Method for generating photovoltaic output scene in extreme weather based on deep learning

Abstract

The invention discloses a deep learning-based photovoltaic output scene generation method in extreme weather, which comprises the following steps: collecting long-time series meteorological data, inputting a meteorological data preprocessing model, and generating preprocessed meteorological data; inputting the preprocessed meteorological data into an extreme weather event identification model, extracting meteorological features, change trends and lagging features when the extreme weather occurs, and identifying and marking an extreme weather event time period; establishing a mixed scene generation model according to the meteorological characteristics, the change trend and the lagging characteristics of the extreme weather event time period, and generating an extreme weather scene; carrying out dimension reduction visualization and quality evaluation comparison on the generated extreme weather scene and the real extreme weather scene through a multi-stage dimension reduction visualization model, and calculating a correction factor according to a quality evaluation comparison result; and establishing a photovoltaic output scene generation model according to the correction factor, and simulating a photovoltaic output condition in an extreme weather scene. According to the invention, the diversity and verisimilitude of extreme weather scene generation can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical fields of meteorological data processing and artificial intelligence, and particularly relates to a method for generating photovoltaic output scenarios under extreme weather based on deep learning. Background Technique

[0002] With the intensification of climate change, the frequency and intensity of extreme weather events are increasing continuously, and their impact on society and economy is becoming increasingly significant. Accurately simulating and generating extreme weather scenarios is of great significance for fields such as disaster prevention, emergency management, and climate research. Weather changes, such as light intensity, temperature, humidity, etc., directly affect the power generation efficiency of photovoltaic power stations. Especially in extreme weather conditions, the volatility of photovoltaic output will be further exacerbated. In order to accurately evaluate the photovoltaic output under extreme weather, an efficient and accurate method for generating photovoltaic output scenarios is needed, considering weather factors, to provide a scientific basis for the planning, design, and operation of the power grid.

[0003] In the existing technical system, the generation of photovoltaic output scenarios usually relies on numerical weather prediction and physical models. Although traditional physical model methods can accurately model according to the physical characteristics and performance parameters of solar panels, their generalization ability is limited when dealing with complex and changeable weather conditions. Especially in extreme weather conditions, the simulation accuracy of physical models will decrease significantly. Moreover, traditional statistical model methods often have difficulty effectively capturing the dynamic characteristics in time series data. In addition, traditional methods are difficult to generate diverse extreme weather scenarios and cannot meet the needs of different application scenarios. Summary of the Invention

[0004] Aiming at the above deficiencies in the prior art, the method for generating photovoltaic output scenarios under extreme weather based on deep learning provided by the present invention can improve the diversity and authenticity of extreme weather scenario generation, and solves the problem that traditional methods are difficult to generate diverse extreme weather scenarios.

[0005] In order to achieve the above invention purpose, the technical solution adopted by the present invention is: A method for generating photovoltaic output scenarios under extreme weather based on deep learning, including the following steps: S1. Collect long-time series meteorological data and input it into a meteorological data preprocessing model to generate preprocessed meteorological data; S2. Input the preprocessed meteorological data into an extreme weather event recognition model to extract meteorological features, change trends, and lag features when extreme weather occurs, and identify and mark extreme weather event periods; S3. Establish a hybrid scenario generation model according to the meteorological features, change trends, and lag features of extreme weather event periods to generate extreme weather scenarios; S4. Perform dimensionality reduction visualization and quality assessment comparison on the generated extreme weather scenarios and real extreme weather scenarios through a multi-level dimensionality reduction visualization model, and calculate the correction factor according to the quality assessment comparison results; S5. Establish a photovoltaic power output scenario generation model based on the correction factor to simulate the photovoltaic power output situation under extreme weather scenarios.

[0006] Furthermore: In S1, the long-term sequence meteorological data includes: temperature, humidity, wind speed, solar radiation intensity, precipitation, snow depth, and air pressure; The method for generating the preprocessed meteorological data is specifically as follows: Use linear interpolation or the mean value of historical data to fill the missing data in the long-term sequence meteorological data; Identify and classify the outliers in the long-term sequence meteorological data through the Z-score method, including deleting and interpolating the obviously wrong values, and retaining and marking the real values; Extract the key features related to photovoltaic power output and perform normalization processing to adapt to the extreme weather event recognition model. The key features related to photovoltaic power output include solar radiation intensity, temperature change rate, and the correlation between humidity and solar radiation.

[0007] Furthermore: In S2, extract the meteorological features when extreme weather occurs The expression is specifically as follows:

[0008] In the formula, is the temperature when extreme weather occurs, is the dew point temperature when extreme weather occurs, is the precipitation when extreme weather occurs, is the air pressure when extreme weather occurs, is the snow depth when extreme weather occurs, is the solar radiation when extreme weather occurs, is the visibility when extreme weather occurs, is the wind speed when extreme weather occurs, is the relative humidity when extreme weather occurs, is the absolute humidity when extreme weather occurs, is the cloud amount when extreme weather occurs, is the cloud height when extreme weather occurs, is the water vapor pressure when extreme weather occurs, is the precipitation intensity when extreme weather occurs, is the evaporation amount feature when extreme weather occurs; Extract the change trend The expression is specifically as follows:

[0009] In the formula, is the temperature change before or after the occurrence of extreme weather, is the dew point temperature change before or after the occurrence of extreme weather, is the precipitation change before or after the occurrence of extreme weather, is the air pressure change before or after the occurrence of extreme weather, is the snow depth change before or after the occurrence of extreme weather, is the solar radiation change before or after the occurrence of extreme weather, is the visibility change before or after the occurrence of extreme weather, is the wind speed change before or after the occurrence of extreme weather, is the relative humidity change before or after the occurrence of extreme weather, is the absolute humidity change before or after the occurrence of extreme weather, is the cloud amount change before or after the occurrence of extreme weather, is the cloud height change before or after the occurrence of extreme weather, is the water vapor pressure change before or after the occurrence of extreme weather, is the precipitation intensity change before or after the occurrence of extreme weather, is the evaporation change before or after the occurrence of extreme weather; Extract lag features The specific expression is:

[0010] In the formula, is The photovoltaic output scenario at time is the time when extreme weather occurs, is the photovoltaic output scenario within several hours after the occurrence of extreme weather, is the lag time window.

[0011] Furthermore: In S3, the hybrid scenario generation model includes a generator and a discriminator. The method for generating an extreme weather scenario is specifically: jointly simulate and generate an extreme weather scenario through the generator and the discriminator; Among them, the loss function of the generator and the loss function of the discriminator The specific expressions are:

[0012] In the formula, represents the distribution Random noise vector sampled Perform expected value calculation, and respectively represent the expected value calculation for sampling from the distribution sampled from and the expected value calculation for sampling from the distribution sampled from Perform expected value calculation, is the random noise vector, 、 、 are the distributions of random noise, real data, and generated data respectively, is the extreme weather label, and each item is: high temperature, cold wave, heavy rain, heavy snow, strong wind, drought, thunderstorm, hail, fog, extreme precipitation, and sandstorm, 、 are the discriminator and the generator respectively, 、 are the weights of the mean square error and the gradient penalty term, is the set of real extreme weather scenario features, and each item is: meteorological features such as air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud amount, cloud height, water vapor pressure, precipitation intensity, and evaporation; is the set of generated extreme weather scenario features, and each item is: meteorological features such as air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud amount, cloud height, water vapor pressure, precipitation intensity, and evaporation; is the mean square error between the set of generated extreme weather scenario features and the set of real extreme weather scenario features, represents the gradient penalty term, and are the probabilities of the real extreme weather scenario and the generated extreme weather scenario respectively, is a smoothing parameter less than 1, used to control the confidence of the discriminator in the generated scenario.

[0013] Furthermore: In S4, the expression for dimensionality reduction visualization of the generated extreme weather scenario and the real extreme weather scenario is specifically:

[0014] where, is the scenario after dimensionality reduction, is the principal component analysis, is the t-distributed stochastic neighbor embedding, is the uniform manifold approximation and projection, and The set of real and generated extreme weather scenario features, respectively, includes meteorological features such as air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud amount, cloud height, water vapor pressure, precipitation intensity, and evaporation.

[0015] Furthermore: In S4, the method for quality assessment and comparison between the generated extreme weather scenario and the real extreme weather scenario is specifically as follows: Calculate the evaluation index to obtain the comparison result between the generated extreme weather scenario and the real extreme weather scenario, calculate the correction factor based on the comparison result, and the correction factor is dynamically adjusted according to the extreme weather type and intensity to ensure that the model can accurately simulate the impact of extreme weather on photovoltaic power output; Among them, the correction factors include direct radiation of extreme weather, solar azimuth angle, diffuse radiation, ground reflected radiation, and photovoltaic power correction factor.

[0016] Furthermore: In S4, the evaluation indexes for quality assessment and comparison between the generated extreme weather scenario and the real extreme weather scenario include:

[0017]

[0018]

[0019]

[0020]

[0021]

[0022] In the formula, 、 、 and are the comprehensive evaluation index, Wasserstein distance, mean square error, and mutual information between the generated extreme weather scenario feature set and the real extreme weather scenario feature set, respectively; is the constraint condition of the Lipschitz continuous function ; is the expected value of the real extreme weather scenario feature set, is the expected value of the generated extreme weather scenario feature set, is the total number of scenarios, and are the th real extreme weather scenario feature set and the generated extreme weather scenario feature set, respectively, is the joint probability distribution, and The marginal probability distributions of the real extreme weather scenario feature set and the generated extreme weather scenario feature set, respectively, 、 、 are weight coefficients, which are dynamically adjusted according to the extreme weather type to reflect the focus of attention under different conditions.

[0023] Furthermore: In S5, the method for simulating the PV output under extreme weather scenarios is specifically as follows: S51. Divide the solar elevation angle into several intervals, each interval being 15 degrees. Calculate the radiation value for each 15-degree solar elevation angle through the direct radiation under extreme weather, solar azimuth angle, diffuse radiation, and ground reflection radiation correction factors, and then calculate the total solar radiation received ;

[0024]

[0025]

[0026]

[0027] Among them, is the solar elevation angle corresponding to the th interval. Divide 180° into 12 intervals, 、 、 and are the direct radiation, diffuse radiation, ground reflection radiation, and radiation attenuation caused by extreme weather received in the th interval, respectively. The radiation attenuation is related to the extreme weather type and intensity, 、 and are the direct radiation, diffuse radiation, and total radiation received on the ground in the th interval, respectively, is the tilt angle of the PV panel, 、 、 and are the direct radiation under extreme weather, solar azimuth angle, diffuse radiation, and ground reflection radiation correction factors, respectively, and their values are determined according to the extreme weather type and intensity, is the albedo, is the serial number of the day, is the latitude of the PV power station, is the solar declination; S52. Based on the total solar radiation received , the influence of extreme weather on photovoltaic power is considered through a photovoltaic power correction factor, and the photovoltaic output under extreme weather scenarios is calculated ;

[0028] In the formula, is the conversion efficiency of the photovoltaic system, is the area of the photovoltaic panel, is the total solar radiation received by the photovoltaic system, is the extreme weather photovoltaic power correction factor, and its value is determined according to the type and intensity of extreme weather.

[0029] The beneficial effects of the present invention are as follows: (1) The present invention provides a method for generating photovoltaic output scenarios under extreme weather based on deep learning. Based on a hybrid scenario generation model architecture of long short-term memory network, generative adversarial network, Transformer architecture and attention mechanism, the generation accuracy of photovoltaic output scenarios under extreme weather is significantly improved. It not only effectively solves the problem of scarce extreme weather scenarios, improves the generalization ability of the model through data augmentation technology, but also comprehensively considers the direct impact and lag effect of extreme weather on photovoltaic output, making the generation results closer to reality and solving the problem that traditional methods are difficult to generate diverse extreme weather scenarios.

[0030] (2) The present invention provides a more reliable basis for power grid planning, operation optimization and risk management, helps to optimize power dispatching, reduce the operation cost of the power grid, and enhance the stability and resilience of the power grid under extreme weather conditions, promoting the healthy development of the photovoltaic industry. Description of the Drawings

[0031] Figure 1 is a flowchart of a method for generating photovoltaic output scenarios under extreme weather based on deep learning according to the present invention. Specific Embodiments

[0032] The following describes the specific embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0033] As Figure 1 shown, in an embodiment of the present invention, a method for generating photovoltaic output scenarios under extreme weather based on deep learning includes the following steps: S1. Collect long-term time series meteorological data and input it into a meteorological data preprocessing model to generate preprocessed meteorological data; S2. Input the preprocessed meteorological data into the extreme weather event recognition model, extract the meteorological features, change trends, and lag features during extreme weather occurrences, and identify and mark the time periods of extreme weather events; S3. Establish a hybrid scenario generation model based on the meteorological features, change trends, and lag features of the extreme weather event time periods to generate extreme weather scenarios; S4. Perform dimensionality reduction visualization and quality assessment comparison on the generated extreme weather scenarios and real extreme weather scenarios through a multi-level dimensionality reduction visualization model, and calculate the correction factor according to the quality assessment comparison results; S5. Establish a photovoltaic output scenario generation model based on the correction factor to simulate the photovoltaic output situation under extreme weather scenarios.

[0034] In S1, the long-term time series meteorological data includes: temperature, humidity, wind speed, solar radiation intensity, precipitation, snow depth, and air pressure; In this embodiment, the present invention extracts long-term time series meteorological data in real time to continuously capture the impact of day-night and seasonal changes on photovoltaic output for 24 hours a day.

[0035] The data is collected from multiple channels, including meteorological stations, satellite remote sensing data, on-site monitoring equipment of photovoltaic power plants, and historical meteorological archives. These data cover hourly meteorological records in the past 10 years, ensuring the comprehensiveness and diversity of the data. The collected meteorological data includes, but is not limited to, temperature, humidity, wind speed, solar radiation intensity, precipitation, snow depth, air pressure, etc. Particular attention is paid to extreme weather events, such as high temperature (above 35°C), low temperature (below -10°C), heavy rain (daily precipitation exceeding 50 mm), snowstorm (wind speed exceeding 20 m / s and accompanied by snowfall), etc., and the time series integrity of the data is ensured, covering 365 days a year with continuous recording for 24 hours a day to capture the impact of day-night changes and seasonal patterns on photovoltaic output.

[0036] The method for generating the preprocessed meteorological data is specifically as follows: Use linear interpolation or the mean value of historical data to fill in the missing data in the long-term time series meteorological data; for example, if the solar radiation data for a certain hour is missing, the data for the two adjacent hours will be used for linear interpolation.

[0037] Identify and classify outliers in long - term meteorological data through the Z - score method, including deleting and interpolating obviously incorrect values, and retaining and marking real values; for example, if the maximum temperature on a certain day exceeds 3 standard deviations of the historical maximum temperature, it is regarded as an outlier. For outliers, the following methods are used according to specific situations: if the outlier is an obvious error (such as an abnormally low temperature record of - 50°C), the data point is directly removed and filled by interpolation. If the outlier is an extreme but possibly real situation (such as a rare high temperature or heavy rain), the data point is retained and specially marked in subsequent analysis.

[0038] Extract key features related to photovoltaic power output and perform normalization to adapt to the extreme weather event recognition model. The key features related to photovoltaic power output include solar radiation intensity, temperature change rate, and the correlation between humidity and solar radiation.

[0039] Through the above processing, the data missing rate after pre - processing is reduced, the outlier ratio is reduced, and the integrity and accuracy of the data are significantly improved. The normalized data can better adapt to the input requirements of deep learning models, improving the training efficiency of the models. The generalization ability of the models under extreme weather conditions is significantly enhanced, and they can more accurately generate photovoltaic power output scenarios under different regions and different types of extreme weather.

[0040] In S2, extract meteorological features when extreme weather occurs The specific expression is:

[0041] In the formula, is the temperature when extreme weather occurs, is the dew point temperature when extreme weather occurs, is the precipitation when extreme weather occurs, is the air pressure when extreme weather occurs, is the snow depth when extreme weather occurs, is the solar radiation when extreme weather occurs, is the visibility when extreme weather occurs, is the wind speed when extreme weather occurs, is the relative humidity when extreme weather occurs, is the absolute humidity when extreme weather occurs, is the cloud amount when extreme weather occurs, is the cloud height when extreme weather occurs, is the vapor pressure when extreme weather occurs, is the precipitation intensity when extreme weather occurs, is the evaporation amount feature when extreme weather occurs; Extract the change trend The specific expression is:

[0042] In the formula, is the temperature change before or after the occurrence of extreme weather, is the dew point temperature change before or after the occurrence of extreme weather, is the precipitation change before or after the occurrence of extreme weather, is the air pressure change before or after the occurrence of extreme weather, is the snow depth change before or after the occurrence of extreme weather, is the solar radiation change before or after the occurrence of extreme weather, is the visibility change before or after the occurrence of extreme weather, is the wind speed change before or after the occurrence of extreme weather, is the relative humidity change before or after the occurrence of extreme weather, is the absolute humidity change before or after the occurrence of extreme weather, is the cloud amount change before or after the occurrence of extreme weather, is the cloud height change before or after the occurrence of extreme weather, is the vapor pressure change before or after the occurrence of extreme weather, is the precipitation intensity change before or after the occurrence of extreme weather, is the evaporation change before or after the occurrence of extreme weather; Extract lag features The specific expression is:

[0043] In the formula, is the photovoltaic output scenario at time is the moment when extreme weather occurs, is the photovoltaic output scenario within several hours after the occurrence of extreme weather, is the lag time window.

[0044] In S3, the hybrid scenario generation model includes a generator and a discriminator. The method for generating an extreme weather scenario is specifically as follows: jointly simulate and generate an extreme weather scenario through the generator and the discriminator; Among them, the loss function of the generator and the loss function of the discriminator are specifically expressed as:

[0045]

[0046] In the formula, represents the calculation of the expected value for the random noise vector sampled from the distribution , and respectively represent the calculation of the expected value for sampled from the distribution and for sampled from the distribution , is the random noise vector, , , are respectively the distributions of the random noise, real data, and generated data, is the extreme weather label, and each item is: high temperature, cold snap, heavy rain, heavy snow, strong wind, drought, thunderstorm, hail, fog, extreme precipitation, and sandstorm, , are respectively the discriminator and the generator, , are the weights of the mean square error and the gradient penalty term, is the set of real extreme weather scenario features, and each item is: meteorological features such as air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud amount, cloud height, water vapor pressure, precipitation intensity, and evaporation; is the set of generated extreme weather scenario features, and each item is: meteorological features such as air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud amount, cloud height, water vapor pressure, precipitation intensity, and evaporation; is the mean square error between the set of generated extreme weather scenario features and the set of real extreme weather scenario features, represents the gradient penalty term, and are respectively the probabilities of the real extreme weather scenario and the generated extreme weather scenario, is a smoothing parameter less than 1, which is used to control the confidence of the discriminator in the generated scenario.

[0047] In this embodiment, the hybrid scenario generation model (LGTA) is a model architecture based on the long short-term memory network (LSTM), generative adversarial network (GAN), Transformer architecture, and attention mechanism. By alternately training the generator and discriminator of the generative adversarial network (GAN), the ability of the discriminator to distinguish real and false data is trained, and the ability of the generator to generate realistic data is trained. The model is continuously optimized to generate more realistic extreme weather scenarios.

[0048] The model parameters are set as follows: the dimension of the random noise vector z is Lz, and the dimension of the extreme weather label y is Ly. The dimension of the LSTM hidden layer is H, and the output dimension is consistent with the real scenario, including hourly meteorological features. Both the generator and the discriminator adopt an L-layer LSTM structure, the number of Transformer layers is T, and the number of heads of the multi-head attention mechanism is A. The Adam optimizer is used, and the learning rate decays to Dr times the original value every 100 iterations.

[0049] In S4, the expression for dimensionality reduction visualization of the generated extreme weather scenario and the real extreme weather scenario is specifically:

[0050] Among them, is the scenario after dimensionality reduction, is the principal component analysis, is the t-distributed stochastic neighbor embedding, is the uniform manifold approximation and projection, and are the feature sets of the real and generated extreme weather scenarios respectively, which include meteorological features such as air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud amount, cloud height, water vapor pressure, precipitation intensity, and evaporation.

[0051] In this embodiment, through dimensionality reduction visualization, the high-dimensional extreme weather scenario feature set can be mapped to a low-dimensional space, and the similarity and difference between the generated extreme weather scenario and the real extreme weather scenario can be intuitively observed for subsequent quality assessment and comparison.

[0052] In S4, the method for quality assessment and comparison of the generated extreme weather scenario and the real extreme weather scenario is specifically: Calculate the evaluation index to obtain the comparison result between the generated extreme weather scenario and the real extreme weather scenario, calculate the correction factor according to the comparison result, and the correction factor is dynamically adjusted according to the extreme weather type and intensity to ensure that the model can accurately simulate the impact of extreme weather on photovoltaic power output; Among them, the correction factors include direct radiation of extreme weather, solar azimuth angle, diffuse radiation, ground reflected radiation, and photovoltaic power correction factor.

[0053] In this embodiment, the value of the correction factor will be dynamically adjusted according to the extreme weather type (including high temperature, cold wave, heavy rain, heavy snow, strong wind, drought, thunderstorm, hail, fog, extreme precipitation, and sandstorm) and intensity to ensure that the model can accurately simulate the impact of extreme weather on photovoltaic power output.

[0054] In S4, the evaluation indexes for quality assessment and comparison of the generated extreme weather scenario and the real extreme weather scenario include:

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] In the formula, , , and are respectively the comprehensive evaluation index, Wasserstein distance, mean square error and mutual information between the generated extreme weather scenario feature set and the real extreme weather scenario feature set; is the constraint condition of the Lipschitz continuous function , is the expected value of the real extreme weather scenario feature set, is the expected value of the generated extreme weather scenario feature set, is the total number of scenarios, and are respectively the th real extreme weather scenario feature set and the generated extreme weather scenario feature set, is the joint probability distribution, and are respectively the marginal probability distributions of the real extreme weather scenario feature set and the generated extreme weather scenario feature set, , , are weight coefficients, and the weight coefficients are dynamically adjusted according to the extreme weather type to reflect the focus of attention under different conditions.

[0061] In S5, the method for simulating the PV output under extreme weather scenarios is specifically as follows: S51. Divide the solar elevation angle into several intervals, each interval being 15 degrees. Calculate the radiation value for every 15-degree solar elevation angle through the direct radiation under extreme weather, solar azimuth angle, diffuse radiation, and ground reflection radiation correction factors, and then calculate the total solar radiation ;

[0062]

[0063]

[0064]

[0065] Among them, is the solar elevation angle corresponding to the th interval. The 180° is divided into 12 intervals. , , and are the direct radiation, diffuse radiation, ground-reflected radiation and radiation attenuation caused by extreme weather received in the th interval respectively. The radiation attenuation is related to the type and intensity of extreme weather. , and are the direct radiation, diffuse radiation and total radiation received on the ground in the th interval respectively. is the tilt angle of the photovoltaic panel. , , and are the direct radiation of extreme weather, solar azimuth angle, diffuse radiation and ground-reflected radiation correction factors respectively, and their values are determined according to the type and intensity of extreme weather. is the ground albedo. is the serial number of the day. is the latitude of the photovoltaic power station. is the solar declination; In this embodiment, the total solar radiation received by the photovoltaic system is composed of direct solar radiation, diffuse solar radiation and ground-reflected radiation. At the same time, considering the shielding and attenuation effects of extreme weather on solar radiation, a correction factor is introduced. In order to make the result more accurate, the solar elevation angle is divided into several intervals, each interval is 15 degrees, and the radiation value of every 15-degree solar elevation angle is calculated.

[0066] S52. Based on the total solar radiation received, consider the influence of extreme weather on the photovoltaic power through the photovoltaic power correction factor, and calculate the photovoltaic output under extreme weather scenarios;

[0067] In the formula, is the conversion efficiency of the photovoltaic system. is the area of the photovoltaic panel. is the total solar radiation received by the photovoltaic system. is the photovoltaic power correction factor for extreme weather, and its value is determined according to the type and intensity of extreme weather.

[0068] In this embodiment, the photovoltaic power under extreme weather conditions is affected by the increase in the temperature of the photovoltaic panel (high temperature) and shading (heavy rain, snowstorm). The influence of extreme weather on the photovoltaic power is considered by introducing a photovoltaic power correction factor.

[0069] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the number of technical features. Therefore, the features defined by "first", "second", "third" may explicitly or implicitly include one or more of such features.

Claims

1. A method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning, characterized in that: The following steps are involved: S1, collect long-term series meteorological data, input it into the meteorological data preprocessing model, and generate preprocessed meteorological data; S2. Input the preprocessed meteorological data into the extreme weather event recognition model, extract the meteorological characteristics, change trends and hysteresis characteristics when extreme weather occurs, and identify and mark the extreme weather event period; S3. Establish a hybrid scenario generation model based on the meteorological characteristics, change trends and hysteresis characteristics during the extreme weather event period to generate extreme weather scenarios; S4. Perform dimensionality reduction visualization and quality assessment comparison on the generated extreme weather scenario and the real extreme weather scenario through a multi-level dimensionality reduction visualization model, and calculate the correction factor based on the quality assessment comparison results; S5. A photovoltaic output scenario generation model is established based on the correction factor to simulate the photovoltaic output under extreme weather scenarios.

2. The method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning according to claim 1, characterized in that: In S1, long-term series meteorological data include: temperature, humidity, wind speed, solar radiation intensity, precipitation, snow depth and air pressure; The method for generating preprocessed meteorological data is as follows: Use linear interpolation or historical data mean to fill in missing data in long-term series meteorological data; Identify and classify outliers in long-term meteorological data series using the Z-score method, including deleting and interpolating obviously erroneous values, and retaining and marking true values; The key features related to photovoltaic output are extracted and normalized to adapt to the extreme weather event recognition model. The key features related to photovoltaic output include solar radiation intensity, temperature change rate, and the correlation between humidity and solar radiation.

3. The method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning according to claim 1, characterized in that: In S2, the meteorological characteristics when extreme weather occurs are extracted The specific expression is: In the formula, is the temperature when extreme weather occurs. is the dew point temperature when extreme weather occurs. is the amount of precipitation when extreme weather occurs. is the air pressure when extreme weather occurs. is the snow depth when extreme weather occurs. is the solar radiation when extreme weather occurs. For visibility during extreme weather events, is the wind speed when extreme weather occurs. is the relative humidity when extreme weather occurs, is the absolute humidity when extreme weather occurs, is the cloud cover when extreme weather occurs. is the cloud height when extreme weather occurs. is the water vapor pressure when extreme weather occurs, is the precipitation intensity when extreme weather occurs. It is the evaporation characteristics when extreme weather occurs; Extracting changing trends The specific expression is: In the formula, is the temperature change before or after extreme weather occurs. is the change in dew point temperature before or after extreme weather occurs. is the change in precipitation before or after extreme weather occurs. is the change in air pressure before or after extreme weather occurs. is the change in snow depth before or after extreme weather occurs. is the change in solar radiation before or after extreme weather occurs. is the visibility change before or after extreme weather occurs. is the change in wind speed before or after extreme weather occurs. is the relative humidity change before or after extreme weather occurs. is the absolute humidity change before or after extreme weather occurs. is the change in cloud cover before or after extreme weather occurs. is the cloud height change before or after extreme weather occurs. is the change in water vapor pressure before or after extreme weather occurs. is the change in precipitation intensity before or after extreme weather occurs. It is the change in evaporation before or after the occurrence of extreme weather; Extracting hysteresis features The specific expression is: In the formula, for Photovoltaic output scene at all times, When extreme weather occurs, This is the photovoltaic output scenario within a few hours after extreme weather occurs. is the lag time window.

4. The method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning according to claim 1, characterized in that: In S3, the hybrid scenario generation model includes a generator and a discriminator, and the method for generating extreme weather scenarios is specifically: using the generator and the discriminator to jointly simulate and generate extreme weather scenarios; Among them, the loss function of the generator And the loss function of the discriminator The specific expression is: In the formula, Represents the distribution A random noise vector sampled from Calculate the expected value. and Respectively represent the distribution Sampled in and the distribution from Sampled in Calculate the expected value. is a random noise vector, , , are the distributions of random noise, real data, and generated data, respectively, The labels for extreme weather include high temperature, cold wave, heavy rain, heavy snow, strong wind, drought, thunderstorm, hail, heavy fog, extreme precipitation and sandstorm. , are the discriminator and the generator respectively, , is the weight of the mean square error and gradient penalty term, It is a set of real extreme weather scene features, including: air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud cover, cloud height, water vapor pressure, precipitation intensity and evaporation meteorological characteristics; The generated extreme weather scene feature set includes: air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud cover, cloud height, water vapor pressure, precipitation intensity and evaporation meteorological characteristics; is the mean square error between the generated extreme weather scene feature set and the real extreme weather scene feature set, represents the gradient penalty term, and are the probabilities of real extreme weather scenarios and generated extreme weather scenarios, respectively. is a smoothing parameter less than 1, which is used to control the confidence of the discriminator in the generated scene.

5. The method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning according to claim 4 is characterized in that: In S4, the expression for dimensionality reduction visualization of the generated extreme weather scenarios and the real extreme weather scenarios is as follows: in, For the scene after dimensionality reduction, For principal component analysis, is a t-distributed random neighborhood embedding, is the uniform manifold approximation and projection, and They are the feature sets of real and generated extreme weather scenarios, respectively, including the meteorological characteristics of air temperature, dew point temperature, precipitation, air pressure, snow depth, solar radiation, visibility, wind speed, relative humidity, absolute humidity, cloud cover, cloud height, water vapor pressure, precipitation intensity and evaporation.

6. The method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning according to claim 5, characterized in that: In S4, the method for comparing the quality of the generated extreme weather scene and the real extreme weather scene is as follows: Calculate the evaluation index to obtain the comparison results of the generated extreme weather scenario and the real extreme weather scenario, and calculate the correction factor based on the comparison results. The correction factor is dynamically adjusted according to the type and intensity of extreme weather to ensure that the model can accurately simulate the impact of extreme weather on photovoltaic output; Among them, the correction factors include direct radiation in extreme weather, solar azimuth, diffuse radiation, ground reflected radiation and photovoltaic power correction factors.

7. The method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning according to claim 6, characterized in that: In S4, the evaluation indicators for quality evaluation and comparison of the generated extreme weather scenarios and the real extreme weather scenarios include: In the formula, , , and They are the comprehensive evaluation index, Wasserstein distance, mean square error and mutual information between the generated extreme weather scene feature set and the real extreme weather scene feature set; is a Lipschitz continuous function The constraints of is the expected value of the real extreme weather scene feature set, is the expected value of the generated extreme weather scene feature set, is the total number of scenes, and Respectively A real extreme weather scene feature set and a generated extreme weather scene feature set, is the joint probability distribution, and are the marginal probability distributions of the real extreme weather scene feature set and the generated extreme weather scene feature set, respectively. , , is the weight coefficient, which is dynamically adjusted according to the extreme weather type to reflect the focus under different conditions.

8. The method for generating photovoltaic output scenarios under extreme weather conditions based on deep learning according to claim 7, characterized in that: In S5, the method for simulating the photovoltaic output under extreme weather scenarios is as follows: S51, the sun elevation angle It is divided into several intervals, each interval is 15 degrees. The radiation value of each 15-degree solar elevation angle is calculated through extreme weather direct radiation, solar azimuth, diffuse radiation and ground reflected radiation correction factors, and then the total solar radiation received is calculated. ; in, For the The solar elevation angle corresponding to each interval divides 180° into 12 intervals. , , and They are The radiation attenuation caused by direct radiation, diffuse radiation, ground reflected radiation and extreme weather received in each interval is related to the type and intensity of extreme weather. , and They are The direct radiation, diffuse radiation and total radiation received on the ground in each interval, is the tilt angle of the photovoltaic panel, , , and They are the correction factors for extreme weather direct radiation, solar azimuth, diffuse radiation and ground reflected radiation, and their values ​​are determined according to the type and intensity of extreme weather. is the ground albedo, It is the serial number of the day. is the latitude of the photovoltaic power station, is the solar declination; S52, based on the total solar radiation received , the photovoltaic power correction factor is used to consider the impact of extreme weather on photovoltaic power and calculate the photovoltaic output under extreme weather scenarios ; In the formula, is the conversion efficiency of the photovoltaic system, is the area of ​​the photovoltaic panel, is the total solar radiation received by the PV system, It is the extreme weather photovoltaic power correction factor, and its value is determined according to the type and intensity of extreme weather.

Images