, perform feature compression, and convert the high-dimensional vector representation into a low-dimensional dense vector. The sigmoid normalization normalizes the low-dimensional dense vector to 0-1, which corresponds to N device encoding features. The embodiment of the present application introduces an embedded feature encoding model to perform interpretable modeling on unknown device features, thereby enhancing the interpretability and robustness of the model.

[0077] Please refer to Figure 3 , Figure 3 is a structural diagram of the deep learning model 50 provided by the embodiment of the present application. As shown in Figure 3 , the deep learning model 50 includes a CNN module, a position encoding module, an LSTM module, a Transformer module and a full connection layer.

[0078] ​​​The CNN module is used for local feature extraction, and the input feature is the feature after fusion of historical weather features and device encoding features. In an embodiment, the structure of the historical weather features is (B, 60, 50), where B is the sample size, in units of days; 60 is the number of time intervals at intervals of 15 minutes, from 5 am to 9 pm; and 50 is 50 weather features. Assuming that the number of device encoding features is 8, the feature structure after fusion of the historical weather features and the device encoding features is (B, 60, 58). The processing process of the CNN module is as follows: 1) adjusting the dimensions of the input data, and exchanging the data contents of the second dimension and the third dimension. For example, the input feature (B, 60, 58) is adjusted to (B, 58, 60). 2) Convolution processing. For example, after convolution processing of the input feature (B, 58, 60) using a convolution kernel of 3 and an output channel of 128, the output feature structure is (B, 128, 60). 3) Batch normalization processing. 4) Restoring the data dimensions, that is, exchanging the data contents of the second dimension and the third dimension again. For example, the output feature structure of the input feature (B, 128, 60) after restoration is (B, 60, 128).

[0079] The position encoding module is used to provide time sequence information for the model, which is crucial for capturing the periodicity and trend of photovoltaic power generation. Photovoltaic power generation follows the law of the sun's movement and has obvious daily periodicity. The position encoding module can provide absolute position information for each time point for the model, helping the model understand the sequence and periodicity in the time series, and the model can more accurately learn the relationship between photovoltaic power generation and time position, thereby improving the accuracy of prediction. Specifically, the formula of the position encoding module is as follows:

[0080]

[0081] wherein, is the time step position index, is the index, 10000 is the frequency base, used to control the frequency of the sine function and the cosine function, dim is the total dimension of the feature. For example, when the input feature structure of the position encoding module is (B, 60, 128), is 0, 1, …, 59; is 0, 1, …, 63; dim is 128.

[0082] The LSTM module is used to capture long-term dependencies in time series data, prevent gradient vanishing and gradient explosion, and the input features thereof are the features fused from the output features of the CNN module and the output features of the position encoding module. In an embodiment, the structure of the features extracted by the CNN module is (B, 60, 128), the structure of the features extracted by the position encoding module is (B, 60, 128), and the two are fused by addition, and the output feature structure is (B, 60, 128). The output feature structure of the LSTM module is the same as the input feature structure, which is also (B, 60, 128).

[0083] The Transformer module is used to capture global dependencies in time series data based on an attention mechanism. In an embodiment, the output feature structure of the Transformer module and the input feature structure are both (B, 60, 128).

[0084] The fully connected layer is used to linearly transform and reduce the dimension of the high-dimensional output features, and convert them into the final prediction results. In an embodiment, the input feature structure of the fully connected layer is (B, 60, 128), and the output feature structure is (B, 60, 1). Further, the last dimension of the output data is removed to obtain the photovoltaic power generation prediction result (B, 60), wherein 60 represents the photovoltaic power generation prediction result of every 15-minute interval.

[0085] Please refer to Figure 4 , Figure 4 is a flowchart of a photovoltaic power generation prediction method provided by an embodiment of the present application, and the method comprises:

[0086] In step S401, historical weather features, device identification information, and historical real power generation are obtained.

[0087] Specifically, historical prediction meteorological data, historical power generation data, and related time data of a photovoltaic power station are collected to obtain historical weather features with a dimension of A B C and historical real power generation with a dimension of A B. Wherein, A is the number of days, B is the number of sunshine time intervals per day, and C is the number of weather features. The historical weather features can include rainfall, temperature, humidity, radiation, air pressure, and wind speed. The historical real power generation is the real power generation of the corresponding date and time interval of the historical prediction weather features.

[0088] In an embodiment, the device identification information is a device SN code, which is usually a string containing letters and numbers.

[0089] In step S402, the device identification information is converted into a numerical serial number and input into a feature encoding model to obtain device encoding features.

[0090] Since the device identification information contains letters and numbers, it cannot be directly input into the neural network, and therefore needs to be converted into numerical sequence numbers such as 0, 1, 2, …, 9999 first. The feature encoding model maps the discrete numerical sequence numbers into continuous vector representations, then converts the high-dimensional vector representations into low-dimensional dense vectors, and obtains the device encoding features through normalization processing. In an embodiment, the device encoding features include conversion efficiency, photovoltaic panel area, temperature coefficient, photovoltaic panel inclination angle, photovoltaic panel pitch angle, and several device characteristic features; the device characteristic features include electronic characteristics, photovoltaic panel dust, and shading.

[0091] In step S403, the historical weather features and the device encoding features are fused and input into the deep learning model to obtain the photovoltaic predicted power generation.

[0092] Specifically, the historical weather features and the device encoding features are spliced and fused. For example, if the number of historical weather features is 50 and the number of device encoding features is 8, the number of features after the fusion of the historical weather features and the device encoding features is 58.

[0093] In step S404, based on a preset physical estimation model, part of the weather features is extracted from the historical weather features and part of the encoding features is extracted from the device encoding features to calculate the photovoltaic estimated power generation.

[0094] In an embodiment, the calculation formula of the preset physical estimation model is:

[0095]

[0096] wherein, is the photovoltaic estimated power generation, is the conversion efficiency, is the radiation, is the photovoltaic panel area, is the temperature coefficient, is the temperature, is a preset reference temperature.

[0097] According to the parameter information required by the physical estimation model, the part of the weather features extracted from the historical weather features and the part of the encoding features extracted from the device encoding features include: the radiation and the temperature extracted from the historical weather features, and the conversion efficiency, the photovoltaic panel area, and the temperature coefficient extracted from the device encoding features.

[0098] In step S405, the fusion loss is calculated based on the photovoltaic predicted power generation, the photovoltaic estimated power generation, and the historical real power generation, and the deep learning model and the feature encoding model are trained based on the fusion loss to obtain the trained deep learning model and the trained feature encoding model.

[0099] Specifically, the physical constraint loss is calculated based on the photovoltaic predicted power generation and the photovoltaic estimated power generation, and the model prediction loss is calculated based on the photovoltaic predicted power generation and the historical real power generation, and then the fusion loss is calculated based on the physical constraint loss and the model prediction loss. The module is mainly applied in the training process of the photovoltaic power generation prediction model, and is used to adjust the related parameters of the deep learning model and the feature encoding model according to the fusion loss, so as to provide guidance for the learning direction of the model.

[0100] In an embodiment, the calculation formula of the fusion loss is:

[0101]

[0102] wherein, is the fusion loss, is the physical constraint loss, is the model prediction loss, , is the weight coefficient, is the sample number, is the true value of the i th sample, is the estimated value of the i th sample by the physical estimation model, is the predicted value of the i th sample by the deep learning model. In the model training, based on the fusion loss, the model parameters of the deep learning model and the feature encoding model are adjusted by using the AdamW optimization algorithm, and and are optimized so that

[0103] is minimum.

[0104] In step S406, the photovoltaic power generation is predicted based on the trained feature encoding model and the trained deep learning model.

[0105] When predicting the power generation of the photovoltaic energy storage device, the prediction weather features of the to-be-predicted date and the device identification information of the photovoltaic energy storage device are obtained, the device identification information is converted into a numerical sequence number and then encoded by the trained feature encoding model to obtain device encoding features; the prediction weather features and the device encoding features are fused and input into the trained deep learning model to obtain the photovoltaic predicted power generation. Specifically, assuming that the prediction weather feature structure of a day is (1, 60, 50) and the device encoding feature structure is (1, 8), the fused features (1, 60, 58) are input into the trained deep learning model, and the prediction result (1, 60) is output, wherein 60 represents the photovoltaic predicted power generation of each 15-minute interval.

[0106] ​​​The photovoltaic power generation prediction method provided by the embodiments of the present application comprises the following steps: obtaining historical weather features, equipment identification information and historical actual power generation; inputting the equipment identification information into a feature coding model after being converted into a numerical sequence to obtain equipment coding features; inputting the historical weather features and the equipment coding features into a deep learning model to obtain photovoltaic predicted power generation; extracting part of the weather features from the historical weather features and part of the coding features from the equipment coding features, and calculating the photovoltaic estimated power generation based on the part of the weather features and the part of the coding features by using a physical equation; and finally, training the deep learning model and the feature coding model based on the fusion loss of the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical actual power generation, obtaining the trained deep learning model and the trained feature coding model, and predicting the photovoltaic power generation based on the trained feature coding model and the trained deep learning model. The method of the present application is different from the traditional data-driven neural network. In the learning process, not only the individualized features of the equipment are added, but also the model is guided by using the physical knowledge law in the learning process, thereby improving the generalization ability of the model, improving the prediction accuracy in the case of less data or more noise, and enhancing the robustness of the model.

[0107] According to the embodiments of the present application, an electronic device is provided, such as Figure 5 The electronic device 100 provided by the embodiments of the present application can comprise a processor 10, a communication interface 30, a memory 20 and a communication bus, wherein the processor 10, the communication interface 30 and the memory 20 complete mutual communication through the communication bus. The processor 10 can call the logical instructions in the memory 20 to execute the photovoltaic power generation prediction method described above.

[0108] In addition, the logical instructions in the memory 20 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in several computer readable storage media. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the photovoltaic power generation prediction method described above. The storage medium described above includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0109] According to the embodiments of the present application, a computer readable storage medium is provided, which is of the type described above, and stores a computer program, when the computer program is executed by a processor, the processor executes the steps of the photovoltaic power generation prediction method described above.

[0110] Through the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus a general hardware platform, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part that contributes to the related art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0111] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application. Therefore, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A photovoltaic power generation prediction method characterized by, The method comprises: acquiring historical weather characteristics, device identification information and historical actual power generation; inputting the device identification information into a feature coding model after converting the device identification information into a numerical sequence to obtain device coding characteristics; inputting the historical weather characteristics and the device coding characteristics into a deep learning model after fusing the historical weather characteristics and the device coding characteristics to obtain photovoltaic predicted power generation; extracting part of the weather characteristics from the historical weather characteristics and part of the coding characteristics from the device coding characteristics based on a preset physical estimation model to calculate photovoltaic estimated power generation; calculating fusion loss based on the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical actual power generation, training the deep learning model and the feature coding model based on the fusion loss to obtain a trained deep learning model and a trained feature coding model; predicting photovoltaic power generation based on the trained feature coding model and the trained deep learning model; the calculation of the fusion loss based on the photovoltaic predicted power generation, the photovoltaic estimated power generation and the historical actual power generation comprises: calculating physical constraint loss based on the photovoltaic predicted power generation and the photovoltaic estimated power generation, the calculation formula of the physical constraint loss being: ； calculating model prediction loss based on the photovoltaic predicted power generation and the historical actual power generation, the calculation formula of the model prediction loss being: ； calculating the fusion loss based on the physical constraint loss and the model prediction loss, the calculation formula of the fusion loss being: ； in, For the loss of fusion, For physical constraint loss, To predict the loss for the model, , These are the weighting coefficients. For the sample size, It is the first The true value of each sample For the physical estimation model of the first The estimated value of a sample, For deep learning models to the first The predicted value for each sample.

2. The method of claim 1, wherein, The feature coding model comprises an embedding layer, a batch normalization layer, a fully connected layer and a Sigmoid normalization layer.

3. The method of claim 1, wherein, The device coding characteristics comprise conversion efficiency, photovoltaic panel area, temperature coefficient, photovoltaic panel inclination angle, photovoltaic panel pitch angle and a plurality of device characteristic features.

4. The method of claim 1, wherein, The extraction of part of the weather characteristics from the historical weather characteristics and part of the coding characteristics from the device coding characteristics based on the preset physical estimation model comprises: The extraction of radiation and temperature from the historical weather characteristics and the extraction of conversion efficiency, photovoltaic panel area and temperature coefficient from the device coding characteristics based on the preset physical estimation model.

5. The method of claim 4, wherein, The calculation formula of the physical estimation model is: wherein, is the power production for a photovoltaic, is the conversion efficiency, is the radiation, is the photovoltaic panel area, is the temperature coefficient, is the temperature, is a preset reference temperature.

6. The method of claim 1, wherein, The deep learning model comprises a convolutional neural network module, a position coding module, a long short-term memory network module, a Transformer module and a fully connected layer.

7. The method of claim 6, wherein, The input feature of the long short-term memory network module is the feature obtained by adding and fusing the output feature of the convolutional neural network module and the output feature of the position coding module.

8. An electronic device, comprising: The device comprises at least one processor and a memory connected to the at least one processor in communication, the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the photovoltaic power generation prediction method according to any one of claims 1 to 7.

9. A computer storage medium, characterized in that The computer storage medium stores instructions or programs, when the instructions or programs are executed by at least one processor, the at least one processor executes the photovoltaic power generation prediction method according to any one of claims 1 to 7.

Images