Photovoltaic generating capacity prediction method and device based on intelligent learning model

Abstract

The invention relates to a photovoltaic generating capacity prediction method and device based on an intelligent learning model, and the method comprises the steps: obtaining the meteorological data and the radiation power data of a target region, and enabling the meteorological data and the radiation power data to serve as a feature value and a target value respectively; performing trend analysis and distribution analysis on the characteristic value and the target value, and preprocessing the characteristic value based on an analysis result; constructing a training data set based on the preprocessed feature value and target value; training a long short-term memory neural network based on the training data set; and inputting the real-time meteorological data of the target area into the trained long-short-term memory neural network to obtain a predicted value of the photovoltaic power generation capacity. According to the invention, the intrinsic law of the meteorological data and the photovoltaic power generation is learned through the long-short-term memory neural network, so that the data requirement is reduced, and the photovoltaic power generation prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power calculation and deep learning, and particularly relates to a photovoltaic power generation prediction method and device based on an intelligent learning model. Background Art

[0002] Photovoltaic power generation occupies an important position in the new energy field due to its excellent flexibility and scalability. Photovoltaic power generation prediction can help energy planners better plan and manage the utilization of photovoltaic resources to ensure stable output of photovoltaic power generation. When the weather is sunny, solar energy resources should be utilized as much as possible to generate electricity, and the excess electricity should be stored; when sunlight is insufficient, other power generation methods should be planned to make up for the photovoltaic power generation gap to avoid power supply shortages and ensure power supply-demand balance. Therefore, accurate photovoltaic power generation prediction is very important when formulating a photovoltaic power generation plan.

[0003] Currently, models commonly used for predicting photovoltaic power generation include the radiation balance model (RBM), the radiation transfer theory model (RTT), etc. These technologies use the principles of atmospheric physics and meteorology to establish models for predicting meteorological factors such as solar radiation, cloud cover, and temperature that have important impacts on photovoltaic power generation; they usually combine a light model, a cloud cover model, and a temperature model to comprehensively predict photovoltaic power generation and provide relatively accurate prediction values. However, the currently popular photovoltaic power generation prediction technologies still have some defects. For example, weather models usually require parameterization of physical processes in the atmosphere and on the ground, and certain errors will be generated during the processing; moreover, they involve knowledge of multiple disciplines, the model itself is relatively complex, and a large amount of computing resources and time are required. Therefore, it is very necessary to explore a more efficient and stable photovoltaic power generation prediction model. Summary of the Invention

[0004] To improve the accuracy of photovoltaic power generation prediction and reduce the model complexity, in the first aspect of the present invention, a photovoltaic power generation prediction method based on an intelligent learning model is provided, including: obtaining meteorological data and radiation power data of a target area, and using the meteorological data and the radiation power data as feature values and target values respectively; performing trend analysis and distribution analysis on both the feature values and the target values, and preprocessing the feature values based on the analysis results; constructing a training data set based on the preprocessed feature values and target values; training a long short-term memory neural network based on the training data set; and inputting the real-time meteorological data of the target area into the trained long short-term memory neural network to obtain a predicted value of photovoltaic power generation.

[0005] In some embodiments of the present invention, the preprocessing of the meteorological data and the radiation power data based on the results of the trend analysis and the distribution analysis includes: converting the time eigenvalue in the eigenvalues into a numerical type; filtering out the eigenvalues with a correlation lower than a preset value based on the correlation between the eigenvalues and the target value; and normalizing the filtered eigenvalues and the target value.

[0006] Further, the converting the time eigenvalue in the eigenvalues into a numerical type includes: performing regularization decomposition on the time eigenvalue to obtain the hour component and the minute component of the time eigenvalue; and converting the time eigenvalue into a numerical type that fluctuates with time through the hour component and the minute component according to a preset coefficient and a bias.

[0007] In some embodiments of the present invention, the long short-term memory neural network includes: an LSTM network for obtaining the training data of the training data set; and a neural network for extracting a plurality of eigenvalues from the training data and predicting the target value according to a loss function.

[0008] Further, the neural network includes a plurality of linear layers with activation functions, and the input size of each linear layer is half of the output size of the previous linear layer.

[0009] In the above embodiments, the performing trend analysis and distribution analysis on both the eigenvalues and the target value includes: drawing a trend graph according to the time change trend of the eigenvalues; drawing a distribution graph according to the time change distribution of the target value; drawing a distribution graph of the eigenvalues on the target value based on the trend graph and the distribution graph, and judging the correlation between the eigenvalues and the target value.

[0010] In a second aspect of the present invention, there is provided a photovoltaic power generation prediction system based on an intelligent learning model, including: an acquisition module for acquiring meteorological data and radiation power data of a target area, and respectively using the meteorological data and the radiation power data as eigenvalues and a target value; a construction module for performing trend analysis and distribution analysis on both the eigenvalues and the target value, and preprocessing the eigenvalues based on the analysis results; constructing a training data set based on the preprocessed eigenvalues and the target value; a training module for training a long short-term memory neural network based on the training data set; and a prediction module for inputting the real-time meteorological data of the target area into the trained long short-term memory neural network to obtain a predicted value of the photovoltaic power generation.

[0011] Further, the construction module includes: a conversion unit for converting the time eigenvalue in the eigenvalues into a numerical type; a filtering unit for filtering out the eigenvalues with a correlation lower than a preset value based on the correlation between the eigenvalues and the target value; and a normalization unit for normalizing the filtered eigenvalues and the target value.

[0012] In a third aspect of the present invention, there is provided an electronic device, comprising: one or more processors; a storage device for storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the photovoltaic power generation prediction method based on an intelligent learning model provided in the first aspect of the present invention.

[0013] In a fourth aspect of the present invention, there is provided a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the photovoltaic power generation prediction method based on an intelligent learning model provided in the first aspect of the present invention.

[0014] The beneficial effects of the present invention are as follows: The present invention adopts a long short-term memory network, which can make full use of information such as trends and cycles in historical data; it can also learn and represent complex non-linear relationships, and is suitable for dealing with non-linear characteristics and mutual influences that may exist in weather data.

[0015] The long short-term memory neural network has strong adaptability. Through the backpropagation algorithm, the artificial neural network can continuously adjust the weights and biases according to the input data, enabling the model to adapt to different weather conditions and having good generalization ability. When dealing with unknown weather conditions, the generalization ability of other models may not be as strong as that of the artificial neural network, especially when facing some sudden weather events.

[0016] Other models may have higher requirements for data quality and need accurate and complete weather data to train and verify the model. Poor data quality may affect the model effect, which may pose certain challenges in practical applications. The artificial neural network can use hardware such as GPUs for large-scale parallel computing to accelerate the model training and inference processes. During the continuous traversal of data, it can find the latent laws between data, ignore the influence of individual data with relatively large errors, and make the results more accurate; The present invention provides a new idea and a new platform for photovoltaic power generation prediction. In the verification set data, the coefficient of determination between the predicted data and the actual measured data is 97.16%, which can well fit the actual curve of photovoltaic power generation, providing a new method for photovoltaic power generation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the basic process of the photovoltaic power generation prediction method based on an intelligent learning model in some embodiments of the present invention; Figure 2 is a schematic diagram of the specific process of the photovoltaic power generation prediction method based on an intelligent learning model in some embodiments of the present invention; Figure 3The variation trend chart of the target value Solar_Rad within 72 hours in some embodiments of the present invention.

[0018] Figure 4 The polar coordinate diagram of the characteristic values Wind_Dir and Hi_Dir in some embodiments of the present invention.

[0019] Figure 5 The change diagram and box plot of rainfall and rainfall rate in some embodiments of the present invention.

[0020] Figure 6 The interface diagram of the CSV file uploaded by users in some embodiments of the present invention.

[0021] Figure 7 The online data visualization interface diagram in some embodiments of the present invention.

[0022] Figure 8 The comparison diagram of the actual solar radiation and the model prediction results in the validation set in some embodiments of the present invention.

[0023] Figure 9 In some embodiments of the present invention, the actual results and the predicted results in the validation set Distribution diagram and scatter plot.

[0024] Figure 10 The structural schematic diagram of the photovoltaic power generation prediction device based on an intelligent learning model in some embodiments of the present invention; Figure 11 The structural schematic diagram of an electronic device in some embodiments of the present invention. Detailed implementation manners

[0025] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not intended to limit the scope of the present invention.

[0026] Reference Figure 1 And Figure 2 , in the first aspect of the present invention, a method for predicting photovoltaic power generation based on an intelligent learning model is provided, including: S100. Obtain meteorological data and radiation power data of a target area, and use the meteorological data and the radiation power data as characteristic values and target values respectively; S200. Perform trend analysis and distribution analysis on both the characteristic values and the target values, and preprocess the characteristic values based on the analysis results; Based on the preprocessed characteristic values and target values, construct a training data set; S300. Train a long short-term memory neural network based on the training data set; S400. Input the real-time meteorological data of the target area into the trained long short-term memory neural network to obtain a predicted value of photovoltaic power generation.

[0027] In step S100 of some embodiments of the present invention, meteorological data and radiation power data of the target area are acquired, and the meteorological data and radiation power data are used as eigenvalue and target value respectively; Specifically, relevant data of meteorology and photovoltaic power generation are acquired, and the potential law between meteorological factors and photovoltaic power generation amount in the target area is explored. The eigenvalues of the experimental data used are time, outdoor temperature, outdoor relative humidity, dew point temperature, wind speed, wind direction, wind chill index, heat index, temperature-humidity index, temperature-humidity-sun index, air pressure, rainfall, rainfall rate, a total of thirteen eigenvalues; the target value is solar radiation power, with the unit of watt per square meter (W / m²), which describes the power received on a unit area of solar panel.

[0028] The dataset records meteorological data and radiation power every minute. 60 data points can be collected in one hour, and 1440 data points in one day. In the same month, if the weather doesn't change much, we can observe the periodicity between every 1440 data points, that is, some eigenvalues or target values between one day and the next day. We can use this periodicity to adjust some eigenvalues. The dataset obtained from the website has already specified the training dataset and the validation dataset for us, which are 50,000 data points and 40,000 data points respectively. The data used in the training dataset are 49,998 data from January 1, 2010 to February 4, 2010, and the data used in the validation dataset are 44,636 data from December 1, 2010 to January 1, 2011. Although they are not continuous, there is an annual periodicity. If this area is located in the Northern Hemisphere, it is in winter during these two time periods. The recurrent neural network model is trained using the training set, and the hyperparameters of the model are adjusted and the model's capabilities are evaluated using the validation set.

[0029] The eigenvalue names include: Time, time, using the 12-hour clock method.

[0030] Temp_Out, outdoor temperature, referring to the measured environmental temperature, usually in degrees Celsius (°C) or degrees Fahrenheit (°F).

[0031] Out_Hum, outdoor relative humidity, representing the percentage of water vapor content in the air relative to the current temperature.

[0032] Dew_Pt, dew point temperature, representing the temperature at which the air needs to be cooled to reach the condensation point when saturated. It is the temperature when the water vapor in the air cools to saturation.

[0033] Wind_Speed, wind speed, representing the movement speed of the wind, usually in miles per hour (mph) or meters per second (m / s).

[0034] Wind_Dir, Wind direction, refers to the direction in which the wind blows, usually expressed using angles or azimuths (such as north, south, east, west).

[0035] Hi_Dir, Wind direction, similar to Wind_Dir, represents the direction in which the wind blows and may be the wind direction value from another measurement or source.

[0036] Wind_Chill, Wind chill index, is used to describe the perceived temperature under the combined effect of wind speed and temperature. It is the relatively lower temperature felt when the wind speed is high.

[0037] Heat_Index, Heat index, is used to describe the perceived temperature under the combined effect of high temperature and humidity. It is the relatively higher temperature felt in a high-temperature and high-humidity environment.

[0038] THW_Index, Temperature-Humidity Index, combines temperature, humidity, and wind speed and is used to measure the human perception of heat.

[0039] THSW_Index, Temperature-Humidity-Solar Index, is similar to THW_Index but also takes into account the influence of solar radiation.

[0040] Bar, Atmospheric pressure, represents the atmospheric pressure, usually measured in pascals (Pa) or inches of mercury (inHg).

[0041] Rain, Rainfall, represents the amount of precipitation measured over a certain period of time, usually measured in millimeters (mm) or inches (in).

[0042] Rain_Rate, Rainfall rate, represents the speed of precipitation per unit time, usually measured in millimeters per hour (mm / h) or inches per hour (in / h).

[0043] Solar_Rad, Solar radiation, represents the intensity and distribution of solar energy, usually measured in watts per square meter (W / m²). This value can be used to measure the intensity of solar radiation and its impact on the environment.

[0044] Intercept data samples: Randomly select data for three days on the training dataset, that is, 4320 data points for analysis. In this case, the data from January 1, 2010 to January 3, 2010 is used.

[0045] Reference Figures 3 to 5 , in step S200 of some embodiments of the present invention, the performing trend analysis and distribution analysis on both the eigenvalue and the target value includes: S201. Draw a trend graph based on the time variation trend of the eigenvalue; Specifically, first draw a time trend chart of the target value Solar_Rad, such as Figure 2 As shown. Every 24 hours, we draw a red vertical line to distinguish each day, so that we can observe the periodicity. From the picture, we can see that there is basically no solar radiation at night. During the day, if the weather is clear, the solar radiation shows a shape similar to a quadratic curve. However, if the meteorological factors change suddenly, it will cause great fluctuations in the trend of solar radiation.

[0046] When drawing a wind speed and direction chart, the data given for wind direction is the angle data centered at the measurement point. We draw a radar chart in polar coordinates, and the results show that most of the wind directions here fall between 315° and 45°. S202. Draw a distribution graph based on the time variation distribution of the target value; like Figure 5 As shown, the obtained box plot is drawn to observe the distribution of the characteristic value itself, whether there are outliers and some other characteristics of the data itself. In the box plots of Rain rainfall and Rain_Rate rainfall rate, it can be observed that there are a large number of outliers; combined with the actual situation, there is only a small part of the time in the three days here, and there is no rainfall for most of the time, resulting in a very small upper limit of the box plot, and almost all the data of rainfall and rainfall rate are outliers.

[0047] S203. Based on the trend graph and the distribution graph, draw a distribution graph of the characteristic value on the target value, and determine the correlation between the characteristic value and the target value.

[0048] Specifically, we draw a distribution graph of the eigenvalues ​​on the target values. It can be observed from the graph that most of the eigenvalues ​​have a certain positive correlation with the target values; while the two eigenvalues ​​of rainfall and rainfall rate have a strong negative correlation with the target value of solar radiation. This is understandable because when it rains, clouds usually block solar radiation, resulting in a decrease in solar radiation received by the ground. In addition, the two eigenvalues ​​of wind speed and wind direction are considered to be eliminated because their measured values ​​are discrete and their distribution is not closely related to the target value.

[0049] In step S200 of some embodiments of the present invention, the preprocessing of the meteorological data and the radiation power data based on the results of the trend analysis and the distribution analysis includes: S204. Convert the time characteristic value in the characteristic value into a numerical type; Specifically, among the eigenvalues, only time uses the 12-hour clock notation and is of a non-data type. We use the datetime library in Python to perform regularization decomposition on it and obtain the numerical type hour time_obj.hour and minute time_obj.minute in the 24-hour clock notation.

[0050] Perform the following transformation on it to obtain the numerical type eigenvalue within a day:

[0051] The obtained time eigenvalue is of numerical type, and its value range is from 0 to 23.983.

[0052] However, in the internship training, the prediction results obtained by this data processing method are not satisfactory, so we adopt another processing method to replace the above method.

[0053]

[0054] Using the above processing method, the variation law of the eigenvalue time on the time scale is similar to a sine function; when the day just begins, the value of time is -12, and when it reaches 12 noon, the value of time is 0, reaching the maximum value. Then, as time goes by, the value of time gradually decreases. When it decreases to 12 midnight, the value of time returns to 0 again; this data processing method can effectively solve the mutation phenomenon of the time eigenvalue at midnight every day in terms of the training prediction results and can effectively improve the prediction accuracy.

[0055] S205. Based on the correlation between the eigenvalue and the target value, filter out the eigenvalues with a correlation lower than the preset value; Observing the correlation coefficients between each eigenvalue and the target value, it is found that the correlations between wind speed and wind direction and the target value are not strong, and it doesn't make much sense to normalize the wind direction eigenvalue; first, it is a discrete data, and second, it is the wind direction presented in angles. If the data collector of the validation dataset is different from its orientation, it will cause negative impacts. Therefore, we delete the eigenvalues of wind direction and wind speed.

[0056] S206. Normalize the filtered eigenvalues and the target value.

[0057] Specifically, perform min-max normalization on all eigenvalues and the target value to prevent gradient explosion during the process of inputting into the neural network for training, which may lead to training failure: Let be a column of eigenvalues, which contains data, is the total number of data, are respectively the maximum and minimum values in, and we perform the following operations on the data: , The new obtained in-situ replaces the previous , and perform the above operations on all eigenvalues and target values, and then perform the above operations on the data in the validation set.

[0058] In step S200 or S300 of some embodiments of the present invention, the long short-term memory neural network includes: an LSTM network for obtaining training data of a training data set; a neural network for extracting multiple eigenvalue from the training data and predicting a target value according to a loss function.

[0059] Specifically, during the process of building the network of the present invention, a two-layer LSTM network is used, followed by a neural network including multiple linear layers and an activation function ReLU, and finally a linear output layer. The input size of the first linear layer is 32, and its output size is 16, exactly half of its input. After that, the output of each layer is half of its input to gradually increase the depth of the network. Since there are only 50,000 pieces of data in the training data set, the data volume is not large, so we prevent overfitting by increasing the model depth.

[0060] The specific parameters of the model are shown in Table 1:

[0061] It can be seen from the above model the forward propagation process of data in the model. First, the input data is processed by the LSTM layer, and then the output is shape-transformed and the final output is obtained through the neural network and the linear layer.

[0062] During the actual training process, we also added an L2 regularization term for the model parameters to control the complexity of the model and avoid the occurrence of overfitting. Then the model and the model weights are synchronized to the database. The website background is built using flask, flask_sqlalchemy is used to connect and operate the database, and flask-cli is used to provide an interface for modifying the database to the outside. Each time the database is updated and the model and the model weights are added to the database, we retrieve whether this model information already exists in the database. If the model information exists, then skip this model because we have already added this model and there is no need to add it again. We only need to add the new models we added.

[0063] Embodiment 2 Refer to Figure 10, the second aspect of the present invention provides a photovoltaic power generation prediction system 1 based on an intelligent learning model, including: an acquisition module 11, configured to acquire meteorological data and radiation power data of a target area, and use the meteorological data and radiation power data as feature values and target values respectively; a construction module 12, configured to perform trend analysis and distribution analysis on both the feature values and the target values, and preprocess the feature values based on the analysis results; construct a training data set based on the preprocessed feature values and target values; a training module 13, configured to train a long short-term memory neural network based on the training data set; a prediction module 14, configured to input the real-time meteorological data of the target area into the trained long short-term memory neural network to obtain a predicted value of the photovoltaic power generation.

[0064] Further, the construction module 12 includes: a conversion unit, configured to convert the time feature value in the feature values into a numerical type; a filtering unit, configured to filter out the feature values with a correlation lower than a preset value based on the correlation between the feature values and the target values; a normalization unit, configured to normalize the filtered feature values and target values.

[0065] In an embodiment of the present invention, the system includes a data transmission module, a data visualization analysis module, a neural network module, and a prediction result evaluation module.

[0066] Data transmission module: used to receive the CSV format file uploaded to the platform server. After the server obtains the file uploaded by the user, it uses the timestamp of the file uploaded by the user to perform md5 encryption to generate the file name of the server-side CSV file, avoiding confusion caused by overwriting due to the user uploading files with the same name. We also open an interface to download the CSV format file stored on the server to the local. The user uploads the CSV file interface as Figure 6 shown.

[0067] Data visualization analysis module: After the server receives the file, it reads the file and organizes the read data into data that can be accepted by the front-end element-plus component el-table-v2. el-table-v2 supports virtual scrolling and can only render the content of the visible area when processing large amounts of data, rather than rendering all of it. This can significantly reduce page loading time and memory usage, and our data is often tens of thousands of data, which can significantly improve page performance. It also supports visual analysis of uploaded data and provides options for drawing line graphs, heat maps, scatter plots, and violin plots. Line graphs can be used to observe the changing trends of eigenvalues ​​over time. Heat maps allow us to intuitively see the correlation between each eigenvalue and the target value. Scatter plots can better observe the distribution of discrete eigenvalues ​​than line graphs. Violin plots can be used to observe the distribution of each variable and the kernel density curve; the online visualization heat map interface is as follows: Figure 7 shown.

[0068] Neural network module: used to predict the uploaded test set. Users can choose from a variety of neural networks such as MLP, LSTM, and GRU provided in the server database. For each different neural network, several good weight data are selected and updated in the database. Use the DataLoader provided in the pytorch library and the abstract class DataSet we implemented to divide the data uploaded by the user into the format required by the network input, input it into the network, and feed back the predicted value to the user; Prediction result evaluation module: It inherits the above neural network module and is used to evaluate the prediction results from different angles. The prediction interval viewed by the user is divided into half a day as the minimum unit, and evaluation indicators such as MSE, RMSE, and MAE are given; the trend of the prediction results and the actual results on the time scale are plotted on the same graph, such as Figure 8 As shown, it provides a more intuitive feeling and comparison; finally, the distribution histogram and scatter plot of the coefficient of determination (R²) of the predicted results and the actual results are given, as shown in Figure 9 shown.

[0069] In the validation set we selected, we intercepted 5760 data points from four days and divided them into intervals of 720 data points each. We calculated the value of each indicator in each interval, and then took the average of the indicator values ​​in each interval to obtain the performance of the indicator on the entire data. The specific evaluation indicators are shown in Table 2.

[0070]

[0071] The training and testing processes of the long short-term memory neural network model were both completed in the Python 3.9 and torch 1.13.1+cu116 environment. The back-end system was built in the flask 3.0.0 environment, and the front-end page was built under vue 3.2.13 and element-plus 2.3.5.

[0072] Example 3 Reference Figure 11 In the third aspect of the present invention, there is provided an electronic device, including: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the photovoltaic power generation prediction method based on the intelligent learning model in the first aspect of the present invention.

[0073] The electronic device 500 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 502 or the program loaded from the storage device 508 into the random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. The input / output (I / O) interface 505 is also connected to the bus 504.

[0074] Generally, the following devices may be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the electronic device 500 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 11 the electronic device 500 with various devices is shown, it should be understood that it is not required to implement or have all the shown devices. Instead, more or fewer devices may be implemented or had. Figure 11 Each block shown in

[0075] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program code for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-described functions defined in the methods of the embodiments of the present disclosure are performed. It should be noted that the computer-readable medium described in the embodiments of the present disclosure can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the embodiments of the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device. In the embodiments of the present disclosure, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and the computer-readable signal medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0076] The above computer-readable medium can be included in the above electronic device; or it can exist separately and not be assembled into the electronic device. The above computer-readable medium carries one or more computer programs, and when the above one or more programs are executed by the electronic device, the electronic device is caused to: Computer program code for performing the operations of the embodiments of the present disclosure may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, Python, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, execute as a stand-alone software package, partly on the user's computer and partly on a remote computer, or execute entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0077] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functions, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions denoted in the blocks may occur in a different order than that denoted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0078] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A photovoltaic power generation prediction method based on an intelligent learning model, characterized in that: include: Acquire meteorological data and radiation power data of the target area, and use the meteorological data and radiation power data as characteristic values ​​and target values ​​respectively; Performing trend analysis and distribution analysis on both the characteristic value and the target value, and preprocessing the characteristic value based on the analysis results; Construct a training data set based on the preprocessed feature values ​​and target values; Based on the training data set, training a long short-term memory neural network; The real-time meteorological data of the target area is input into the trained long short-term memory neural network to obtain the predicted value of photovoltaic power generation.

2. The photovoltaic power generation prediction method based on the intelligent learning model according to claim 1 is characterized in that: The preprocessing of the meteorological data and the radiation power data based on the results of the trend analysis and the distribution analysis comprises: Convert the time characteristic value in the characteristic value into a numerical type; Based on the correlation between the feature value and the target value, the feature values ​​with correlation lower than the preset value are filtered out; Normalize the filtered feature values ​​and target values.

3. The photovoltaic power generation prediction method based on the intelligent learning model according to claim 2 is characterized in that: The converting of the time characteristic value in the characteristic value into a numerical type comprises: Regularize and decompose the time characteristic value to obtain the hour component and minute component of the time characteristic value; According to the preset coefficient and the offset, the time characteristic value is converted into a numerical type that fluctuates with time through the hour component and the minute component.

4. The photovoltaic power generation prediction method based on the intelligent learning model according to claim 1 is characterized in that: The long short-term memory neural network comprises: LSTM network, used to obtain training data for the training dataset; A neural network is used to extract multiple feature values ​​from the training data and predict the target value based on the loss function.

5. The photovoltaic power generation prediction method based on the intelligent learning model according to claim 4 is characterized in that: The neural network includes multiple linear layers with activation functions, and the input size of each linear layer is half of the output size of the previous linear layer.

6. The photovoltaic power generation prediction method based on the intelligent learning model according to claim 1 is characterized in that: The performing trend analysis and distribution analysis on both the characteristic value and the target value comprises: Draw a trend graph based on the time variation trend of the characteristic value; Draw a distribution graph based on the time-varying distribution of the target value; Based on the trend graph and the distribution graph, a distribution graph of the characteristic value on the target value is drawn, and the correlation between the characteristic value and the target value is determined.

7. A photovoltaic power generation prediction system based on an intelligent learning model, characterized in that: include: An acquisition module is used to acquire meteorological data and radiation power data of a target area, and use the meteorological data and radiation power data as characteristic values ​​and target values ​​respectively; A construction module is used to perform trend analysis and distribution analysis on both the characteristic value and the target value, and preprocess the characteristic value based on the analysis results; Construct a training data set based on the preprocessed feature values ​​and target values; A training module, used for training a long short-term memory neural network based on the training data set; The prediction module is used to input the real-time meteorological data of the target area into the trained long short-term memory neural network to obtain the predicted value of photovoltaic power generation.

8. According to the photovoltaic power generation prediction method based on the intelligent learning model according to claim 7, the building module comprises: A conversion unit, used for converting the time characteristic value in the characteristic value into a numerical type; A filtering unit, configured to filter out feature values ​​whose correlation is lower than a preset value based on the correlation between the feature value and the target value; The normalization unit is used to normalize the filtered feature values ​​and target values.

9. An electronic device, comprising: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement the photovoltaic power generation prediction method based on the intelligent learning model as described in any one of claims 1 to 6.

10. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the photovoltaic power generation prediction method based on the intelligent learning model as described in any one of claims 1 to 6 is implemented.

Images