Method, device and equipment for weather type identification of distributed photovoltaic prediction uncertainty and medium

Abstract

The application discloses a weather type identification method and device for distributed photovoltaic prediction uncertainty, equipment and medium, relates to the photovoltaic power generation technical field, carries out training to bayesian neural network based on photovoltaic output sample data set, obtains the probability distribution parameter of photovoltaic output;Random sampling is carried out on the probability density function, a noise sequence is generated, the noise sequence and the historical photovoltaic output sequence are input into the generative adversarial network, and the photovoltaic output sequence is output;The photovoltaic output sequence is divided into local class Gaussian distribution subset by using regression decision tree, a Gaussian mixture model is constructed for the local class Gaussian distribution subset, and the Gaussian mixture model is iteratively updated;The probability density and the abnormal score of the photovoltaic output sequence corresponding to the weather to be identified are calculated by using the updated Gaussian mixture model and the isolated forest, the weather type of the weather to be identified is determined based on the probability density and the abnormal score, the identification precision and reliability are enhanced, and the large-scale distributed photovoltaic output is effectively predicted and controlled.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and in particular to a method, apparatus, equipment and medium for identifying weather type uncertainty in distributed photovoltaic forecasting. Background Technology

[0002] With the continuous grid connection of distributed photovoltaic (PV) systems, their penetration rate is constantly increasing, and their application in the power industry is becoming increasingly widespread. However, after large-scale distributed PV systems are connected to the grid, their output is affected by changes in weather factors such as irradiance and relative humidity due to varying weather conditions, exhibiting significant volatility and randomness. This poses a significant challenge to the safety and reliability of power system planning and operation. Considering the impact of extreme weather on large-scale distributed PV operation, the prediction of PV output is subject to considerable uncertainty.

[0003] Existing power prediction methods mainly include mathematical statistical prediction and artificial intelligence prediction. Mathematical statistical prediction methods are mostly based on modeling the statistical patterns of historical data, but they often fail to capture changes in weather types. Artificial intelligence prediction methods rely on techniques such as neural networks and data mining to fit complex nonlinear relationships, learn their potential distribution patterns, and use them for prediction. However, in practical engineering, the predictions of these methods often depend on a large amount of high-quality data. Due to the suddenness of various extreme weather events, historical data on photovoltaic power plants under such weather conditions is scarce, affecting subsequent analysis and making it difficult to effectively predict the trend of photovoltaic power output changes under extreme weather scenarios.

[0004] As can be seen from the above, how to overcome the problem of insufficient sample size under extreme weather conditions, accurately identify weather types, enhance the accuracy and reliability of identification, and effectively predict and regulate the output of large-scale distributed photovoltaic power is a problem that needs to be solved in this field. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for weather type identification in the face of uncertainty in distributed photovoltaic forecasting. This method overcomes the problem of insufficient sample size under extreme weather conditions, accurately identifies weather types, enhances the accuracy and reliability of identification, and effectively predicts and controls the output of large-scale distributed photovoltaic power. The specific solution is as follows: In a first aspect, this application discloses a method for weather type identification of forecast uncertainty in distributed photovoltaic systems, including: A photovoltaic power output sample dataset is constructed using historical operating data of the target area. A Bayesian neural network is trained based on the photovoltaic power output sample dataset to obtain the probability distribution parameters of photovoltaic power output. A probability density function is constructed using the probability distribution parameters and historical photovoltaic power data. The probability density function is randomly sampled to generate a noise sequence. The noise sequence and the historical photovoltaic power output sequence are input into a generative adversarial network to output a photovoltaic power output sequence that varies with time. Based on the photovoltaic power output sequence, power characteristics and time series characteristics are determined. Using a regression decision tree, and based on the power characteristics and time series characteristics, the photovoltaic power output sequence is divided into local Gaussian distribution subsets. A Gaussian mixture model is constructed for the local Gaussian distribution subsets. The Gaussian mixture model is iteratively updated to obtain the updated Gaussian mixture model. An isolated forest is constructed based on the local Gaussian distribution subsets. Using the updated Gaussian mixture model and the isolated forest, the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified are calculated respectively. Based on the probability density and the anomaly score, the weather type of the weather to be identified is determined.

[0006] Optionally, constructing a photovoltaic output sample dataset using historical operating data of the target area includes: Meteorological element data are extracted from historical operational data of the target area at fixed time intervals; the meteorological element data includes photovoltaic power, wind speed, temperature, relative humidity, and irradiance. Using the meteorological data mentioned above, a photovoltaic power output sample dataset is constructed; Set the window size and step size of the first sliding window; Using the window size and step size of the first sliding window, the photovoltaic power output sample dataset is traversed to calculate the average power and peak-valley difference within each first window, and the average power and peak-valley difference within each first window are used as indicators for judging abnormal power. Construct indicators for judging abnormal weather conditions; Based on the abnormal power judgment index and the abnormal weather judgment index, an abnormal weather dataset is selected from the photovoltaic power output sample dataset. The datasets other than the abnormal weather dataset in the photovoltaic power output sample dataset are used as the normal weather dataset.

[0007] Optionally, training the Bayesian neural network based on the photovoltaic output sample dataset includes: Construct a Bayesian neural network; the Bayesian neural network includes an input layer, an output layer, and several hidden layers; Construct a loss function using KL divergence; Using a stochastic variational inference method, and with the goal of minimizing the loss function, normal weather datasets and abnormal weather datasets are input into a Bayesian neural network for training, so as to output the probability distribution parameters of photovoltaic power output.

[0008] Optionally, the step of inputting the noise sequence and the historical photovoltaic power output sequence into a generative adversarial network to output a photovoltaic power output sequence that varies with time includes: The noisy sequence is input into the generator of the generative adversarial network to output generated samples; Select real samples from the photovoltaic power output sample dataset; The generated sample and the real sample are input into the discriminator in the generative adversarial network to output the discrimination probability of the real sample and the discrimination probability of the generated sample; The generator and discriminator are trained using the discrimination probabilities of the real samples and the discrimination probabilities of the generated samples to obtain the trained generative adversarial network. The mapping relationship between noise sequences and historical photovoltaic power output sequences is analyzed using a trained generative adversarial network to obtain the photovoltaic power output sequence that varies with time.

[0009] Optionally, the step of determining power characteristics and time-series characteristics based on the photovoltaic output sequence, utilizing a regression decision tree, and dividing the photovoltaic output sequence into locally Gaussian-like subsets based on the power characteristics and time-series characteristics includes: Set the window size and step size of the second sliding window; Using the window size and step size of the second sliding window, the photovoltaic power output sequence is traversed to calculate the average power and peak-to-valley difference within each second window; Calculate the first-order difference of the photovoltaic power output sequence, and use the first-order difference, the average power within each second window, and the peak-valley difference as power characteristics and time series characteristics, respectively. The power characteristics and the timing characteristics are normalized to obtain a normalized feature set; Based on the normalized feature set and using a regression decision tree, the photovoltaic power output sequence is divided into a subset of local Gaussian distributions.

[0010] Optionally, the iterative update of the Gaussian mixture model to obtain the updated Gaussian mixture model, and the construction of an isolated forest based on the local Gaussian distribution subset, includes: The Gaussian mixture model is iteratively updated using the expectation-maximization algorithm, and it is determined whether the preset iteration termination condition has been triggered. If the preset iteration termination condition is triggered, the iteration update will stop, and the updated Gaussian mixture model will be obtained. Samples for constructing an isolated forest are selected from a subset of local Gaussian-like distributions. Target features are randomly selected from all features of the sample, and the cut value of the target features is determined. The sample is then segmented using the cut value to obtain the segmented sample. Determine whether the preset segmentation termination condition has been triggered. If the preset segmentation termination condition is triggered, the segmentation will stop, resulting in an isolated tree. The isolated trees are used to form an isolated forest.

[0011] Optionally, the step of using the updated Gaussian mixture model and the isolated forest to calculate the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified, and determining the weather type based on the probability density and the anomaly score, includes: The photovoltaic power output sequence corresponding to the weather to be identified is input into the updated Gaussian mixture model to output the probability density of the photovoltaic power output sequence corresponding to the weather to be identified. The photovoltaic power output sequence corresponding to the weather to be identified is input into the isolated forest to output the anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified. Set the probability density and the weighting coefficients for the anomaly scores; The probability density and the anomaly score are weighted using the weighting coefficients to obtain a comprehensive score; The weather type to be identified is determined based on the comprehensive score.

[0012] Secondly, this application discloses a weather type identification device for distributed photovoltaic forecasting uncertainty, comprising: The training module is used to construct a photovoltaic power output sample dataset using historical operating data of the target area, and to train a Bayesian neural network based on the photovoltaic power output sample dataset to obtain the probability distribution parameters of photovoltaic power output. The photovoltaic power output sequence determination module is used to construct a probability density function using the probability distribution parameters and historical photovoltaic power data, randomly sample the probability density function to generate a noise sequence, and input the noise sequence and historical photovoltaic power output sequence into a generative adversarial network to output a photovoltaic power output sequence that varies with time. An isolated forest construction module is used to determine power characteristics and time-series characteristics based on the photovoltaic power output sequence, utilize a regression decision tree, and divide the photovoltaic power output sequence into local Gaussian-like subsets based on the power characteristics and time-series characteristics, construct a Gaussian mixture model for the local Gaussian-like subsets, iteratively update the Gaussian mixture model to obtain the updated Gaussian mixture model, and construct an isolated forest based on the local Gaussian-like subsets. The weather type determination module is used to calculate the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified using the updated Gaussian mixture model and the isolated forest, and to determine the weather type of the weather to be identified based on the probability density and the anomaly score.

[0013] Thirdly, this application discloses an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned method for identifying weather types with uncertainty in distributed photovoltaic forecasting.

[0014] Fourthly, this application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed method for identifying weather type uncertainties in distributed photovoltaic forecasting.

[0015] As can be seen, this application provides a weather type identification method for distributed photovoltaic (PV) forecast uncertainty. This method includes constructing a PV output sample dataset using historical operational data of the target area, training a Bayesian neural network based on the dataset to obtain probability distribution parameters for PV output. This quantifies the uncertainty of PV output, improving the realism and reliability of subsequent scenario generation. A probability density function is constructed using the probability distribution parameters and historical PV power data. Random sampling is performed on the probability density function to generate a noise sequence. The noise sequence and historical PV output sequences are input into a generative adversarial network (GAN) to output a PV output sequence showing the change in PV power over time. Therefore, this method can generate PV output sequences that conform to the true probability distribution and time-series characteristics, effectively expanding the training samples for extreme weather scenarios and improving the diversity and representativeness of the training data. Based on the photovoltaic (PV) power output sequence, power and time-series characteristics are determined. A regression decision tree is used to divide the PV power output sequence into local Gaussian-like subsets. A Gaussian mixture model is constructed for these subsets and iteratively updated. An isolated forest is then built based on these subsets. This improves the fitting accuracy of the Gaussian mixture model to the local probability distribution of PV power output, enhances the ability to identify anomalous power output characteristics under extreme weather conditions, and improves the accuracy and robustness of PV power output characteristic modeling in complex weather scenarios. Using the updated Gaussian mixture model and the isolated forest, the probability density and anomaly score of the PV power output sequence corresponding to the weather to be identified are calculated. Based on the probability density and anomaly score, the weather type of the weather to be identified is determined, which simultaneously considers the degree of weather type matching and anomalous characteristics, improving the accuracy, stability, and reliability of weather type identification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This application discloses a flowchart for a weather type identification method for distributed photovoltaic forecasting uncertainty. Figure 2 A schematic diagram of the overall strategy framework for weather type identification of distributed photovoltaic forecasting uncertainty disclosed in this application; Figure 3 This is a network structure diagram of a generative adversarial network disclosed in this application; Figure 4 This is a data identification diagram disclosed in this application; Figure 5 This is a graph showing the regression analysis results of predicted power and actual power on a training set disclosed in this application; Figure 6 This is a graph showing the regression analysis results of predicted power and actual power on a test set disclosed in this application; Figure 7 This is a schematic diagram of the weather type identification device for distributed photovoltaic forecasting uncertainty disclosed in this application; Figure 8 This application provides a structural diagram of an electronic device. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] With the continuous grid connection of distributed photovoltaic (PV) systems, their penetration rate is constantly increasing, and their application in the power industry is becoming increasingly widespread. However, after large-scale distributed PV systems are connected to the grid, their output is affected by changes in weather factors such as irradiance and relative humidity due to different weather types, exhibiting significant fluctuations and randomness. This poses a huge challenge to the safety and reliability of power system planning and operation. Considering the impact of extreme weather on large-scale distributed PV operation, the prediction of PV output is subject to considerable uncertainty. Existing power prediction methods mainly include mathematical statistical prediction methods and artificial intelligence prediction methods. Mathematical statistical prediction methods are mostly based on historical data statistical patterns for modeling, but they often fail to capture changes in weather types. Artificial intelligence prediction methods rely on neural networks and data mining techniques to fit complex nonlinear relationships, learn their potential distribution patterns, and use them for prediction. However, in practical engineering, the prediction of these methods often depends on a large amount of high-quality data. Due to the suddenness of various extreme weather events, historical data on PV power plants under such weather conditions is scarce, affecting subsequent analysis and making it difficult to effectively predict the trend of PV output changes under extreme weather scenarios. As can be seen from the above, how to overcome the problem of insufficient sample size under extreme weather conditions, accurately identify weather types, enhance the accuracy and reliability of identification, and effectively predict and regulate the output of large-scale distributed photovoltaic power is a problem that needs to be solved in this field.

[0020] See Figure 1 As shown in the embodiments of this application, a weather type identification method for distributed photovoltaic forecasting uncertainty is disclosed, which may specifically include: Step S11: Construct a photovoltaic power output sample dataset using historical operating data of the target area, and train a Bayesian neural network based on the photovoltaic power output sample dataset to obtain the probability distribution parameters of photovoltaic power output.

[0021] In this embodiment, a schematic diagram of the overall strategy framework for weather type identification of distributed photovoltaic forecasting uncertainty is shown below. Figure 2 As shown, the first step is to construct the dataset and perform distribution learning. Specifically, meteorological element data is extracted from the historical operational data of the target area at fixed time intervals. These meteorological element data include photovoltaic power, wind speed, temperature, relative humidity, and irradiance. That is, based on the historical photovoltaic operational data of the target area, the total number of days and the number of sampling points per day are selected. Data points are collected at fixed time intervals, and the aforementioned meteorological element data are extracted. Then, using these meteorological element data, the various types of data are merged to construct a photovoltaic power output sample dataset.

[0022] In this application, high power values, low power values, and peak-to-valley differences from the power curve are used as extreme indicators for sample collection. A sliding window method is employed, with the window size set accordingly. N and step length SBy traversing the photovoltaic power output sequence, the average power output and peak-to-valley difference within each window are calculated. The high and low power values ​​are determined by calculating the average power of the power curve, while the peak-to-valley difference is defined as the difference between the maximum and minimum power in the power curve within the sliding window, using the following formula: ; ; in, For peak-valley difference, This represents the average power.

[0023] Furthermore, an abnormal weather judgment index is constructed, and based on the abnormal power judgment index and the abnormal weather judgment index, an abnormal weather dataset is selected from the photovoltaic power output sample dataset, and the other datasets in the photovoltaic power output sample dataset other than the abnormal weather dataset are used as normal weather datasets.

[0024] Specifically, the screening is based on abnormal power judgment indicators, and the judgment criteria are set as follows: the window average power is higher than the daily average. a Double or below the daily average a Furthermore, to avoid misjudgment due to short-term fluctuations, continuous [measurement / analysis] is required. m A window satisfies this condition; the peak-to-valley difference of the window is greater than the maximum peak-to-valley difference of the photovoltaic output sequence. b The time index of the extreme data is retained, and the determination is made by the following formula: If any one of the abnormal power judgment indicators meets the condition, it is considered an extreme weather scenario; otherwise, it is included in the normal weather dataset, and the time index of the extreme data is retained. ; in, For the first i Average power of each window This is the daily average. This represents the maximum peak-to-valley difference in the photovoltaic power output sequence.

[0025] Specifically, based on the abnormal meteorological judgment indicators, four meteorological indicators—temperature, irradiance, wind speed, and relative humidity—are selected to construct the abnormal weather judgment criteria. If any one of the meteorological indicators meets the conditions, it is considered an extreme weather scenario, which can be specifically divided into four types of weather: snow cover, fog / haze, severe convection, and dense fog. The corresponding power sequences are included in the abnormal weather dataset. The judgment criteria for the four types of weather—snow cover, fog / haze, severe convection, and dense fog—are given by the following formulas: ; in, Irradiance, The average temperature over 1 day. Relative humidity, The average wind speed, For each indicator to be considered as an extreme weather event, it must meet the corresponding conditions for the duration to be determined.

[0026] Subsequently, a Bayesian neural network is trained based on the photovoltaic power output sample dataset. Specifically, a Bayesian neural network is constructed, comprising an input layer, an output layer, and several hidden layers. The input layer receives meteorological and power characteristics from normal and abnormal weather datasets, and the output layer outputs the probability distribution parameters of photovoltaic power output. First, the prior distributions of the network's weights and bias parameters are... Set as normal distribution The model performance is optimized by adjusting the hyperparameters of the network structure and prior distribution.

[0027] In the Bayesian neural network optimization stage, variational inference is employed, combined with KL divergence to calculate the relative entropy of the two distributions to represent the difference in random variables. Specifically, the loss function is constructed using KL divergence. For the predicted distribution of photovoltaic power output, the loss function is given by the following equation: ; ; in, y For the target value, x For input values, Let the variational posterior distribution of the weights be... Let KL divergence be denoted as KL divergence.

[0028] Furthermore, using a stochastic variational inference method and aiming to minimize the aforementioned loss function, normal weather datasets and abnormal weather datasets are input into a Bayesian neural network for training to output the probability distribution parameters of photovoltaic power output. During the iteration process, the two distribution functions gradually approximate each other, and the KL divergence gradually decreases. When the divergence value is minimized, the corresponding approximate distribution is the equivalent true function distribution. After training, the Bayesian neural network can provide the probability distribution of photovoltaic power output and output the mean. and variance At the same time, at a given confidence level The following formula can be used to give the prediction interval: ; in, a , b These are the upper and lower bounds of the prediction interval, respectively. For two-sided quantiles of a normal distribution, the corresponding values ​​are taken according to the confidence level.

[0029] Furthermore, the root mean square error and mean absolute error are used to measure the difference between the frequency distribution curves of the photovoltaic power output prediction model and the actual power output, and are calculated using the following formulas respectively: ; ; in, MAE and RMSE These are the mean absolute error and the root mean square error, respectively.

[0030] Step S12: Construct a probability density function using the probability distribution parameters and historical photovoltaic power data, randomly sample the probability density function to generate a noise sequence, and input the noise sequence and historical photovoltaic power output sequence into a generative adversarial network to output a photovoltaic power output sequence that varies with time.

[0031] In this embodiment, based on the obtained distribution parameters, a probability density function is constructed by combining historical photovoltaic power data. A noise sequence with prior distribution information is generated through random sampling. This prior noise sequence and the historical photovoltaic power output sequence are then used as input to train a generative adversarial network (GAN) incorporating EM distance, resulting in a photovoltaic power output scenario. The generated normal scenario is then fused with the extreme scenario to obtain a photovoltaic power output scenario covering the complete time series. The GAN used is WGAN-GP (Wasserstein Generative Adversarial Network with Gradient Penalty).

[0032] Specifically, the network structure of generative adversarial networks is as follows: Figure 3 As shown, from the noise distribution Extracting noise z Input generator G Output generated samples Drawing real samples from the sample distribution x and generate samples Input discriminator D The parameter settings of the generative model that can significantly improve the quality of the generated scene are determined. This involves inputting a noisy sequence into the generator of a generative adversarial network (GAN) to output generated samples; selecting real samples from a photovoltaic power output sample dataset; and inputting the generated samples and real samples into the discriminator in the GAN to output the discrimination probabilities of the real samples and the generated samples. The discriminator distinguishes between real samples and generated samples, outputting corresponding results based on the different sources of the input data, as given by the following formula: ; in, For discriminator D The probability of discrimination against real samples. For discriminator D The discrimination probability of the generated sample.

[0033] Furthermore, the generator and discriminator are trained using the discrimination probabilities of the real samples and the generated samples to obtain the trained generative adversarial network. The generator aims to increase the discriminator's discrimination probability for generated samples, while the discriminator aims to minimize the discrimination probability for generated samples and maximize the discrimination probability for real samples. This is achieved through iterative training using a minimax game until convergence, and the estimation function is given by the following formula: ; in, for z Follows the sample distribution Expectations for x Follows the sample distribution Expectations; Furthermore, the Earth-Mover (EM) distance is introduced to calculate the distance between generated samples and real samples. Using Lipschitz continuity, it can be defined by the following formula: ; in, Discriminator D Lipschitz continuity should be ensured.

[0034] Finally, to optimize stability, the generator of the generative adversarial network here introduces gradient penalty instead of weight clipping. After combining the EM distance, the estimation function can be rewritten as follows: ; ; in, For neural network parameters, For gradient penalty, The generator produces a difference sample obtained by mixing real samples with the generated samples in a certain proportion. The gradient of the discriminator with respect to the interpolated samples. This represents the gradient penalty strength coefficient.

[0035] After training, the generator of the generative adversarial network can analyze the mapping relationship between noise sequences and historical photovoltaic power output sequences, that is, learn noise based on prior knowledge. z The mapping relationship between the output power and the output distribution is used to output a photovoltaic output sequence that shows the change of photovoltaic power over time.

[0036] Step S13: Based on the photovoltaic power output sequence, determine the power characteristics and time series characteristics, use a regression decision tree, and based on the power characteristics and time series characteristics, divide the photovoltaic power output sequence into a local Gaussian distribution subset, construct a Gaussian mixture model for the local Gaussian distribution subset, iteratively update the Gaussian mixture model to obtain the updated Gaussian mixture model, and construct an isolated forest based on the local Gaussian distribution subset.

[0037] In this embodiment, see Figure 2 The offline model training process, as shown, first uses a sliding window method based on the obtained photovoltaic power output sequences covering both normal and extreme scenarios to obtain power and time-series features. Specifically, the window size and step size of the second sliding window are set. Using the window size and step size of the second sliding window, the photovoltaic power output sequence is traversed to calculate the average power and peak-to-valley difference within each second window. Simultaneously, the first-order difference of the photovoltaic power output sequence is calculated to represent the changing trend. The first-order difference, the average power within each second window, and the peak-to-valley difference are used as the power and time-series features, respectively, and calculated using the following formulas: ; ; ; in, and These represent the peak-to-valley difference and average power of the photovoltaic power output sequence, respectively. This is the first-order difference of the photovoltaic power output sequence.

[0038] Furthermore, the power characteristics and time-series characteristics are normalized to obtain a normalized feature set. Based on this normalized feature set, a regression decision tree is used to divide the photovoltaic power output sequence into a subset of locally Gaussian-like distributions. Specifically, to segment the highly nonlinear power output scenario into a subset of locally Gaussian-like distributions adapted to a Gaussian mixture model, the features are first normalized by eliminating dimensions using the following formula: ; in, and These are the corresponding feature mean and feature standard deviation, respectively.

[0039] Furthermore, the input is the obtained photovoltaic power output sequence, and the feature set is extracted. Based on the criteria for judging extreme weather, the target is set as an extreme weather type label. The corresponding values ​​represent normal, snow-covered, haze, severe convection, and dense fog weather, respectively. By iterating through all features and their corresponding values, the space is progressively divided down to the mean squared error. MSEMinimize the value of the feature; at this point, the value corresponding to the feature is the final result of the node inside the decision tree, expressed by the following formula: ; ; in, For the sample i The input feature vector, j For feature index, s For corresponding features j The separator value, and The mean of the subset objective. and The dataset was selected The partitioned subset.

[0040] Furthermore, a grid search method is used to prune the decision tree, and k-fold cross-validation is used to improve the model's generalization ability, thereby preventing overfitting. The first step is to determine the hyperparameters to be tuned and their value ranges, and then traverse them using an exhaustive method. These hyperparameters are then input into the model to be trained, and cross-validation is used to select the optimal parameters. The finally selected hyperparameters are the maximum depth of the decision tree and the minimum number of samples in the leaf nodes. Assume the data contains... N Each sample is used to divide the data into... k Divide into equal parts, each subset containing N / k Samples, total number of iterations k One cycle; in each cycle i From this, select a subset as the test set. The union of the remaining subsets is used as the training set, and the mean squared error is used as the evaluation metric. The average mean squared error is taken to complete the model evaluation.

[0041] Subsequently, a Gaussian Mixture Model (GMM) is constructed for the local Gaussian distribution subset. The Gaussian Mixture Model is fitted to each segment subset, and each component of the Gaussian Mixture Model corresponds to a weather type. The formula for the Gaussian Mixture Model is shown below: ; ; in, For the first k The weights of a Gaussian distribution Let be a Gaussian probability density function. d For feature dimension, It is the mean vector. It is the covariance matrix; Furthermore, the Gaussian mixture model is iteratively updated. Specifically, the expectation-maximization (EM) algorithm is used to iteratively update the Gaussian mixture model, determining whether a preset iteration termination condition has been triggered. If the preset iteration termination condition is triggered, the iterative update is stopped, and the updated Gaussian mixture model is obtained. The EM algorithm consists of an expectation step and a maximization step. First, the posterior probability of a sample belonging to each Gaussian distribution is calculated: ; in, These are latent variables, representing samples. Belongs to the k A Gaussian-distributed indicator variable.

[0042] Then update the Gaussian mixture model parameters using the following formula: .

[0043] The expected step and the maximization step are performed alternately until the error is minimal and the preset iteration termination condition is triggered. The iteration stops, the model and data fitting are completed, and the updated Gaussian mixture model is obtained.

[0044] Finally, an isolated forest is constructed based on the aforementioned subset of local Gaussian-like distributions. Specifically, based on outlier detection using a segmentation-based Gaussian mixture model, samples for constructing the isolated forest are selected from the subset of local Gaussian-like distributions. First, data sampling is performed, selecting a subset of samples for each weather type, the number of which is much smaller than the dataset. These selected samples are used to construct isolated trees. Then, target features are randomly selected from all features of the samples, and the cutoff value of the target features is determined. The samples are then segmented using the cutoff value to obtain segmented samples. It is then determined whether a preset segmentation termination condition has been triggered. If the preset segmentation termination condition has been triggered, segmentation is stopped, resulting in isolated trees. These isolated trees are then used to form the isolated forest. That is, features are randomly selected. j and split values s The process is recursively segmented until the number of samples equals 1 (i.e., the preset segmentation termination condition is triggered). The above steps are repeated, and isolated trees form an isolated forest.

[0045] Furthermore, the sample path length is defined as the number of steps along the path from the root node to the leaf node. Outlier identification is based on the path length, which is calculated using the following formula: ; ; in, e This represents the number of paths from the root node to an isolated leaf node for a given sample, i.e., the number of cuts. This indicates the number of other samples located at the same leaf node as the given sample. The average path length of normal samples. is Euler's constant.

[0046] Furthermore, outlier detection is performed using the constructed isolated forest. For the target scene, anomaly score (i.e., type anomaly score) is used as the standard for judging outliers, with an overall range of (0, 1). The type anomaly score for each isolated tree is calculated first, and then the average is taken. The higher the type anomaly score, the closer the node is to the root node, and the more likely it is to be an outlier. Weights are assigned to the sample path lengths for different weather types, with higher weights for extreme weather types. The type anomaly score can be defined as follows: ; in, The average path length is defined as the mean path length of all isolated trees. Weather type , For type The weight of extreme weather events is determined by the score; the higher the score, the more likely the weather event is to fall into that category.

[0047] Step S14: Using the updated Gaussian mixture model and the isolated forest, calculate the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified, and determine the weather type of the weather to be identified based on the probability density and the anomaly score.

[0048] In this embodiment, the GMM-IF (Gaussian Mixture Model-Isolation Forest) algorithm is used to identify extreme scene types. First, a pre-trained Gaussian Mixture Model is called to identify the weather conditions of the target scene. The photovoltaic power output sequence corresponding to the weather to be identified is input into the updated Gaussian Mixture Model to output the probability density of the photovoltaic power output sequence corresponding to the weather to be identified. That is, based on the target scene input feature set into the pre-trained Gaussian Mixture Model, when the weather type is t_i, the GMM probability density of that weather type is calculated. As shown in the following formula: .

[0049] Furthermore, the photovoltaic power output sequence corresponding to the weather to be identified is input into the Isolation Forest to output the anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified. That is, the target scene is input into the trained Isolation Forest, and the type anomaly score of its weather type t_i is calculated using the aforementioned type anomaly score formula. .

[0050] Furthermore, weighting coefficients are set for the probability density and the anomaly score. These weighting coefficients are used to weight the probability density and the anomaly score to obtain a comprehensive score. Based on this comprehensive score, the weather type to be identified is determined. That is, the scene type of the target scene is determined through joint decision-making using GMM probability density and type anomaly score. A comprehensive score is then proposed by combining both, and finally, the weather type with the highest comprehensive score is output. The comprehensive score for the scene with weather type t_i is given by the following formula: ; in, These are weighting coefficients, which can be adjusted through cross-validation.

[0051] This step dynamically weights the probability density of the Gaussian mixture model with the anomaly score of the isolated forest, which enhances the noise resistance to extreme weather disturbances while maintaining local probability modeling, thus obtaining more accurate and stable weather type identification results.

[0052] To further illustrate the recognition effect of the embodiments of this application, the following examples are provided in conjunction with the accompanying drawings. A schematic diagram of data recognition is shown below. Figure 4 As shown, the horizontal axis represents the number of days, and the vertical axis represents whether the day is an extreme weather event. It can be seen that the method described in this application embodiment can effectively identify extreme weather days distributed in different periods of the year from photovoltaic power output scenarios covering the complete time series, demonstrating the ability to capture the spatiotemporal distribution characteristics of extreme weather.

[0053] The regression analysis results of predicted power and actual power on the training set are as follows: Figure 5 As shown, the regression analysis results of predicted power and actual power on the test set are as follows: Figure 6 As shown. Figure 5 and Figure 6 The scattered points are closely distributed on both sides of the baseline. Both the training set and the test set achieved high coefficients of determination and maintained low root mean square error and mean absolute error. This indicates that the method described in this application embodiment maintains a high degree of consistency between the predicted value and the actual value in both low-value and medium-to-high-value regions, and has good fitting accuracy and stability for photovoltaic power output under extreme weather scenarios.

[0054] This application provides a weather type identification method for distributed photovoltaic (PV) forecast uncertainty. The method includes constructing a PV output sample dataset using historical operational data of the target area, training a Bayesian neural network based on this dataset to obtain probability distribution parameters for PV output. This quantifies the uncertainty of PV output, improving the realism and reliability of subsequent scenario generation. A probability density function is constructed using the probability distribution parameters and historical PV power data. Random sampling is performed on the probability density function to generate a noise sequence. The noise sequence and historical PV output sequences are input into a generative adversarial network (GAN) to output a PV output sequence that varies with time. Therefore, this method can generate PV output sequences that conform to the true probability distribution and time series characteristics, effectively expanding the training samples for extreme weather scenarios and improving the diversity and representativeness of the training data. The power and time-series characteristics of photovoltaic (PV) output sequences are determined. Regression decision trees are used to divide the PV output sequences into subsets of local Gaussian distributions. A Gaussian mixture model is constructed for these subsets and iteratively updated. An isolated forest is then built based on these subsets. This improves the fitting accuracy of the Gaussian mixture model to the local probability distribution of PV output, enhances the ability to identify anomalous output characteristics under extreme weather conditions, and improves the accuracy and robustness of PV output characteristic modeling in complex weather scenarios. Using the updated Gaussian mixture model and the isolated forest, the probability density and anomaly score of the PV output sequence corresponding to the weather to be identified are calculated. Based on the probability density and anomaly score, the weather type of the weather to be identified is determined, which simultaneously considers the degree of weather type matching and anomalous characteristics, improving the accuracy, stability, and reliability of weather type identification.

[0055] See Figure 7 As shown in the figure, this application discloses a weather type identification device for distributed photovoltaic forecasting uncertainty, which may specifically include: Training module 11 is used to construct a photovoltaic power output sample dataset using historical operating data of the target area, and to train a Bayesian neural network based on the photovoltaic power output sample dataset to obtain the probability distribution parameters of photovoltaic power output. The photovoltaic power output sequence determination module 12 is used to construct a probability density function using the probability distribution parameters and historical photovoltaic power data, randomly sample the probability density function to generate a noise sequence, and input the noise sequence and historical photovoltaic power output sequence into a generative adversarial network to output a photovoltaic power output sequence that varies with time. The isolated forest construction module 13 is used to determine the power characteristics and time series characteristics based on the photovoltaic power output sequence, use a regression decision tree, and divide the photovoltaic power output sequence into local Gaussian distribution subsets based on the power characteristics and the time series characteristics, construct a Gaussian mixture model for the local Gaussian distribution subsets, iteratively update the Gaussian mixture model to obtain the updated Gaussian mixture model, and construct an isolated forest based on the local Gaussian distribution subsets. The weather type determination module 14 is used to calculate the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified using the updated Gaussian mixture model and the isolated forest, and to determine the weather type of the weather to be identified based on the probability density and the anomaly score.

[0056] In some specific embodiments, the training module 11 may specifically include: The meteorological element extraction module is used to extract various meteorological element data from the historical operational data of the target area at fixed time intervals; the meteorological element data includes photovoltaic power, wind speed, temperature, relative humidity, and irradiance; The dataset construction module is used to construct a photovoltaic power output sample dataset using the meteorological element data mentioned above. The first window setting module is used to set the window size and step size of the first sliding window; An abnormal power index construction module is used to traverse the photovoltaic power output sample dataset using the window size and step size of the first sliding window to calculate the average power and peak-valley difference within each first window, and use the average power and peak-valley difference within each first window as an abnormal power judgment index. The abnormal weather indicator construction module is used to construct abnormal weather judgment indicators; An abnormal weather dataset filtering module is used to filter out abnormal weather datasets from the photovoltaic power output sample dataset based on the abnormal power judgment index and the abnormal meteorological judgment index. The normal weather dataset determination module is used to identify other datasets in the photovoltaic power output sample dataset besides the abnormal weather dataset as normal weather datasets.

[0057] In some specific embodiments, the training module 11 may specifically include: A Bayesian neural network construction module is used to construct a Bayesian neural network; the Bayesian neural network includes an input layer, an output layer, and several hidden layers; The loss function construction module is used to construct loss functions using KL divergence; The probability distribution parameter output module is used to train a Bayesian neural network by inputting normal weather datasets and abnormal weather datasets into the network with the goal of minimizing the loss function, using a stochastic variational inference method to output the probability distribution parameters of photovoltaic power output.

[0058] In some specific embodiments, the photovoltaic power output sequence determination module 12 may specifically include: The sample output module is used to input the noise sequence into the generator of the generative adversarial network to output generated samples; The real sample filtering module is used to filter out real samples from the photovoltaic power output sample dataset. The discrimination probability output module is used to input the generated sample and the real sample into the discriminator in the generative adversarial network, so as to output the discrimination probability of the real sample and the discrimination probability of the generated sample; A generative adversarial network training module is used to train the generator and the discriminator using the discrimination probabilities of the real samples and the discrimination probabilities of the generated samples, so as to obtain the trained generative adversarial network. The mapping relationship analysis module is used to analyze the mapping relationship between noise sequences and historical photovoltaic power output sequences using a trained generative adversarial network, so as to obtain the photovoltaic power output sequence that varies with time.

[0059] In some specific embodiments, the isolated forest construction module 13 may specifically include: The second window setting module is used to set the window size and step size of the second sliding window; The window feature calculation module is used to traverse the photovoltaic power output sequence by utilizing the window size and step size of the second sliding window to calculate the average power and peak-valley difference within each second window; The feature determination module is used to calculate the first-order difference of the photovoltaic power output sequence, and uses the first-order difference, the average power in each second window, and the peak-valley difference as power features and time series features. The normalization processing module is used to normalize the power characteristics and the time-series characteristics to obtain a normalized feature set; The subset partitioning module is used to partition the photovoltaic power output sequence into locally Gaussian-like subsets based on the normalized feature set and using a regression decision tree.

[0060] In some specific embodiments, the isolated forest construction module 13 may specifically include: The iterative update module is used to iteratively update the Gaussian mixture model using the expectation-maximization algorithm and determine whether the preset iteration termination condition has been triggered. The iteration stop module is used to stop the iteration update if the preset iteration end condition is triggered, and obtain the updated Gaussian mixture model. The sample selection module is used to select samples from a subset of local Gaussian-like distributions for constructing an isolated forest. The sample segmentation module is used to randomly select target features from all features of the sample, determine the cut value of the target features, and use the cut value to segment the sample to obtain segmented samples. The segmentation judgment module is used to determine whether the preset segmentation end condition has been triggered. The isolated tree acquisition module is used to stop the splitting process and obtain an isolated tree if the preset splitting end condition is triggered. An isolated forest forming module is used to form an isolated forest using the isolated trees.

[0061] In some specific embodiments, the weather type determination module 14 may specifically include: The probability density output module is used to input the photovoltaic power output sequence corresponding to the weather to be identified into the updated Gaussian mixture model to output the probability density of the photovoltaic power output sequence corresponding to the weather to be identified. The anomaly score output module is used to input the photovoltaic power output sequence corresponding to the weather to be identified into the isolated forest, so as to output the anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified; The weighting coefficient setting module is used to set the weighting coefficients of the probability density and the anomaly score; The comprehensive score calculation module is used to weight the probability density and the outlier score using the weighting coefficients to obtain a comprehensive score; The weather type determination submodule is used to determine the weather type of the weather to be identified based on the comprehensive score.

[0062] Figure 8 This is a schematic diagram of an electronic device provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the weather type identification method for distributed photovoltaic forecast uncertainty disclosed in any of the foregoing embodiments.

[0063] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0064] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0065] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the data 223 in the memory 22. The operating system 221 can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the weather type identification method for distributed photovoltaic forecast uncertainty executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the weather type identification device for distributed photovoltaic forecast uncertainty transmitted from external devices, as well as data collected by its own input / output interface 25.

[0066] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main 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 known in the art.

[0067] Furthermore, this application also discloses a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the weather type identification method steps for distributed photovoltaic forecast uncertainty disclosed in any of the foregoing embodiments.

[0068] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0069] The above provides a detailed description of the weather type identification method, apparatus, equipment, and medium for distributed photovoltaic forecasting uncertainty provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for weather type identification in the face of uncertainties in distributed photovoltaic forecasting, characterized in that, include: A photovoltaic power output sample dataset is constructed using historical operating data of the target area. A Bayesian neural network is trained based on the photovoltaic power output sample dataset to obtain the probability distribution parameters of photovoltaic power output. A probability density function is constructed using the probability distribution parameters and historical photovoltaic power data. The probability density function is randomly sampled to generate a noise sequence. The noise sequence and the historical photovoltaic power output sequence are input into a generative adversarial network to output a photovoltaic power output sequence that varies with time. Based on the photovoltaic power output sequence, power characteristics and time series characteristics are determined. Using a regression decision tree, and based on the power characteristics and time series characteristics, the photovoltaic power output sequence is divided into local Gaussian distribution subsets. A Gaussian mixture model is constructed for the local Gaussian distribution subsets. The Gaussian mixture model is iteratively updated to obtain the updated Gaussian mixture model. An isolated forest is constructed based on the local Gaussian distribution subsets. Using the updated Gaussian mixture model and the isolated forest, the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified are calculated respectively. Based on the probability density and the anomaly score, the weather type of the weather to be identified is determined.

2. The weather type identification method for distributed photovoltaic forecasting uncertainty according to claim 1, characterized in that, The process of constructing a photovoltaic power output sample dataset using historical operating data of the target area includes: Meteorological element data are extracted from historical operational data of the target area at fixed time intervals; the meteorological element data includes photovoltaic power, wind speed, temperature, relative humidity, and irradiance. Using the meteorological data mentioned above, a photovoltaic power output sample dataset is constructed; Set the window size and step size of the first sliding window; Using the window size and step size of the first sliding window, the photovoltaic power output sample dataset is traversed to calculate the average power and peak-valley difference within each first window, and the average power and peak-valley difference within each first window are used as indicators for judging abnormal power. Construct indicators for judging abnormal weather conditions; Based on the abnormal power judgment index and the abnormal weather judgment index, an abnormal weather dataset is selected from the photovoltaic power output sample dataset. The datasets other than the abnormal weather dataset in the photovoltaic power output sample dataset are used as the normal weather dataset.

3. The weather type identification method for distributed photovoltaic forecasting uncertainty according to claim 2, characterized in that, The training of the Bayesian neural network based on the photovoltaic power output sample dataset includes: Construct a Bayesian neural network; the Bayesian neural network includes an input layer, an output layer, and several hidden layers; Construct a loss function using KL divergence; Using a stochastic variational inference method, and with the goal of minimizing the loss function, normal weather datasets and abnormal weather datasets are input into a Bayesian neural network for training, so as to output the probability distribution parameters of photovoltaic power output.

4. The weather type identification method for distributed photovoltaic forecasting uncertainty according to claim 1, characterized in that, The step of inputting the noise sequence and the historical photovoltaic power output sequence into a generative adversarial network to output a photovoltaic power output sequence that varies with time includes: The noisy sequence is input into the generator of the generative adversarial network to output generated samples; Select real samples from the photovoltaic power output sample dataset; The generated sample and the real sample are input into the discriminator in the generative adversarial network to output the discrimination probability of the real sample and the discrimination probability of the generated sample; The generator and discriminator are trained using the discrimination probabilities of the real samples and the discrimination probabilities of the generated samples to obtain the trained generative adversarial network. The mapping relationship between noise sequences and historical photovoltaic power output sequences is analyzed using a trained generative adversarial network to obtain the photovoltaic power output sequence that varies with time.

5. The weather type identification method for distributed photovoltaic forecasting uncertainty according to claim 1, characterized in that, The process involves determining power and time-series characteristics based on the photovoltaic output sequence, utilizing a regression decision tree, and dividing the photovoltaic output sequence into locally Gaussian-like subsets based on the power and time-series characteristics, including: Set the window size and step size of the second sliding window; Using the window size and step size of the second sliding window, the photovoltaic power output sequence is traversed to calculate the average power and peak-to-valley difference within each second window; Calculate the first-order difference of the photovoltaic power output sequence, and use the first-order difference, the average power within each second window, and the peak-valley difference as power characteristics and time series characteristics, respectively. The power characteristics and the timing characteristics are normalized to obtain a normalized feature set; Based on the normalized feature set and using a regression decision tree, the photovoltaic power output sequence is divided into a subset of local Gaussian distributions.

6. The weather type identification method for distributed photovoltaic forecasting uncertainty according to claim 1, characterized in that, The iterative update of the Gaussian mixture model to obtain the updated Gaussian mixture model, and the construction of an isolated forest based on the local Gaussian distribution subset, includes: The Gaussian mixture model is iteratively updated using the expectation-maximization algorithm, and it is determined whether the preset iteration termination condition has been triggered. If the preset iteration termination condition is triggered, the iteration update will stop, and the updated Gaussian mixture model will be obtained. Samples for constructing an isolated forest are selected from a subset of local Gaussian-like distributions. Target features are randomly selected from all features of the sample, and the cut value of the target features is determined. The sample is then segmented using the cut value to obtain the segmented sample. Determine whether the preset segmentation termination condition has been triggered. If the preset segmentation termination condition is triggered, the segmentation will stop, resulting in an isolated tree. The isolated trees are used to form an isolated forest.

7. The weather type identification method for distributed photovoltaic forecasting uncertainty according to any one of claims 1 to 6, characterized in that, The process involves using the updated Gaussian mixture model and the isolated forest to calculate the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified, and determining the weather type based on the probability density and the anomaly score, including: The photovoltaic power output sequence corresponding to the weather to be identified is input into the updated Gaussian mixture model to output the probability density of the photovoltaic power output sequence corresponding to the weather to be identified. The photovoltaic power output sequence corresponding to the weather to be identified is input into the isolated forest to output the anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified. Set the probability density and the weighting coefficients for the anomaly scores; The probability density and the anomaly score are weighted using the weighting coefficients to obtain a comprehensive score; The weather type to be identified is determined based on the comprehensive score.

8. A weather type identification device for distributed photovoltaic forecasting uncertainty, characterized in that, include: The training module is used to construct a photovoltaic power output sample dataset using historical operating data of the target area, and to train a Bayesian neural network based on the photovoltaic power output sample dataset to obtain the probability distribution parameters of photovoltaic power output. The photovoltaic power output sequence determination module is used to construct a probability density function using the probability distribution parameters and historical photovoltaic power data, randomly sample the probability density function to generate a noise sequence, and input the noise sequence and historical photovoltaic power output sequence into a generative adversarial network to output a photovoltaic power output sequence that varies with time. An isolated forest construction module is used to determine power characteristics and time-series characteristics based on the photovoltaic power output sequence, utilize a regression decision tree, and divide the photovoltaic power output sequence into local Gaussian-like subsets based on the power characteristics and time-series characteristics, construct a Gaussian mixture model for the local Gaussian-like subsets, iteratively update the Gaussian mixture model to obtain the updated Gaussian mixture model, and construct an isolated forest based on the local Gaussian-like subsets. The weather type determination module is used to calculate the probability density and anomaly score of the photovoltaic power output sequence corresponding to the weather to be identified using the updated Gaussian mixture model and the isolated forest, and to determine the weather type of the weather to be identified based on the probability density and the anomaly score.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the weather type identification method for distributed photovoltaic forecast uncertainty as described in any one of claims 1 to 7.

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, it implements the weather type identification method for distributed photovoltaic forecast uncertainty as described in any one of claims 1 to 7.

Images