Photovoltaic power generation power prediction method and related device
By integrating meteorological and power generation data, using interquartile range method and interpolation method to process abnormal and missing values, and building an optimized LSTM model, the problem of insufficient modeling of traditional photovoltaic power generation prediction is solved, and higher prediction accuracy and data continuity are achieved.
Patent Information
- Application Number
- CN202510508064.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-22
- Publication Date
- 2025-07-22
AI Technical Summary
Traditional photovoltaic power generation prediction methods lack the ability to model the nonlinearity and long-range dependence of photovoltaic power generation, and ignore the outliers and missing problems in the data acquisition process, resulting in low prediction accuracy.
Integrate meteorological data and power generation data, use the interquartile distance method to eliminate outliers, and add missing values to the cubic spline interpolation method to build a photovoltaic power generation prediction model based on LSTM, combining adaptive momentum optimization and Dropout layer to optimize network parameters.
It improves the accuracy of photovoltaic power generation prediction, reduces data errors, and enhances the modeling ability of long-term and short-term dependencies. The RMSE has dropped to 2.3%, and the MAE has dropped to 1.8%, an increase of about 44% compared with traditional methods.
Smart Images
Figure CN120357450A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of new energy power generation, and particularly to a photovoltaic power prediction method and related devices. Background Art
[0002] With the rapid development of the global new energy industry, photovoltaic power generation has become an important part of renewable energy power generation. However, due to the influence of various factors such as weather conditions, seasonal changes, and geographical locations, the output power of photovoltaic power generation systems shows significant volatility and uncertainty.
[0003] Traditional prediction methods, including physical models, statistical models, and machine learning models, etc., although can improve the accuracy of photovoltaic power prediction to a certain extent, have weak modeling capabilities for the strong non-linearity and long-range dependence (such as diurnal cycle, seasonal change) of photovoltaic power generation; secondly, traditional prediction models usually directly adopt original meteorological and power generation data, ignoring outliers and missing problems in the data collection process. Existing technologies mostly use simple linear interpolation or mean filling methods to handle missing values, but such methods cannot adapt to the non-linear characteristics of photovoltaic data, resulting in insufficient continuity and authenticity of the interpolated data sequence. Summary of the Invention
[0004] Aiming at the problems mentioned in the prior art, the present invention proposes a photovoltaic power prediction method and related devices, which integrate meteorological data and photovoltaic power generation data, and combine probability statistical analysis and long short-term memory network (LSTM) model to achieve accurate prediction of photovoltaic power generation, and solve the problems of low data quality, incomplete feature representation, and weak modeling ability in the prior art.
[0005] To achieve the above object, the present invention adopts the following technical solutions.
[0006] In a first aspect, a photovoltaic power prediction method of the present invention includes the following steps: Obtain meteorological data and power generation data, and preprocess the meteorological data and power generation data; Based on the processed meteorological data and power generation data, construct a probability density function of power generation, and analyze the probability density function to obtain the frequency domain characteristics of power; Construct an LSTM-based photovoltaic power prediction model according to the frequency domain characteristics of power; Input the processed meteorological data and power generation data into the trained photovoltaic power prediction model for photovoltaic power prediction.
[0007] As a further improvement of the present invention, the meteorological data includes light intensity, temperature, wind speed and direction, cloud amount, and humidity; The power generation data includes current, voltage, active power, and grid frequency.
[0008] As a further improvement of the present invention, the preprocessing process is to clean the collected data, remove abnormal values and missing values, supplement the missing data by interpolation, and ensure data continuity.
[0009] As a further improvement of the present invention, the interquartile range method is used to eliminate outliers; Cubic spline interpolation method was used to fill in the missing values.
[0010] As a further improvement of the present invention, the process of obtaining the frequency domain characteristics of power is: Probabilistic statistical analysis is performed on the processed meteorological data and power generation data to obtain the probability density function curve of photovoltaic power generation. The multi-peak distribution characteristics of photovoltaic power are identified based on the probability density function curve, and the power distribution is correlated with the time period change of light intensity. According to the results of power spectrum analysis, the dynamic range of power fluctuation amplitude changing with solar radiation intensity is determined.
[0011] As a further improvement of the present invention, the process of constructing a photovoltaic power generation prediction model includes: A stacked LSTM network structure is used, the input layer contains meteorological parameters and power generation data, and the output layer is the power prediction value of the future time step; An adaptive momentum optimization algorithm is used to dynamically adjust the learning rate and weight parameters of the network.
[0012] As a further improvement of the present invention, the optimization of the photovoltaic power generation prediction model includes introducing a Dropout layer in the LSTM network to prevent overfitting; The number of hidden layer neurons, batch size, and learning rate are optimized through grid search to minimize the RMSE and MAE indicators.
[0013] In a second aspect, the present invention provides a photovoltaic power generation power prediction system, comprising: An acquisition module is used to acquire meteorological data and power generation data, and pre-process the meteorological data and power generation data; An analysis module is used to construct a probability density function of power generation based on the processed meteorological data and power generation data, and to analyze the probability density function to obtain the frequency domain characteristics of power; A building module is used to construct a photovoltaic power prediction model based on LSTM according to the frequency domain characteristics of power.
[0014] In a third aspect, the present invention provides a photovoltaic power generation prediction device, comprising a processor and a memory, wherein the processor implements the photovoltaic power generation prediction method as described above when executing a computer program stored in the memory.
[0015] In a fourth aspect, the present invention provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the photovoltaic power generation power prediction method as described above.
[0016] Compared with the prior art, the present invention has achieved the following technical effects: The present invention integrates multi-dimensional meteorological data and power generation data to construct a unified standardized data set, eliminates the deviation of a single data source, and eliminates outliers in the data through the interquartile range method, and combines the interpolation method to supplement missing data, thereby solving the data distortion problem caused by traditional linear interpolation and ensuring the continuity and reliability of the data series. The continuity error of the preprocessed data series is reduced to less than 2%.
[0017] The present invention reveals the multi-peak distribution characteristics of photovoltaic power (such as multi-peaks during the day and single peak at night) by constructing a probability density function of generated power, and clarifies the correlation between power distribution and light intensity. It can accurately reflect the dynamic correlation between light intensity and power output, and extract the main frequency component of power fluctuations through fast Fourier transform, providing the prediction model with frequency domain feature input with clear physical meaning.
[0018] The present invention adopts a stacked LSTM network structure, which can capture features and time dependencies. It combines adaptive momentum optimization with the Dropout layer to enable the model to learn more robust feature representations, thereby improving its generalization ability on the test set. The ability to model the long-term and short-term dependencies of photovoltaic power is significantly enhanced. After optimizing the hyperparameters using the grid search method, the root mean square error (RMSE) of the model on the test set is reduced to 2.3%, and the mean absolute error (MAE) is 1.8%, which is about 44% higher than the traditional prediction model. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a schematic diagram of the process of the photovoltaic power prediction method of the present invention; Figure 2 It is a schematic diagram of the photovoltaic power generation prediction system of the present invention; Figure 3 The figure is a schematic diagram of the structure of a computer device according to the present invention. DETAILED DESCRIPTION
[0020] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.
[0021] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments: like Figure 1As shown in the figure, the present invention proposes a photovoltaic power prediction method, which includes the following steps: Obtain meteorological data and power generation data, and preprocess the meteorological data and power generation data; the meteorological data includes light intensity, temperature, wind speed and direction, cloud cover, and humidity; the power generation data includes current, voltage, active power, and grid frequency.
[0022] Perform probability statistical analysis on the processed meteorological data and power generation data to obtain the probability density function curve of photovoltaic power generation. Identify the multi-peak distribution characteristics of photovoltaic power according to the probability density function curve, and combine the power distribution with the hourly change of light intensity. According to the power spectrum analysis results, determine the dynamic range of the power fluctuation amplitude with the change of solar radiation intensity.
[0023] Construct an LSTM-based photovoltaic power prediction model according to the frequency domain characteristics of power. When establishing the model, a stacked LSTM network structure is used. The input layer includes meteorological parameters and power generation data, and the output layer is the power prediction value for future time steps. The adaptive momentum optimization algorithm is used to dynamically adjust the learning rate and weight parameters of the network.
[0024] Input the processed meteorological data and power generation data into the trained photovoltaic power prediction model for photovoltaic power prediction.
[0025] Step 1: Data collection and preprocessing In this embodiment, the monitoring systems of 5 meteorological stations and 10 photovoltaic power stations in a certain province are used as data sources to collect multi-dimensional data from 2018 to 2023, including: Meteorological data: hourly light intensity (unit: W / m²), temperature (°C), humidity (%), wind speed (m / s); Power generation data: real-time power generation of photovoltaic power stations (unit: MW).
[0026] Data collection is realized through the API interface and Internet of Things sensors to ensure real-time and synchronization. For example, the monitoring system of a certain photovoltaic power station uploads power generation data once an hour, and meteorological station data is obtained through the provincial meteorological data platform.
[0027] The following problems generally exist in the original data, such as outliers: for example, the light intensity suddenly increases to 2000 W / m² during a thunderstorm (normal range 0 - 1200 W / m²); missing values: data missing due to sensor failures or communication interruptions (average missing rate is about 3%), and data deviation may occur according to the geographical location differences between meteorological stations and power stations. Therefore, the spatial interpolation method is used in this embodiment to perform regional averaging on meteorological data to reduce local interference.
[0028] In this embodiment, the interquartile range method (IQR) is used to identify outliers. Specifically, the first quartile (Q1) is the value at the lower quarter of the dataset, that is, 25% of the data is less than or equal to it. The third quartile (Q3) is the value at the upper quarter of the dataset, that is, 75% of the data is less than or equal to it. Calculate the interquartile range (IQR): IQR = Q3 - Q1.
[0029] IQR reflects the dispersion degree of the middle 50% of the data. The smaller the value, the more concentrated the middle data is, and the larger the value, the more dispersed the middle data is. Determine the outlier range: Usually, values less than Q1 - 1.5 * IQR or greater than Q3 + 1.5 * IQR are regarded as outliers. This range can be adjusted according to the actual situation.
[0030] Substitute and calculate the statistic for each data column: In the power generation data, Q1 = 200 W / m², Q3 = 800 W / m², IQR = 600 W / m², and the normal range is 700 - 1700. Since the physical upper limit of the light intensity is 1200 W / m², only the data > 1700 W / m² is excluded.
[0031] In the embodiment, for asymmetric distribution data such as wind speed, the modified IQR method is used to exclude outliers.
[0032] In this embodiment, the cubic spline interpolation method is used to fill in the missing data, and the formula is as follows:
[0033] Sort the data by time, and construct a piecewise cubic polynomial with the timestamp as the independent variable; ensure the curve is smooth through the boundary conditions (natural spline: the second derivative is 0); The average error rate between the interpolated data and the measured value ≤ 1.5%.
[0034] To eliminate the influence of different dimensions, normalize the meteorological data and power data, and compress the range of the normalized data to [0, 1] for convenient model training.
[0035] Step 2. Power characteristic analysis Analyze the distribution law of the photovoltaic power generation through statistical methods. Specifically, use tools such as histograms to observe the distribution of the data, initially judge the unimodal, bimodal or multimodal characteristics of the data, and calculate the central tendency (such as mean, median) and dispersion tendency (such as variance, standard deviation) of the data to further understand the statistical characteristics of the data.
[0036] Based on the results of the probability statistical analysis, use methods such as kernel density estimation to fit the probability density function curve of the data. By observing the shape and characteristics of the probability density function curve, the distribution of the data and the multimodal characteristics can be more intuitively understood.
[0037] On the probability density function curve, observe whether there are multiple obvious peaks. If there are multiple peaks, it indicates that the data has a multi-modal distribution characteristic; further analyze the positions and heights of each peak to understand the photovoltaic power levels represented by different peaks and their occurrence frequencies.
[0038] Conduct power spectrum analysis on the power generation data to understand the frequency components and amplitude distributions of power fluctuations.
[0039] Through the power spectrum diagram, the power fluctuations at different frequencies can be visually observed. According to the results of the power spectrum analysis, combined with the change information of solar radiation intensity, determine the dynamic range of the power fluctuation amplitude with the change of solar radiation intensity. Analyze the change trends and laws of the power fluctuation amplitude under different solar radiation intensities, as well as possible influencing factors.
[0040] Data grouping: Segment the power data by hour (such as 0:00 - 1:00, 1:00 - 2:00, etc.), and calculate the power mean and standard deviation of each time period.
[0041] Use Kernel Density Estimation (KDE) to fit the power distribution curve, and select the Silverman criterion for the bandwidth; The results are shown as follows: Morning peak (8:00 - 10:00): The light intensity gradually increases, and the power density peak is 40% of the rated power; Afternoon peak (13:00 - 15:00): The solar radiation is the strongest, and the power density peak reaches 85% of the rated power; Night (20:00 - 6:00): The power density is concentrated in the low value area of 0 - 5%.
[0042] To quantify the frequency domain characteristics of power fluctuations, perform spectral analysis using the Fast Fourier Transform (FFT): Select power data for a continuous week (sampling frequency 1Hz), remove the trend term and then intercept a 10 - minute segment; calculate the power spectral density (PSD) and identify the main frequency components; Main frequency range: The main frequency from 13:00 - 15:00 in the afternoon is 0.6Hz, and the fluctuation period is about 1.67 seconds; Amplitude change: The fluctuation amplitude reaches ±15% of the rated power (for example, if the rated power of a power station is 100MW, the fluctuation range is 85 - 115MW).
[0043] Step 3: Construction and optimization of the LSTM prediction model Adopt a stacked LSTM network structure, and the specific configuration is as follows: Input layer: Contains meteorological data (light, temperature, humidity, wind speed, power) and power generation power in the past 24 hours, with a total of 5 features; Hidden layer: The first layer of LSTM: 128 neurons, returning the complete sequence; The second layer of LSTM: 64 neurons, returning the complete sequence; The third layer of LSTM: 32 neurons, only returning the output of the last time step; Dropout layer: The dropout rate is 0.2 to prevent overfitting; Output layer: Predicted values of power generation power for the next 6 hours (6 time steps).
[0044] Dataset division: Training set (70%): Data from 2018 to 2021; Validation set (15%): Historical data; Test set (15%): Historical data.
[0045] Use Adaptive Momentum Optimization (Adam), with an initial learning rate of 0.001, β1 = 0.9, β2 = 0.999; Loss function: Mean Squared Error (MSE).
[0046] Adopt the grid search method to traverse the following parameter combinations: Batch size: 32, 64, 128; Learning rate: 0.1, 0.01, 0.001; Number of training epochs: 100, 200, 300. Optimal parameter combination: Batch size = 64, Learning rate = 0.001, Number of training epochs = 200.
[0047] The evaluation results on the test set are as follows: RMSE: 2.3% (4.1% for the traditional RNN model); MAE: 1.8%; R²: 0.98, indicating a very high goodness of fit of the model.
[0048] Through the design and optimization of the above steps, the data preprocessing error rate of the present invention is reduced to 1.2%. The power characteristic analysis reveals the multi-peak distribution and frequency domain fluctuation law. The RMSE of the LSTM prediction model reaches 2.3%, which is 44% higher than the traditional method. This solution provides a complete solution for photovoltaic power prediction and has industry application value.
[0049] The present invention integrates multi-dimensional meteorological data and power generation data, constructs a unified standardized dataset, eliminates the bias of a single data source, and at the same time eliminates outliers in the data through the interquartile range method and supplements missing data by interpolation, solving the problem of data distortion caused by traditional linear interpolation, ensuring the continuity and reliability of the data sequence, and reducing the continuity error of the preprocessed data sequence to less than 2%.
[0050] By constructing the probability density function of the power generation, the present invention reveals the multi-peak distribution characteristics of the photovoltaic power (such as the multi-peak during the day and the single-peak at night), clarifies the correlation between the power distribution and the light intensity, accurately reflects the dynamic correlation between the light intensity and the power output, and extracts the main frequency components of the power fluctuation through the fast Fourier transform, providing the input of the frequency domain characteristics with clear physical meaning for the prediction model.
[0051] Based on the same inventive concept, the embodiment of the present invention also provides a photovoltaic power generation prediction system. Since the principle of solving problems by this photovoltaic power generation prediction system is similar to that of the aforementioned photovoltaic power generation prediction method, the implementation of this photovoltaic power generation prediction system can refer to the implementation of the photovoltaic power generation prediction method, and the repeated parts will not be described again.
[0052] In specific implementation, the photovoltaic power generation prediction system provided by the embodiment of the present invention specifically includes: An acquisition module, configured to acquire meteorological data and power generation data, and preprocess the meteorological data and power generation data; An analysis module, configured to construct a probability density function of the power generation based on the processed meteorological data and power generation data, and analyze the probability density function to obtain the frequency domain characteristics of the power; A construction module, configured to construct an LSTM-based photovoltaic power generation prediction model according to the frequency domain characteristics of the power.
[0053] Correspondingly, the embodiment of the present invention also provides a photovoltaic power generation prediction device, including a processor and a memory. Wherein, when the processor executes the computer program stored in the memory, the photovoltaic power generation prediction method provided by the embodiment of the present invention is implemented.
[0054] For a more specific process of the above method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be described herein again.
[0055] Correspondingly, the embodiment of the present invention also provides a computer-readable storage medium, configured to store a computer program. Wherein, when the computer program is executed by a processor, the above-mentioned photovoltaic power generation prediction method provided by the embodiment of the present invention is implemented.
[0056] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the systems, devices, and storage media disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.
[0057] Those skilled in the art may further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.
[0058] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.
[0059] Finally, it should also be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.
[0060] The above has introduced in detail the photovoltaic power prediction method, system, device and storage medium provided by the present invention. Specific examples are used herein to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.
Claims
1. A photovoltaic power generation prediction method, characterized in that, The following steps are involved: Acquire meteorological data and power generation data, and pre-process the meteorological data and power generation data; Based on the processed meteorological data and power generation data, the probability density function of power generation is constructed, and the frequency domain characteristics of power are obtained by analyzing the probability density function; According to the frequency domain characteristics of power, a photovoltaic power generation prediction model based on LSTM is constructed; The processed meteorological data and power generation data are input into the trained photovoltaic power generation prediction model to predict the photovoltaic power generation.
2. The photovoltaic power generation prediction method according to claim 1, characterized in that The meteorological data include light intensity, temperature, wind speed and direction, cloud cover and humidity; The power generation data includes current, voltage, active power and grid frequency.
3. The photovoltaic power generation prediction method according to claim 1, wherein The preprocessing process is to clean the collected data, remove outliers and missing values, supplement the missing data by interpolation, and ensure data continuity.
4. The photovoltaic power prediction method according to claim 3, characterized in that The interquartile range method was used to eliminate outliers; Cubic spline interpolation method was used to fill in the missing values.
5. The photovoltaic power generation prediction method according to claim 1, characterized in that The process of obtaining the frequency domain characteristics of power is: Probabilistic statistical analysis is performed on the processed meteorological data and power generation data to obtain the probability density function curve of photovoltaic power generation. The multi-peak distribution characteristics of photovoltaic power are identified based on the probability density function curve, and the power distribution is correlated with the time period change of light intensity. According to the results of power spectrum analysis, the dynamic range of power fluctuation amplitude changing with solar radiation intensity is determined.
6. The photovoltaic power generation prediction method according to claim 1, wherein The process of building a photovoltaic power generation prediction model includes: A stacked LSTM network structure is used, the input layer contains meteorological parameters and power generation data, and the output layer is the power prediction value of the future time step; An adaptive momentum optimization algorithm is used to dynamically adjust the learning rate and weight parameters of the network.
7. The method for predicting photovoltaic power generation according to claim 6, wherein The optimization of the photovoltaic power prediction model includes introducing a Dropout layer in the LSTM network to prevent overfitting; The number of hidden layer neurons, batch size, and learning rate are optimized through grid search to minimize the RMSE and MAE indicators.
8. A photovoltaic power generation prediction system, characterized in that, include: An acquisition module is used to acquire meteorological data and power generation data, and pre-process the meteorological data and power generation data; An analysis module is used to construct a probability density function of power generation based on the processed meteorological data and power generation data, and to analyze the probability density function to obtain the frequency domain characteristics of power; A building module is used to construct a photovoltaic power prediction model based on LSTM according to the frequency domain characteristics of power.
9. A photovoltaic power generation power prediction device, characterized in that, The method comprises a processor and a memory, wherein the processor implements the photovoltaic power generation power prediction method according to any one of claims 1 to 7 when executing the computer program stored in the memory.
10. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein when the computer program is executed by a processor, the photovoltaic power generation power prediction method according to any one of claims 1 to 7 is implemented.