Solar power generation power prediction method and device, electronic device, and storage medium

Abstract

The present disclosure provides a solar power generation power prediction method and device, electronic device, and storage medium, belonging to the field of power prediction technology. The method includes: predicting a first power generation power of a solar power generation device using a physical model based on physical parameters of a target area; the physical model is a mapping model between physical parameters and power generation constructed based on the laws of solar celestial motion and photovoltaic power generation; predicting a second power generation power of the solar power generation device using a deep learning model based on meteorological data of the target area and historical power generation data of the solar power generation device; and fusing the first power generation power and the second power generation power based on the weather type of the target area to obtain a third power generation power of the solar power generation device. The solar power generation power prediction method and device, electronic device, and storage medium provided by the present disclosure can improve the reliability of solar power generation power prediction and facilitate accurate grid scheduling and energy management.

Description

Technical Field

[0001] The present disclosure belongs to the technical field of power prediction, and more specifically, relates to a method and device for predicting solar power generation power, an electronic device, and a storage medium. Background Art

[0002] As a clean, renewable energy source, solar power can reduce the consumption of fossil fuels like coal and oil, thereby reducing environmental pollution. However, its intermittent and volatile nature poses challenges to grid dispatch and energy management. Accurate solar power forecasting is crucial for stable grid operation, economic dispatch, ancillary service provisioning, and participation in electricity markets.

[0003] In recent years, artificial intelligence (AI) prediction methods based on machine learning and deep learning have been widely used in the field of solar power forecasting. These methods can handle high-dimensional nonlinear relationships, have strong adaptive learning capabilities, can integrate heterogeneous data from multiple sources, and achieve high prediction accuracy under big data conditions. However, AI methods suffer from "black box" characteristics, resulting in poor interpretability and requiring large amounts of high-quality data for training. Furthermore, AI models lack adaptability when faced with new scenarios, unusual weather conditions, or changes in equipment status, resulting in low reliability of prediction results. Summary of the Invention

[0004] The present invention aims to provide a method and device for predicting solar power generation, an electronic device, and a storage medium to improve the reliability of solar power generation prediction and facilitate accurate grid scheduling and energy management.

[0005] According to a first aspect of the embodiments of the present disclosure, a method for predicting solar power generation is provided, comprising:

[0006] Based on the physical parameters of the target area, a physical model is used to predict the first power generation power of the solar power generation device; the physical model is a mapping model between physical parameters and power generation power constructed based on the movement law of the solar celestial body and the mechanics of photovoltaic power generation, and the target area is the area where the solar power generation device is located;

[0007] Predicting a second power generation capacity of the solar power generation equipment using a deep learning model based on meteorological data of a target area and historical power generation data of the solar power generation equipment;

[0008] The first power generation power and the second power generation power are merged based on the weather type of the target area to obtain a third power generation power of the solar power generation device.

[0009] According to a second aspect of the embodiments of the present disclosure, a solar power generation power prediction device is provided, comprising:

[0010] a first prediction module for predicting a first power generation capacity of the solar power generation device using a physical model based on physical parameters of a target area; the physical model is a mapping model of physical parameters and power generation capacity constructed based on the movement laws of the solar celestial body and the mechanics of photovoltaic power generation; and the target area is the area where the solar power generation device is located;

[0011] a second prediction module, configured to predict a second generated power of the solar power generation device using a deep learning model based on meteorological data of a target area and historical power generation data of the solar power generation device;

[0012] The data fusion module is used to fuse the first power generation power and the second power generation power based on the weather type of the target area to obtain the third power generation power of the solar power generation equipment.

[0013] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned solar power generation power prediction method when executing the computer program.

[0014] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned solar power generation power prediction method are implemented.

[0015] The solar power generation power prediction method and device, electronic device, and storage medium provided by the embodiments of the present disclosure have the following beneficial effects:

[0016] First, the disclosed embodiment introduces a physical model to predict power generation, which provides a physical basis for power generation prediction of solar power generation equipment and makes it more interpretable.

[0017] Secondly, the disclosed embodiment performs two predictions: a first power generation prediction based on a physical model and a second power generation prediction based on a deep learning model, and then combines the prediction results of the two predictions. In this way, the prediction process based on the physical model can effectively compensate for the shortcomings of "deep learning models' lack of adaptability when facing new scenarios, abnormal weather conditions, or changes in equipment status" and "deep learning models' low prediction accuracy when lacking high-quality data," thereby effectively solving the problems of the existing technology.

[0018] In addition, it should be pointed out that the embodiment of the present disclosure does not simply merge the results of the two predictions, but dynamically merges them based on the weather type, thereby further improving the applicability of power generation prediction to actual application scenarios, effectively ensuring the accuracy and reliability of solar power generation prediction, and facilitating accurate grid scheduling and energy management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 A schematic diagram of a process for predicting solar power generation provided by an embodiment of the present disclosure;

[0021] Figure 2 A schematic diagram of a physical model structure provided in one embodiment of the present disclosure;

[0022] Figure 3 A schematic diagram of a linear fusion layer structure provided in one embodiment of the present disclosure;

[0023] Figure 4 A schematic diagram of the deep learning model structure provided in one embodiment of the present disclosure;

[0024] Figure 5 A structural block diagram of a solar power generation power prediction device provided in one embodiment of the present disclosure;

[0025] Figure 6 A schematic block diagram of an electronic device provided in one embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present disclosure. However, it will be apparent to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present disclosure with unnecessary detail.

[0027] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, specific embodiments will be described below with reference to the accompanying drawings.

[0028] Please refer to Figure 1 , Figure 1This is a flow chart of a solar power generation power prediction method provided by an embodiment of the present disclosure, the method comprising:

[0029] S101: Based on the physical parameters of the target area, a physical model is used to predict the first power generation power of the solar power generation equipment; the physical model is a mapping model of physical parameters and power generation power constructed based on the movement laws of the solar celestial body and the mechanics of photovoltaic power generation, and the target area is the area where the solar power generation equipment is located.

[0030] In this embodiment, the physical parameters of the target area may include the target area's geographic location (latitude and longitude), altitude, atmospheric conditions (such as atmospheric transparency and cloud cover), and characteristic parameters of the solar power generation equipment itself (such as photovoltaic cell type, area, and conversion efficiency). These physical parameters directly or indirectly affect the amount of solar radiation received by the solar power generation equipment and its ability to convert solar radiation into electrical energy, thereby affecting the power generated by the solar power generation equipment.

[0031] In this embodiment, the solar motion law refers to the astronomical laws governing the Earth's revolution and rotation around the sun. Based on this law, the changes in the sun's position in the sky over time and geographic location can be calculated. In this embodiment, the solar motion law can be modeled to produce a solar geometry model. Based on this solar geometry model, parameters such as the solar altitude and azimuth at a specific time and target region can be accurately calculated. This allows the angles between the sun's rays and the ground and the surface of the solar power generation device to be determined, and the amount of solar radiation incident on the solar power generation device to be calculated.

[0032] In this example, the photovoltaic mechanism (also known as the photovoltaic effect) is the fundamental principle of solar power generation. When sunlight strikes a photovoltaic cell made of semiconductor materials (such as silicon), the energy from the photons is transferred to electrons in the semiconductor, causing them to transition into free electrons, thereby generating an electric current. Based on this mechanism, a physical model can be established to describe the relationship between solar radiation energy and the current and voltage generated by the photovoltaic cell, thereby deriving the generated power.

[0033] In this embodiment, the power generation power prediction of the solar power generation equipment is performed based on the physical model. It can reflect the theoretical value of the solar power generation power more scientifically based on the physical essence, and the prediction result is more interpretable.

[0034] S102: Predicting a second power generation power of the solar power generation equipment using a deep learning model based on meteorological data of the target area and historical power generation data of the solar power generation equipment.

[0035] In this embodiment, based on the deep learning model, complex power generation patterns can be learned directly from historical power generation data and meteorological data, thereby predicting the power generation power of solar power generation equipment.

[0036] Specifically, the deep learning model may include a feature extraction network, a time series modeling network, and a fully connected layer. On this basis, predicting the second power generation of the target area based on the deep learning model may include:

[0037] Extracting features from historical power generation data and meteorological data of the target area based on a feature extraction network to obtain first feature data;

[0038] Inputting the first feature data into the time series modeling network to obtain second feature data;

[0039] The second feature data is input into the fully connected layer to obtain the second generated power of the target area.

[0040] In this embodiment, the potential patterns and regularities in meteorological data and historical power generation data are mined based on a deep learning model, which can capture nonlinear relationships and random factors that are difficult to accurately describe with physical models, so that the second power generation power of the solar power generation equipment can be predicted with higher accuracy.

[0041] S103: The first power generation power and the second power generation power are integrated based on the weather type of the target area to obtain a third power generation power of the solar power generation equipment.

[0042] In this embodiment, the third power generation power is the predicted result of the solar power generation power in the target area. By fusing the first power generation power and the second power generation power to obtain the third power generation power, the respective advantages of the physical model and the deep learning model can be utilized to improve the accuracy of the solar power generation power prediction.

[0043] Specifically, the first power generation and the second power generation can be fused by weighted summation. At the same time, considering that the prediction effects of the physical model and the deep learning model are different under different weather types, different weights can be selected according to the weather type when performing weighted summation on the physical model and the deep learning model. For example, in clear weather, the prediction results of the physical model are closer to the actual value, and the corresponding weight of the physical model can be increased; while in cloudy, overcast weather or when there are special meteorological phenomena (such as sandstorms, heavy rain), the deep learning model can better handle complex situations, and the weight of its prediction results can be increased. This embodiment adaptively fuses the first power generation and the second power generation based on weather type, which can give full play to the advantages of the two methods, comprehensively consider physical principles and actual operating conditions, and obtain more accurate and reliable power generation prediction values.

[0044] From the above, it can be concluded that this embodiment first introduces a physical model to predict the power generation, so that the power generation prediction of the solar power generation equipment has a physical basis and is more explainable;

[0045] Secondly, the disclosed embodiment performs two predictions: a first power generation prediction based on a physical model and a second power generation prediction based on a deep learning model, and then combines the prediction results of the two predictions. In this way, the prediction process based on the physical model can effectively compensate for the shortcomings of "deep learning models' lack of adaptability when facing new scenarios, abnormal weather conditions, or changes in equipment status" and "deep learning models' low prediction accuracy when lacking high-quality data," thereby effectively solving the problems of the existing technology.

[0046] In addition, it should be pointed out that the embodiment of the present disclosure does not simply merge the results of the two predictions, but dynamically merges them based on the weather type, thereby further improving the applicability of power generation prediction to actual application scenarios, effectively ensuring the accuracy and reliability of solar power generation prediction, and facilitating accurate grid scheduling and energy management.

[0047] Please refer to Figure 2 In one embodiment of the present disclosure, the physical model is used to determine the irradiance of the target area based on the spatiotemporal parameters and the atmospheric optical characteristic parameters, and to determine the first power generation power based on the irradiance and the photoelectric conversion parameters;

[0048] Before predicting the first power generation capacity of the solar power generation equipment using the physical model based on the physical parameters of the target area, the method further includes:

[0049] Based on the error between the first generated power predicted by the physical model on the first historical date and the corresponding actual generated power, the atmospheric optical characteristic parameters and the photoelectric conversion parameters in the physical model are optimized.

[0050] In this embodiment, the spatiotemporal parameters may include timestamp, longitude information and latitude information, and the atmospheric optical characteristic parameters may be atmospheric transmittance. The position of the sun at a specific moment may be accurately calculated based on the timestamp information, the longitude information may reflect the time of solar radiation, the latitude information may reflect the radiation characteristics of the sun, and the atmospheric transmittance may reflect the proportion or degree of solar radiation passing through the atmosphere and reaching the ground. Therefore, this embodiment may obtain the irradiance of the target area based on the timestamp, longitude information, latitude information and atmospheric optical characteristic parameters and the movement law of the solar celestial body.

[0051] In this embodiment, based on the irradiance of the target area, the photoelectric conversion parameter can characterize the conversion relationship between the irradiance and the generated power. Therefore, the corresponding first generated power can be obtained based on the irradiance and the photoelectric conversion parameter.

[0052] This embodiment first calculates the irradiance of the target area based on the timestamp information, longitude information, latitude information and atmospheric transmittance of the target area, and then obtains the first power generation power based on the irradiance and the photoelectric conversion parameters. The above calculation process does not require the input of a large amount of meteorological data (such as cloud cover, temperature, humidity, etc.), the calculation amount is small, and the timestamp information, longitude information and latitude information are relatively fixed, and the calculation process is not easily affected by external factors.

[0053] At the same time, considering the influence of factors such as seasonal changes and aging of power generation equipment (such as aging of photovoltaic modules and dust accumulation), the prediction results of the first power generation power may be inaccurate.

[0054] To solve the above problem, this embodiment can regularly obtain historical data of the physical model, such as the first generated power predicted by the physical model on the first historical date and the corresponding actual generated power. With the goal of minimizing the error between the first generated power and the corresponding actual generated power, the atmospheric optical characteristic parameters and the photoelectric conversion parameters are adjusted to achieve the optimal value. The first generated power is predicted based on the optimized atmospheric optical characteristic parameters and photoelectric conversion parameters, which can improve the accuracy of the first generated power prediction.

[0055] Specifically, the calculation process of the irradiance and the first power generation is shown in the following first formula:

[0056]

[0057]

[0058] in, represents the solar constant (1361 W / m²), represents the zenith angle, Indicates atmospheric transmittance (value between 0-1), represents the photoelectric conversion parameter, represents irradiance, Indicates the predicted value of the first generated power, ” indicates scalar multiplication.

[0059] According to the first formula, the predicted value of the first power generation can be obtained by using the physical parameters such as the zenith angle, atmospheric transmittance and irradiance of the first historical date. On this basis, an error function can be constructed based on the predicted value of the first power generation and the corresponding actual value. With the goal of minimizing the error function, the error function is adjusted. and , finally confirmed and The optimal value of , the error function can adopt the root mean square error function, as shown in the second formula below:

[0060] .

[0061] in, represents the actual value of the first generated power on the i-th date, It represents the predicted value of the first generated power on the i-th day.

[0062] From the above, it can be concluded that this embodiment constructs an error function based on the predicted value of the first power generation power and the corresponding actual value, and optimizes the atmospheric optical characteristic parameters and photoelectric conversion parameters with the goal of minimizing the error function, so that the physical model can more accurately reflect the mapping relationship between the physical parameters of the target area and the solar power generation power, thereby improving the accuracy and reliability of the prediction.

[0063] In one embodiment of the present disclosure, the solar power generation power prediction method further includes:

[0064] The multiple historical dates are screened based on the correlation between the irradiance and the actual power generation values corresponding to the multiple historical dates to obtain a first historical date; the irradiance corresponding to the multiple historical dates is the irradiance of the target area on the multiple historical dates obtained based on the physical model, and the actual power generation value is the actual power generation value of the solar power generation equipment in the target area on the multiple historical dates;

[0065] Multiple historical dates are the N dates before the forecast date.

[0066] In this embodiment, a rolling training strategy can be adopted to optimize the atmospheric optical characteristic parameters and the photoelectric conversion parameters based on the historical data of the N dates before the prediction date (that is, the most recent historical data). Since the most recent historical data can more accurately reflect the current atmospheric environment characteristics and the latest performance status of the photoelectric equipment, the optimization of the atmospheric optical characteristic parameters and the photoelectric conversion parameters based on the historical data of the N dates before the prediction date can improve the accuracy of the first power generation prediction.

[0067] Furthermore, this embodiment selects a date with a higher correlation from the first N dates of the predicted date based on the correlation between the irradiance and the actual power generation value corresponding to multiple historical dates, that is, the first historical date. The correlation corresponding to the first historical date is higher, indicating that the irradiance obtained based on the physical model on these dates can be better mapped to the actual power generation output. Therefore, the first historical date can be called a high-quality day.

[0068] The correlation between the irradiance and the actual power generation values corresponding to multiple historical dates can be obtained by calculating the correlation coefficient between the two. Taking the Pearson correlation coefficient as an example, the specific calculation formula for the correlation is:

[0069]

[0070] in, represents the Pearson correlation coefficient, n represents the n time points of any historical date, represents the irradiance of the target area at the i-th time point, represents the average irradiance of the target area on any historical date, represents the actual value of the power generated by the solar power generation equipment at the i-th time point, It represents the average power generated by the solar power generation equipment on any historical date.

[0071] In this embodiment, a first threshold may be pre-set, and a date with a correlation greater than the first threshold is determined as the first historical date. For example, the first threshold may be set to 0.95, 0.98, 0.99, etc.

[0072] On the basis of obtaining the first historical date, the physical model is retrained based on the physical parameters of the first historical date, historical predicted values and corresponding actual values, and the atmospheric optical characteristic parameters and photoelectric conversion parameters of the physical model can be dynamically optimized.

[0073] As can be seen from the above, this embodiment employs a rolling training strategy and dynamically optimizes the atmospheric optical property parameters and photoelectric conversion parameters of the physical model, enhancing the model's adaptability to seasonal changes and equipment aging, and improving the stability and reliability of predictions. Furthermore, by selecting the first historical date with high relevance, the model can effectively learn from a small amount of high-quality data, further reducing the computational effort during the dynamic optimization process and improving computational accuracy.

[0074] In one embodiment of the present disclosure, determining the irradiance of a target area based on timestamp information, longitude information, latitude information, and atmospheric transmittance includes:

[0075] Determine the solar declination angle and solar hour angle of the target area based on the timestamp information, longitude information and latitude information of the target area;

[0076] Determine the zenith angle of the target area based on the latitude information, solar declination angle and solar hour angle of the target area;

[0077] The irradiance of the target area is determined based on the zenith angle and atmospheric transmittance of the target area.

[0078] In this embodiment, the calculation process of the target area irradiance is:

[0079] (1) Solar time calculation: Convert standard time to solar time, taking into account the longitude correction and the time difference equation:

[0080]

[0081] The standard time is the time obtained from the timestamp information, specifically to the hour, such as 12 o'clock, 1 o'clock, etc. The standard longitude is a preset constant. is the time difference equation, calculated by the following formula:

[0082]

[0083]

[0084] The day number refers to the number of days in a year starting from January 1. For example, the day number of January 1 is 1, and the day number of January 2 is 2.

[0085] Based on the solar time, the solar hour angle can be obtained based on the solar time:

[0086]

[0087] in, is the solar hour angle, Indicates solar time in hours.

[0088] (2) Zenith angle calculation: The zenith angle represents the angle between the sun's rays and the ground normal, and is a key parameter for irradiance calculation:

[0089]

[0090] in, is the latitude information, is the solar declination angle, is the solar hour angle. The solar declination angle is calculated using the following formula:

[0091]

[0092] (3) Irradiance calculation: Calculate the theoretical irradiance based on the zenith angle and atmospheric transmittance:

[0093]

[0094] in, Indicates irradiance, specifically theoretical irradiance, is the solar constant (1361 W / m²), is the atmospheric transmittance (a value between 0-1).

[0095] In one embodiment of the present disclosure, the solar power generation power prediction method further includes:

[0096] The weather type of the target area is determined based on the cloud cover and radiation intensity of the target area.

[0097] In this embodiment, the current weather type can be identified through meteorological data. Specifically, based on the cloud cover and solar radiation intensity, the weather type can be divided into three categories: the first weather type can be sunny, the second weather type can be cloudy, and the third weather type can be overcast. The model is expressed as:

[0098]

[0099] in, Indicates the weather type, is the cloud cover index, is the radiation intensity.

[0100] Specifically, the process of determining the weather type can be described as follows:

[0101] Determine weather in which the cloud cover is less than the first cloud cover and the radiation intensity is greater than the first radiation intensity as a first weather type;

[0102] Determine the weather where the cloud amount is between the first cloud amount and the second cloud amount and the radiation intensity is between the first radiation intensity and the second radiation intensity as the second weather type;

[0103] The weather in which the cloud cover is greater than the second cloud cover and the radiation intensity is less than the second radiation intensity is determined as the third weather type.

[0104] Among them, the first cloud amount, the second cloud amount, the first radiation intensity, and the second radiation intensity are all preset constants, and those skilled in the art can design their specific values according to actual needs.

[0105] Please refer to Figure 3 In one embodiment of the present disclosure, the first power generation and the second power generation are integrated based on the weather type of the target area, including:

[0106] Determine the corresponding fusion parameters based on the weather type of the target area;

[0107] fusing the first generated power and the second generated power based on a fusion parameter;

[0108] Among them, under the first weather type, the fusion parameter of the first power generation power is greater than the fusion parameter of the second power generation power; the first weather type refers to a weather type in which the cloud cover is less than the first cloud cover and the radiation intensity is greater than the first radiation intensity.

[0109] In this example, a separate linear fusion layer (which can be implemented using PyTorch's Linear layer) can be configured for each weather type to form an expert system architecture:

[0110]

[0111] in, Indicates the final predicted value of generated power, i.e. the third generated power; It means sunny day, Indicates cloudy, Indicates cloudy weather, 、 、 are the weight matrices corresponding to weather types, 、 、 is the corresponding bias term, represents the prediction result of the physical model, represents the prediction result of the deep learning model, Indicates that the prediction results of the physical model and the deep learning model are concatenated into a vector. ” indicates matrix multiplication.

[0112] By setting the corresponding weight matrix and bias terms, we can achieve:

[0113] (1) Under sunny conditions, the linear layer can learn to give higher weights to the physical model;

[0114] (2) Under cloudy and overcast conditions, the linear layer can learn the optimal fusion ratio.

[0115] The linear fusion layer parameters for each weather type are obtained by training the historical data of the corresponding weather type:

[0116]

[0117] in, is the weight matrix corresponding to the weather type, is the corresponding bias term, is the number of samples corresponding to the weather type, represents the final predicted value, which is the third power generated by fusing the first power and the second power. Indicates the actual value of generated power.

[0118] In this embodiment, during the prediction stage, the current weather type can be identified, and then the corresponding linear fusion layer model is selected, and the prediction results of the physical model and the deep learning model are input into the linear fusion layer model to obtain the final prediction value.

[0119] This multi-model fusion method based on weather classification has the following advantages:

[0120] (1) Advantages of expert systems: A specially optimized fusion model is used for each weather type;

[0121] (2) Strong model adaptability: Automatically select the optimal fusion strategy under different weather conditions;

[0122] (3) High prediction stability: avoids the performance degradation of a single model under unsuitable conditions;

[0123] (4) Simple and efficient implementation: The linear layer is implemented based on PyTorch, with low computational overhead.

[0124] like Figure 4 As shown, in one embodiment of the present disclosure, the deep learning model receives historical power generation data and meteorological data as input, and finally generates a deep learning prediction value through three main steps: feature extraction, time series modeling, and prediction output. The specific implementation process of each step is detailed below:

[0125] (1) Input feature fusion

[0126] The input of a deep learning model consists of two types of data:

[0127] (1.1) Historical power generation data: The historical power generation sequence containing time pattern information is expressed as:

[0128]

[0129] (1.2) Meteorological data: This includes key meteorological factors that affect power generation, such as temperature, humidity, and cloud cover. These heterogeneous data are integrated through the feature fusion layer to form a unified input tensor:

[0130]

[0131] in, Is the length of The historical power generation series, The present and the future Weather forecast data for each time step.

[0132] (2) Feature extraction network

[0133] Feature extraction uses a one-dimensional convolutional neural network with the following structure:

[0134] (2.1) One-dimensional convolutional layer: extracts features from the input sequence and maps the local pattern of the time dimension to the feature space:

[0135]

[0136] in, represents the convolution operation, and are the convolution kernel and bias.

[0137] (2.2) ReLU activation function: Introducing nonlinear transformation to enhance the model's expressive power:

[0138]

[0139] (2.3) Maximum pooling layer: reduces the dimension and extracts the most significant features, reducing the computational effort of subsequent processing:

[0140]

[0141] The advantage of convolutional networks is that they can automatically extract local temporal features and capture short-term patterns and trends in power generation series.

[0142] (3) Time Series Modeling Network

[0143] Time series modeling uses LSTM network combined with attention mechanism:

[0144] (3.1) LSTM layer: Capturing long-term dependencies in time series:

[0145]

[0146] in, is the hidden state, are unit states that together encode the temporal information of the sequence.

[0147] (3.2) Attention mechanism: Dynamically weight the importance of different time steps:

[0148]

[0149]

[0150]

[0151] in, It is The attention weight of time steps, is the context vector, which represents the weighted sequence representation.

[0152] The advantage of the LSTM network is its ability to capture long-term dependencies, while the attention mechanism enhances the model's perception of key time steps, making predictions more accurate.

[0153] (3) Prediction output network

[0154] The prediction output network consists of fully connected layers:

[0155] (3.1) Fully connected layer: maps the context vector to the prediction space:

[0156]

[0157] in, It's the future The power generation forecast value for each time step.

[0158] (4) Model training and optimization

[0159] The present invention adopts an end-to-end training method and uses a back-propagation algorithm to optimize the model parameters:

[0160] Loss function: Mean square error (MSE) is used as the main loss function:

[0161]

[0162] Optimizer: Use Adam optimizer to adaptively adjust the learning rate:

[0163]

[0164] in, are model parameters, is the learning rate, and are the first and second moment estimates.

[0165] Regularization: Implicit regularization is achieved through soft weight distribution of the attention mechanism to reduce the risk of overfitting.

[0166] The predicted values output by the deep learning model will be combined with the predicted values of the physical model through a linear fusion layer to form the final solar power generation power prediction result.

[0167] In summary, the solar power generation prediction method proposed in this embodiment achieves multiple technical breakthroughs and application value by integrating physical models with deep learning models, and has the following significant beneficial effects:

[0168] (1) Improved prediction accuracy

[0169] This invention adopts a hybrid architecture of "physical model + deep learning", which significantly improves the prediction accuracy compared with traditional single methods:

[0170] Reduced Root Mean Square Error (RMSE): The physical model and deep learning network independently predict, and the two prediction results are intelligently combined through a linear fusion layer, making the overall prediction closer to the actual power generation curve.

[0171] Reduced mean absolute percentage error (MAPE): Especially during transition periods such as sunrise and sunset, the prediction accuracy is significantly better than traditional methods.

[0172] Various evaluation indicators have been improved: the accuracy rate of a certain energy bureau, the southern region, and the national standard CR and QR indicators have all improved.

[0173] (2) Enhanced prediction stability and reliability

[0174] This invention significantly enhances the stability and reliability of predictions through physical constraints and high-quality day identification mechanisms:

[0175] Abnormal predictions are greatly reduced: The celestial motion constraints provided by the physical model effectively prevent overfitting and abnormal predictions of deep learning models.

[0176] Reduced forecast volatility: Compared with pure data-driven methods, the volatility and randomness of the forecast results are significantly reduced, which is more in line with the actual power generation laws.

[0177] Enhanced adaptability to extreme weather conditions: Under extreme weather conditions such as drastic changes in cloud cover and abnormal temperatures, the forecast performance is better than traditional methods.

[0178] (3) Computing efficiency and resource consumption optimization

[0179] The present invention optimizes the calculation process and improves prediction efficiency:

[0180] Computational time is significantly shortened: the physical model is simple and efficient, the deep learning network and the physical model are calculated in parallel, and the overall prediction process is efficient and coordinated.

[0181] Reduced training data requirements: The high-quality day recognition mechanism enables the model to effectively learn from a small amount of high-quality data, reducing the dependence on large-scale training data.

[0182] Reasonable optimization of model parameters: The deep learning module focuses on learning complex patterns from data, while the physical model processes basic laws. The two have a clear division of labor, and the overall system parameters are reasonably optimized.

[0183] (4) Improved explainability and traceability

[0184] The hybrid architecture of the present invention significantly improves the interpretability of prediction results:

[0185] Transparency in the prediction process: Prediction results can be decomposed into contributions from the physical model and the deep learning model. The weight distribution of the linear fusion layer clearly demonstrates the influence of each model, facilitating analysis of the source of prediction deviations.

[0186] The physical meaning of the parameters is clear: the key parameters of the physical model (such as atmospheric transmittance , photoelectric conversion parameters ) has a clear physical meaning, which is easy for engineers to understand and adjust.

[0187] Enhanced abnormality diagnosis capabilities: By comparing the deviation patterns between physical predictions and actual values, problems such as photovoltaic system failures and sensor anomalies can be effectively identified.

[0188] (5) Enhanced adaptability and generalization capabilities

[0189] The present invention has excellent adaptability and generalization ability:

[0190] Strong geographical adaptability: The physical model naturally adapts to the movement of the sun in different geographical locations and can be applied to new locations without retraining.

[0191] Seasonal adaptation: The rolling training strategy enables the model to automatically adapt to changes in power generation characteristics caused by seasonal changes.

[0192] Equipment aging adaptation: By dynamically optimizing mapping parameters, the model can adapt to long-term changing factors such as PV module aging and dust accumulation.

[0193] Strong cross-site generalization capability: In scenarios with newly built power plants or limited data, this method can still provide reliable predictions, and its generalization performance is better than that of purely data-driven methods.

[0194] (6) Improved engineering practicality

[0195] The present invention has significant engineering practical value:

[0196] Low deployment threshold: The physical model part only requires basic geographic information (including longitude and latitude information) to run, and the deep learning part can be flexibly configured according to data availability.

[0197] Low maintenance cost: The model structure is clear and the physical meaning of the parameters is clear, which makes it easy for engineers to maintain and optimize.

[0198] Strong scalability: The modular design of the framework allows for easy integration of new meteorological data sources or forecasting algorithms.

[0199] Wide range of application scenarios: It is suitable for photovoltaic power stations of different sizes and types, from distributed rooftop photovoltaics to large-scale ground power stations.

[0200] In summary, by integrating the physical constraints of celestial motion with deep learning technology, this invention has achieved significant breakthroughs in prediction accuracy, stability, computational efficiency, interpretability, adaptability, and engineering practicality. It provides an innovative solution for solar power generation prediction technology and is of great value in promoting the efficient utilization of new energy and the safe and stable operation of power grids.

[0201] Corresponding to the solar power generation power prediction method of the above embodiment, Figure 5 This is a structural block diagram of a solar power generation prediction device provided by an embodiment of the present disclosure. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. Figure 5 The solar power generation power prediction device 20 includes: a first prediction module 21, a second prediction module 22 and a data fusion module 23.

[0202] The first prediction module 21 is configured to predict a first power generation capacity of the solar power generation device using a physical model based on the physical parameters of the target area. The physical model is a mapping model between physical parameters and power generation capacity based on the motion laws of the solar celestial body and the mechanics of photovoltaic power generation. The target area is the area where the solar power generation device is located.

[0203] A second prediction module 22 is configured to predict a second power generation capacity of the solar power generation equipment using a deep learning model based on the meteorological data of the target area and the historical power generation data of the solar power generation equipment;

[0204] The data fusion module 23 is configured to fuse the first power generation power and the second power generation power based on the weather type of the target area to obtain a third power generation power of the solar power generation equipment.

[0205] In one embodiment of the present disclosure, the physical parameters of the target area include spatiotemporal parameters, atmospheric optical characteristic parameters, and photoelectric conversion parameters.

[0206] The physical model is used to determine the irradiance of the target area based on the spatiotemporal parameters and the atmospheric optical characteristic parameters, and to determine the first generated power based on the irradiance and the photoelectric conversion parameters;

[0207] Before predicting the first power generation power of the solar power generation equipment using the physical model based on the physical parameters of the target area, the first prediction module 21 is specifically used to:

[0208] Based on the error between the first generated power predicted by the physical model on the first historical date and the corresponding actual generated power, the atmospheric optical characteristic parameters and the photoelectric conversion parameters in the physical model are optimized.

[0209] In one embodiment of the present disclosure, the first prediction module 21 is further configured to:

[0210] The multiple historical dates are screened based on the correlation between the irradiance and the actual power generation values corresponding to the multiple historical dates to obtain a first historical date; the irradiance corresponding to the multiple historical dates is the irradiance of the target area on the multiple historical dates obtained based on the physical model, and the actual power generation value is the actual power generation value of the solar power generation equipment in the target area on the multiple historical dates;

[0211] Multiple historical dates are the N dates before the forecast date.

[0212] In one embodiment of the present disclosure, the spatiotemporal parameters include timestamp information, longitude information, and latitude information of the target area, and the atmospheric optical characteristic parameters include atmospheric transmittance. The first prediction module 21 is further configured to:

[0213] The irradiance of the target area is determined based on timestamp information, longitude information, latitude information, and atmospheric transmittance.

[0214] In one embodiment of the present disclosure, the first prediction module 21 is further configured to:

[0215] Determine the solar declination angle and solar hour angle of the target area based on the timestamp information, longitude information and latitude information of the target area;

[0216] Determine the zenith angle of the target area based on the latitude information, solar declination angle and solar hour angle of the target area;

[0217] The irradiance of the target area is determined based on the zenith angle and atmospheric transmittance of the target area.

[0218] In one embodiment of the present disclosure, the data fusion module 23 is specifically configured to:

[0219] The weather type of the target area is determined based on the cloud cover and radiation intensity of the target area.

[0220] In one embodiment of the present disclosure, the data fusion module 23 is specifically configured to:

[0221] Determine the corresponding fusion parameters based on the weather type of the target area;

[0222] fusing the first generated power and the second generated power based on a fusion parameter;

[0223] Among them, under the first weather type, the fusion parameter of the first power generation power is greater than the fusion parameter of the second power generation power; the first weather type refers to a weather type in which the cloud cover is less than the first cloud cover and the radiation intensity is greater than the first radiation intensity.

[0224] See also Figure 6 , Figure 6 This is a schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 6 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules / units in the above-mentioned device embodiments, such as Figure 5 The functions of the first prediction module 21, the second prediction module 22 and the data fusion module 23 are shown.

[0225] It should be understood that in the embodiments of the present disclosure, the processor 301 may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0226] The input device 302 may include a touchpad, a fingerprint collection sensor (for collecting user fingerprint information and fingerprint direction information), a microphone, etc. The output device 303 may include a display (LCD, etc.), a speaker, etc.

[0227] The memory 304 may include a read-only memory and a random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store device type information.

[0228] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present disclosure can execute the implementation methods described in the first and second embodiments of the solar power generation power prediction method provided by the embodiments of the present disclosure, and can also execute the implementation methods of the electronic device described in the embodiments of the present disclosure, which will not be repeated here.

[0229] In another embodiment of the present disclosure, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, all or part of the process of the method in the above embodiment is implemented. The computer program can also be used to instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments are implemented. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium.

[0230] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, the computer-readable storage medium can include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or is about to be output.

[0231] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0232] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0233] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or can be an electrical, mechanical or other form of connection.

[0234] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of these units may be selected based on actual needs to achieve the objectives of the embodiments of the present disclosure.

[0235] In addition, the functional units in the various embodiments of the present disclosure may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0236] The above are only specific embodiments of the present disclosure, but the scope of protection of the present disclosure is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or replacements within the technical scope disclosed in this disclosure, and such modifications or replacements should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure should be based on the scope of protection of the claims.

Claims

1. A solar power generation power prediction method, characterized in that: include: Based on the physical parameters of the target area, a physical model is used to predict the first power generation power of the solar power generation device; the physical model is a mapping model between physical parameters and power generation power constructed based on the movement law of the solar celestial body and the mechanics of photovoltaic power generation, and the target area is the area where the solar power generation device is located; Predicting a second power generation capacity of the solar power generation equipment using a deep learning model based on meteorological data of a target area and historical power generation data of the solar power generation equipment; fusing the first generated power and the second generated power based on a weather type of a target area to obtain a third generated power of the solar power generation device; The fusing of the first power generation and the second power generation based on the weather type of the target area includes: Determining corresponding fusion parameters based on the weather type of the target area; fusing the first generated power and the second generated power based on the fusion parameter; Wherein, in a first weather type, the fusion parameter of the first power generation is greater than the fusion parameter of the second power generation; the first weather type refers to a weather type in which the cloud cover is less than a first cloud cover and the radiation intensity is greater than a first radiation intensity; The physical parameters of the target area include spatiotemporal parameters, atmospheric optical characteristic parameters, and photoelectric conversion parameters, and the physical model is used to determine the irradiance of the target area based on the spatiotemporal parameters and the atmospheric optical characteristic parameters, and to determine the first generated power based on the irradiance and the photoelectric conversion parameters; Before predicting the first power generation power of the solar power generation equipment using the physical model based on the physical parameters of the target area, the method further includes: optimizing the atmospheric optical characteristic parameters and the photoelectric conversion parameters in the physical model based on an error between a first generated power predicted by the physical model on a first historical date and the corresponding actual generated power; Also includes: Filtering multiple historical dates based on the correlation between the irradiance and the actual power generation values corresponding to the multiple historical dates to obtain the first historical date; the irradiance corresponding to the multiple historical dates is the irradiance of the target area on the multiple historical dates obtained based on the physical model, and the actual power generation value is the actual power generation value of the solar power generation equipment in the target area on the multiple historical dates; The multiple historical dates are the N dates before the predicted date.

2. The solar power generation power prediction method according to claim 1, wherein: The spatiotemporal parameters include timestamp information, longitude information, and latitude information of the target area; the atmospheric optical characteristic parameters include atmospheric transmittance; and the physical model is specifically used for: The irradiance of the target area is determined based on the timestamp information, the longitude information, the latitude information, and the atmospheric transmittance.

3. The solar power generation power prediction method according to claim 2, wherein: The determining the irradiance of the target area based on the timestamp information, the longitude information, the latitude information, and the atmospheric transmittance includes: Determine the solar declination angle and solar hour angle of the target area based on the timestamp information, longitude information and latitude information of the target area; Determining the zenith angle of the target area based on the latitude information, the solar declination angle, and the solar hour angle of the target area; The irradiance of the target area is determined based on the zenith angle and atmospheric transmittance of the target area.

4. The solar power generation power prediction method according to claim 1, wherein: Also includes: The weather type of the target area is determined based on the cloud cover and radiation intensity of the target area.

5. A solar power generation power prediction device, characterized in that: include: a first prediction module for predicting a first power generation capacity of the solar power generation device using a physical model based on physical parameters of a target area; the physical model is a mapping model of physical parameters and power generation capacity constructed based on the movement laws of the solar celestial body and the mechanics of photovoltaic power generation; and the target area is the area where the solar power generation device is located; a second prediction module, configured to predict a second generated power of the solar power generation device using a deep learning model based on meteorological data of a target area and historical power generation data of the solar power generation device; a data fusion module, configured to fuse the first power generation and the second power generation based on a weather type of a target area to obtain a third power generation of the solar power generation device; The data fusion module is specifically used for: Determine the corresponding fusion parameters based on the weather type of the target area; fusing the first generated power and the second generated power based on a fusion parameter; Wherein, under the first weather type, the fusion parameter of the first power generation is greater than the fusion parameter of the second power generation; the first weather type refers to a weather type in which the cloud cover is less than the first cloud cover and the radiation intensity is greater than the first radiation intensity; The physical parameters of the target area include temporal and spatial parameters, atmospheric optical characteristic parameters and photoelectric conversion parameters. The physical model is used to determine the irradiance of the target area based on the spatiotemporal parameters and the atmospheric optical characteristic parameters, and to determine the first generated power based on the irradiance and the photoelectric conversion parameters; Before predicting the first power generation power of the solar power generation equipment using the physical model based on the physical parameters of the target area, the first prediction module is specifically used to: Optimizing atmospheric optical characteristic parameters and photoelectric conversion parameters in the physical model based on an error between a first generated power predicted by the physical model on a first historical date and the corresponding actual generated power; The first prediction module is further configured to: The multiple historical dates are screened based on the correlation between the irradiance and the actual power generation values corresponding to the multiple historical dates to obtain a first historical date; the irradiance corresponding to the multiple historical dates is the irradiance of the target area on the multiple historical dates obtained based on the physical model, and the actual power generation value is the actual power generation value of the solar power generation equipment in the target area on the multiple historical dates; Multiple historical dates are the N dates before the forecast date.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. 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 method according to any one of claims 1 to 4 are implemented.

Images