Distributed photovoltaic power prediction method, system and device based on Gaussian mixture model and medium

Abstract

The invention discloses a distributed photovoltaic power prediction method, system and device based on a Gaussian mixture model and a medium, and belongs to the technical field of photovoltaic power prediction.The distributed photovoltaic power prediction method comprises the steps that a time sequence vector is collected, principal component analysis is carried out on the time sequence vector, low-dimensional feature representation is obtained, and a power feature vector of each photovoltaic power station is formed; performing clustering analysis based on a probability model on the power feature vector to generate a plurality of photovoltaic power station clusters; for each cluster, acquiring meteorological input data through a set data source priority rule and a completion mechanism; and constructing a neural network power prediction model based on the accumulated power data in the cluster and the corresponding meteorological features, and outputting a future power generation power prediction value of the photovoltaic power station in the corresponding cluster. According to the invention, N photovoltaic power stations in a region are divided into M clusters through a GMM clustering method, so that the design is simplified; and the power prediction of the whole area is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power prediction, in particular to a distributed photovoltaic power prediction method, system, device and medium based on a Gaussian mixture model. BACKGROUND

[0002] In the field of photovoltaic power generation, power prediction of photovoltaic power stations is crucial for efficient energy management and optimization of power grid operation. Current power prediction methods involve multiple influencing factors such as wind speed, wind direction, temperature, humidity, air pressure, and light intensity, among which the BP neural network method has been widely used. However, if there are numerous photovoltaic power stations in a region, directly using all station data for prediction can lead to excessive computational load and increased model complexity.

[0003] In addition, when there are a large number of photovoltaic power stations in a certain region, some with large installed capacity directly connect to the grid through medium voltage (10kV), and these stations have complete environmental collection systems that can collect data such as wind speed, wind direction, temperature, humidity, air pressure, light intensity, and real-time power, making them suitable for power prediction. However, most of the stations with small installed capacity directly connect to the grid through low voltage (380V), and these stations can only collect real-time power data, making them unsuitable for power prediction.

[0004] Since low-voltage grid-connected photovoltaic power stations do not have the conditions for power prediction, a solution is needed to predict these stations to achieve regional prediction.

[0005] Traditional clustering methods have limitations, such as DBSCAN relying on manual setting of neighborhood parameters (ε, MinPts), which cannot effectively describe the probability distribution characteristics of continuous power data; hard clustering cannot handle boundary ambiguous stations (power characteristics between multiple clusters), leading to grouping errors; and directly processing 365-dimensional single-day power data has high computational complexity and slow model convergence. Therefore, new methods are needed. SUMMARY

[0006] In view of the above problems, the present application is proposed.

[0007] Therefore, the technical problem solved by the present application is: how to achieve efficient grouping of distributed photovoltaic power stations through probability modeling and automatic cluster number selection, and improve the accuracy and robustness of regional power prediction.

[0008] To solve the above technical problems, the present application provides the following technical solutions: a distributed photovoltaic power prediction method based on a Gaussian mixture model, comprising,

[0009] The historical power data of the distributed photovoltaic power station is collected, data preprocessing is performed, and the power data of each power station is divided by day, and the time series vector representing the daily maximum power characteristics is extracted; principal component analysis is performed on the time series vector to obtain a low-dimensional feature representation, which constitutes the power feature vector of each photovoltaic power station; the power feature vector is subjected to clustering analysis based on a probability model to generate a plurality of photovoltaic power station clustering clusters; for each clustering cluster, meteorological input data is obtained through a set data source priority rule and a completion mechanism; based on the cumulative power data and the corresponding meteorological characteristics in the clustering cluster, a neural network power prediction model is constructed, and the parameter weight used for prediction is obtained through training; the weather forecast information is input into the prediction model, and the future power generation power prediction value of the photovoltaic power station in the corresponding clustering cluster is output.

[0010] As a preferred scheme of the distributed photovoltaic power prediction method based on a Gaussian mixture model, wherein: the extraction of the time series vector representing the daily maximum power characteristics includes: normalizing the real-time power data of each power station, constructing an annual power curve based on the normalized data, and dividing the annual power curve into a plurality of data segments by day; for each daily data segment, the average value of a plurality of maximum power points is extracted as a representative feature value of the day, and a time series vector representing the annual power characteristics is generated.

[0011] As a preferred scheme of the distributed photovoltaic power prediction method based on a Gaussian mixture model, wherein: the principal component analysis includes: based on the time series vector of all photovoltaic power stations, calculating the covariance matrix thereof, and extracting principal components through eigenvalue decomposition; the original vector is linearly transformed according to the principal components to obtain a reduced feature representation.

[0012] As a preferred scheme of the distributed photovoltaic power prediction method based on a Gaussian mixture model, wherein: the clustering analysis based on a probability model includes: within a set range of candidate clustering cluster numbers, for each candidate cluster number, a corresponding Gaussian mixture model is constructed, and the model score is calculated based on the Bayesian information criterion, and the cluster number with the lowest score is selected as the optimal clustering number; based on the selected cluster number, the model parameters are iteratively estimated using the expectation maximization algorithm until the preset convergence condition is met, and the model training is completed.

[0013] As a preferred solution of the distributed photovoltaic power prediction method based on the Gaussian mixture model described in the present invention, the obtaining of meteorological input data includes determining whether there is a photovoltaic power station with meteorological monitoring capabilities within the target cluster; if so, selecting the meteorological data of the photovoltaic power station with the largest power generation capacity as the meteorological input data of the cluster; if not, based on the power station with the largest power generation capacity within the cluster, searching for external sites with meteorological data that are geographically adjacent to the power station with the largest power generation capacity, and using a spatial interpolation method to estimate the meteorological input data of the cluster.

[0014] As a preferred solution of the distributed photovoltaic power prediction method based on the Gaussian mixture model described in the present invention, the construction of the neural network power prediction model includes, for each cluster, collecting the cumulative historical power generation power and corresponding meteorological characteristic data of the photovoltaic power station in the cluster to construct a training data set; establishing a single hidden layer feedforward neural network model based on the training data set, the model includes an input layer, a hidden layer and an output layer, wherein: the input features of the input layer include at least one meteorological element of wind speed, wind direction, temperature, humidity, air pressure and light intensity; the output value of the output layer is the future power generation power of the photovoltaic power station in the corresponding cluster; and optimizing the neural network parameters through the back propagation algorithm to minimize the mean square error between the predicted output and the actual power.

[0015] The beneficial effects of this preferred technical solution are as follows: by constructing a single hidden layer feedforward neural network power prediction model for clusters, it is possible to effectively model the historical cumulative power within the clusters and the corresponding meteorological characteristics, thereby improving the model's generalization ability under conditions of spatial heterogeneity. The use of an input layer design that includes multiple key meteorological elements enhances the model's ability to perceive environmental changes, helping to improve the accuracy and robustness of power generation predictions. At the same time, the backpropagation algorithm is used to optimize the model parameters, making model training more efficient and prediction errors smaller, thereby meeting the high-precision prediction requirements of distributed photovoltaic power stations in actual scheduling and operation and maintenance.

[0016] As a preferred embodiment of the distributed photovoltaic power prediction method based on a Gaussian mixture model described in the present invention, the output of the predicted value of future power generation power of the photovoltaic power station in the corresponding cluster includes inputting weather forecast data consistent with the meteorological elements used in the model training phase into the neural network power prediction model; outputting the predicted result of future power generation power of the corresponding cluster through the model; and based on the prediction result, combining the historical power distribution characteristics of each photovoltaic power station in the cluster, allocating or estimating the predicted power among the power stations.

[0017] The beneficial effect of the preferred technical solution is that by inputting weather forecast data of meteorological elements consistent with the structure and training phase into the prediction model, the consistency and effectiveness of the model input features are ensured, and the credibility of the prediction result is improved.

[0018] The application provides a distributed photovoltaic power prediction system based on a Gaussian mixture model.

[0019] To solve the above technical problems, the application provides the following technical scheme: a distributed photovoltaic power prediction system based on a Gaussian mixture model, comprising: a data acquisition module, which is used for acquiring historical power data of distributed photovoltaic power stations, performing data preprocessing, dividing the power data of each power station by day, and extracting a time sequence vector representing daily maximum power characteristics; a feature extraction module, which is used for performing principal component analysis on the time sequence vector, obtaining a low-dimensional feature representation, and constituting a power feature vector of each photovoltaic power station; a clustering analysis module, which is used for performing clustering analysis on the power feature vector based on a probability model, and generating a plurality of photovoltaic power station clustering clusters; a data input module, which is used for obtaining meteorological input data through a set data source priority rule and a completion mechanism for each clustering cluster; a model training module, which is used for constructing a neural network power prediction model based on cumulative power data and corresponding meteorological characteristics in the clustering cluster, and obtaining parameter weights for prediction through training; and a power prediction module, which is used for inputting weather forecast information into the prediction model and outputting a future power generation power prediction value of the photovoltaic power station in the corresponding clustering cluster.

[0020] The application provides a computer device, comprising a memory and a processor, and the memory stores a computer program.

[0021] The application provides a computer readable storage medium, which stores a computer program.

[0022] The beneficial effect of the application is that N photovoltaic power stations in a region are divided into M clusters by the GMM clustering method, which simplifies the design, and finally realizes the power prediction of the entire region. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 An overall flow chart of a distributed photovoltaic power prediction method based on a Gaussian mixture model provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0025] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0026] Example 1, with reference to Figure 1 , is an embodiment of the present invention, which provides a distributed photovoltaic power prediction method based on a Gaussian mixture model, comprising:

[0027] Step 101: Collect historical power data of distributed photovoltaic power stations, perform data preprocessing, and divide the power data of each power station by day to extract the time series vector representing the daily maximum power characteristics; Step 102: Perform principal component analysis on the time series vector to obtain low-dimensional feature representation to form the power feature vector of each photovoltaic power station; Step 103: Perform cluster analysis on the power feature vector based on a probability model to generate multiple photovoltaic power station clusters; Step 104: For each cluster, obtain meteorological input data through the set data source priority rules and completion mechanism; Step 105: Based on the accumulated power data and corresponding meteorological characteristics in the cluster, construct a neural network power prediction model, and obtain the parameter weights used for prediction through training; Step 106: Input weather forecast information into the prediction model and output the future power generation power forecast value of the photovoltaic power station in the corresponding cluster.

[0028] In step 101, the extraction of the time series vector representing the daily maximum power characteristics includes normalizing the real-time power data of each power station, constructing an annual power curve based on the normalized data, and dividing the annual power curve into multiple data segments by day; for each daily data segment, extracting the average value of several maximum power points as the representative characteristic value of the day, and generating a time series vector representing the power characteristics of the whole year.

[0029] In step 102, the principal component analysis includes calculating the covariance matrix based on the time series vectors of all photovoltaic power stations, and extracting principal components by eigenvalue decomposition, performing linear transformation on the original vectors according to the principal components, and obtaining the low-dimensional feature representation.

[0030] In a preferred embodiment of the present application, the principal component analysis on the time series vectors includes constructing a covariance matrix based on the time series vectors representing the daily maximum power features of all photovoltaic power stations, performing eigenvalue decomposition on the covariance matrix, selecting the first several principal components with a cumulative variance contribution rate reaching a set threshold (such as 95%) as the linear combination basis of the low-dimensional representation. The specific steps are as follows: constructing a power time series vector matrix and performing mean centering processing; calculating the covariance matrix; performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors; arranging the first k eigenvectors in descending order of variance contribution rate to form a projection matrix; and multiplying the original data by the projection matrix to obtain the low-dimensional power feature vector of each power station.

[0031] The preferred technical solution has the beneficial effects that: the principal component analysis realizes effective compression of high-dimensional power features, reduces the computational burden of the subsequent clustering algorithm, and significantly improves the stability of the model training process; especially in the scenario where the data dimension is much larger than the sample number, the negative impact of the curse of dimensionality on the model accuracy can be effectively prevented.

[0032] In an optional embodiment of the present application, the dimension reduction process can be replaced by other linear or nonlinear feature dimension reduction methods, including but not limited to independent component analysis (ICA), linear discriminant analysis (LDA), t-distributed stochastic neighbor embedding (t-SNE) or autoencoder (Autoencoder) method. Among them, ICA can separate potential independent source signals and is suitable for scenarios with greater noise influence; LDA is suitable for supervised dimension reduction with known class labels and can enhance the separability between clusters; t-SNE is suitable for two-dimensional visualization and structure preservation; and autoencoder can automatically learn nonlinear dimension reduction representation in an end-to-end framework and is suitable for complex pattern recognition. Such methods provide more adaptive dimension reduction solutions for different power station distribution characteristics, which helps to improve the accuracy and flexibility in the clustering process.

[0033] In step 103, the clustering analysis based on the probability model includes, within a set range of candidate clustering cluster numbers, constructing a corresponding Gaussian mixture model for each candidate cluster number, calculating the model score based on the Bayesian information criterion, and selecting the cluster number with the lowest score as the optimal cluster number; based on the selected cluster number, the expectation maximization algorithm is used to iteratively estimate the model parameters until the preset convergence condition is met, and the model training is completed.

[0034] In step 104, the acquisition of the meteorological input data comprises: determining whether there is a photovoltaic power station with meteorological monitoring capability in the target cluster; if there is, selecting the meteorological data of the photovoltaic power station with the maximum power generation as the meteorological input data of the cluster; and if there is not, searching for an external site with meteorological data based on the photovoltaic power station with the maximum power generation in the cluster and its adjacent geographical position, and estimating the meteorological input data of the cluster by using a spatial interpolation method.

[0035] In a preferred embodiment of the present application, the cluster analysis of the power feature vectors based on the probability model is soft clustering processing by using a Gaussian mixture model. Specifically, it comprises: firstly setting the range of the number of candidate clusters; then constructing a corresponding Gaussian mixture model for each cluster number, calculating the fitting degree of the model on the sample data by using maximum likelihood estimation; then evaluating the advantages and disadvantages of each model structure based on the Bayesian information criterion (BIC), and selecting the cluster number with the lowest score as the optimal cluster number; and finally, based on the selected cluster number, iteratively estimating the mean, covariance matrix and mixing coefficient of the Gaussian mixture model by using the expectation maximization (EM) algorithm until the log-likelihood function converges, and obtaining the clustering result.

[0036] The method allows each photovoltaic power station to be attributed to multiple clusters in a probabilistic form, realizes reasonable division of boundary fuzzy power stations, and improves the robustness of grouping and the model training effect.

[0037] The preferred technical scheme has the beneficial effects that: by introducing the Gaussian mixture model as the clustering algorithm and automatically determining the optimal cluster number based on the Bayesian information criterion, the system can effectively adapt to the feature distribution of the photovoltaic power stations in different regions under unsupervised conditions, and reduce the error caused by manual parameter setting; at the same time, compared with the hard clustering methods such as K-means or DBSCAN, the GMM has stronger expression ability and boundary processing ability, and is especially suitable for the division of power stations in continuous or overlapping regions.

[0038] In an optional embodiment of the present application, the cluster analysis based on the probability model can also use an improved K-means algorithm or a spectral clustering algorithm. Specifically: in the K-means mode, the initial center point selection is performed by K-means++, so as to improve the initial convergence stability; for spectral clustering, a similarity matrix can be constructed based on the power feature vectors, a low-dimensional embedding space is obtained by eigenvalue decomposition of the Laplacian matrix, and then a traditional clustering method is used for clustering and distribution; the number of clusters can be determined by using the silhouette coefficient or the elbow rule.

[0039] Although this mode does not have the soft clustering feature, it has the advantages of low computational complexity and easy implementation, and is suitable for scenarios where the number of photovoltaic power stations is large but the distribution is regular.

[0040] In step 105, the construction of the neural network power prediction model comprises, for each cluster, collecting the cumulative historical power generation of the photovoltaic power station in the cluster and the corresponding meteorological feature data to construct a training data set; a single-hidden-layer feedforward neural network model is established based on the training data set, and the model comprises an input layer, a hidden layer and an output layer, wherein: the input features of the input layer comprise at least one of wind speed, wind direction, temperature, humidity, air pressure and light intensity; the output value of the output layer is the future power generation of the photovoltaic power station in the corresponding cluster; and the neural network parameters are optimized through a back propagation algorithm to minimize the mean square error between the predicted output and the actual power.

[0041] In a preferred embodiment of the present application, the neural network power prediction model is a single-hidden-layer feedforward neural network, and the structure thereof comprises an input layer, a hidden layer and an output layer. Specifically, the structure comprises:

[0042] The input features of the input layer are the meteorological data obtained in step 104, and the input features comprise at least one of wind speed, wind direction, temperature, humidity, air pressure and light intensity; the hidden layer adopts an S-shaped activation function (Sigmoid function) to improve the nonlinear fitting capability; the output of the output layer is the future cumulative power generation of all photovoltaic power stations in the cluster; and the neural network is trained through a back propagation algorithm, and the objective is to minimize the mean square error between the predicted value of the model and the actual power value, and the model parameters are iteratively updated until convergence; in order to prevent overfitting, a cross-validation method can be used to evaluate the generalization performance of the model, or an early stopping mechanism can be introduced to control the number of training rounds.

[0043] The preferred technical solution has the beneficial effects that: the single-hidden-layer feedforward neural network structure adopted achieves a good balance between computational efficiency and fitting capability, and is particularly suitable for practical scenarios with limited meteorological feature dimensions and moderate training data volumes; meanwhile, by standardizing the meteorological data input and the cumulative power output, the convergence speed and the prediction stability of the model are improved, and the power prediction accuracy in the distributed photovoltaic scenario is significantly improved.

[0044] In an optional embodiment of the present application, the neural network power prediction model can also adopt a deep learning model such as a multilayer perceptron, a convolutional neural network or a long short-term memory network (LSTM). Specifically, the multilayer perceptron can be used to model complex nonlinear relationships and is suitable for regions with a large number of feature quantities; the convolutional neural network can extract local correlations in time series meteorological features, which helps to improve the short-term prediction accuracy; and the LSTM network is suitable for modeling input sequences with time correlation and can provide better performance in medium and long-term power prediction scenarios. Through the use of the above alternative models, the structure can be flexibly selected according to the prediction time scale, the number of meteorological features and the computing resources, and the adaptive ability of the system in different environments is enhanced.

[0045] In step 106, the output of the future power generation prediction value of the photovoltaic power station in the corresponding clustering cluster includes inputting weather forecast data consistent with the meteorological elements used in the model training stage into the neural network power prediction model; outputting the future period power generation prediction result of the corresponding clustering cluster through the model; and based on the prediction result, combining the historical power distribution characteristics of each photovoltaic power station in the clustering cluster, the predicted power is allocated or estimated among the power stations.

[0046] Embodiment 2 is an embodiment of the present application, which provides a distributed photovoltaic power prediction method based on Gaussian mixture model based on the previous embodiment, comprising:

[0047] In step 101, it is assumed that there are N photovoltaic power stations in the region. First, collect real-time power and related data of all photovoltaic power stations, if there is meteorological data of a photovoltaic power station, obtain the corresponding meteorological data at the same time, including but not limited to wind speed, wind direction, temperature, humidity, air pressure, and light intensity of the photovoltaic power station.

[0048] After that, the collected data is preprocessed, including data cleaning: processing of abnormal values, missing values and noise values, standardization and normalization of data. After this step, there are N power curves in total, and the power curve completes time regularization (all), that is, a certain time period (all curves are data of the same time period), for example, one year, according to every 15 minutes as a collection point, group power curves. One year accumulates 365*24*4=35040 points. Real-time power is recorded as P ij , where i represents the i-th medium voltage curve, from 1 to M, and j from 1 to 35040.

[0049] The power of each power station is normalized, and the specific method is to calculate the average value of the maximum m points (for example, m is 1000) of each curve, and then divide the real-time power by the average value. Each normalized point is recorded as P ij ′, if the normalized value exceeds 1, it is calculated as 1. In this way, the normalized value becomes in the interval 【0, 1】.

[0050] The normalized data is analyzed by day, and 35040 points are divided into 365 arrays according to 96 points per day. For each day data, the average value of the maximum 4 data points is calculated. In this way, each curve corresponds to 365 average values. The maximum power of the i-th photovoltaic power station (i from 1 to N) on the d-th day (d from 0 to 365) is defined as P max,i,d . After this step, each substation obtains a 365-dimensional vector.

[0051] Principal Component Analysis (PCA) dimension reduction: Calculate the covariance matrix of all power features of power plants, extract the first k principal components (cumulative variance contribution rate ≥ 95%), and obtain the low-dimensional feature vector Zi (dimension k < 365).

[0052] In step 102, GMM-based probability clustering: The foregoing steps obtain the normalized single-day maximum photovoltaic power of N photovoltaic power plants; the single-day maximum photovoltaic power of photovoltaic power plants with the same characteristics should be consistent, and GMM-based probability clustering is used for clustering analysis to classify photovoltaic power plants with the same characteristics into one category. The specific operation is as follows:

[0053] Mathematical definition of Gaussian Mixture Model:

[0054] Assume that the data obeys the weighted sum of M Gaussian distributions, and the overall probability density is:

[0055]

[0056] where, is the model parameter; π m ≥ 0 is the mixing coefficient, satisfying N(·) is the Gaussian distribution, is the mean vector, is the covariance matrix (assuming it is a diagonal matrix to simplify calculation, or a full covariance matrix).

[0057] Bayesian Information Criterion (BIC) determines the optimal cluster number:

[0058] Iterate the candidate cluster number: preset M ∈ {2, 3,..., 10} (adjust according to the scale of regional power plants), and calculate for each M:

[0059]

[0060] where: is the maximum likelihood estimate of the model on the data; the penalty term M·ln N grows with the number of parameters M·(k+k(k+1) / 2+1)), avoiding overfitting.

[0061] Optimal cluster number selection: select the M that makes the BIC value minimum as the final cluster number. *

[0062] Expectation Maximization (EM) algorithm parameter estimation:

[0063] a. Initialization: use the K-means++ algorithm to initialize the mean

[0064] Randomly select the first mean

[0065] Each subsequent mean​ with probability Select, where D(Z i )) is the distance of the data point to the nearest selected mean, ensuring the initial mean distribution is uniform.

[0066] The covariance matrix is initialized as (σ 2 is the data variance), and the mixing coefficients are initialized as

[0067] b. Calculate the posterior probability:

[0068] For each data point Zi and cluster m, calculate the probability of belonging to that cluster:

[0069]

[0070] c. Update model parameters:

[0071] Mixing coefficients:

[0072]

[0073] Mean vector:

[0074]

[0075] Covariance matrix (diagonal):

[0076]

[0077] d. Iteration termination condition: when the rate of change of the log-likelihood function ∣lnL (t+1) - lnL (t) | <10 -5 or reaches the maximum number of iterations (such as 200 times).

[0078] Soft clustering assignment and hard cluster generation:

[0079] For each power station i, assign to the cluster with the maximum posterior probability:

[0080] c i = argmax m＝1,...,M γ(i, m)

[0081] Generate M clusters (C1, C2,..., CM), each containing a number of medium and low voltage power stations.

[0082] In step 103, intra-cluster weather data processing: based on step 102, obtain weather data for the M clusters, and obtain weather information according to the following principles.

[0083] If some photovoltaic power stations in the current cluster have meteorological data, the meteorological data of the photovoltaic power station with the maximum power generation is selected as the meteorological data of the new cluster; if none of the photovoltaic power stations in the current cluster has meteorological data, the photovoltaic power station with the maximum power generation is first found. Then, the meteorological data of the nearest stations around the photovoltaic power station is obtained, and the meteorological data of the cluster is obtained through interpolation.

[0084] In step 104, the prediction model is trained: for the cluster formed in step 102, the meteorological information obtained in step 103 is used to train the model by using the neural network BP algorithm, the cumulative value of the original power of each photovoltaic power station in the cluster, the wind speed, the wind direction, the speed, the temperature, the humidity, the air pressure and the light intensity are used as the characteristics of the model input, and the accumulated predicted power is used as the target output of the model. The model is repeatedly trained in an iterative manner, and the parameters are updated so that the mean square error of the predicted power and the actual power is minimized. Finally, M models are output for M clusters.

[0085] The M photovoltaic power station models connected to the medium voltage power grid are trained by using the BP algorithm: in order to simplify the complexity of the model, a single hidden layer BP neural network is designed, which consists of an input layer, a hidden layer and an output layer; the number of neurons in the input layer is consistent with the input characteristics of the data (such as wind speed, wind direction, temperature, humidity, air pressure and light intensity); since the final goal of the present scheme is to predict the power, which is a regression problem, the number of output neurons is 1.

[0086] The number of hidden layer neurons is generally selected by experience or experimental method, and the calculation formula is: wherein, l represents the number of hidden layer neurons; m is the number of input layer neurons; n is the number of output layer neurons; h is an adjustment constant, generally 1-10. At the same time, the Sigmoid function is selected as the activation function of the hidden layer and output layer neurons of the BP neural network.

[0087] Since the single hidden layer BP neural network is used, the cross-validation method is used to train and verify the model, so as to improve the learning ability of the model.

[0088] In step 105, the predicted power is predicted according to the weather forecast information obtained by the method in step 3.

[0089] It should be noted that the N photovoltaic power stations in the region are divided into M clusters by the DBSCAN method, without the need to specify the number of clusters in advance, simplifying the design; and the power prediction of the entire region is finally realized.

[0090] Probabilistic soft assignment of GMM clustering: Unlike the hard clustering (all or nothing) of DBSCAN, GMM outputs the probability of each power station belonging to each cluster (e.g. the probability of power station A belonging to cluster 1 is 60%, and cluster 2 is 40%), which provides a more reasonable grouping basis for power stations with ambiguous boundaries and reduces the risk of "misjudgment".

[0091] Engineering value of PCA dimension reduction: By retaining 95% of the principal components, the 365-dimensional features are compressed to 10-20 dimensions, reducing the number of GMM parameters from o(3652M) to o(k2M), and the calculation efficiency is improved by more than 90% (k=20, M=10 as an example, the number of parameters is reduced from about 130,000 to 4,000).

[0092] Cross-modal correlation of meteorological data: For clusters of power stations without medium voltage, meteorological data is generated by interpolation based on geographical proximity, establishing an indirect correlation between low-voltage power station power features, geographical space, and medium-voltage power station meteorological data, filling the data gap.

[0093] In addition to the algorithm used in this embodiment, common clustering algorithms include K-Means, DBSCAN, BIRCH, Affinity Propagation, and other methods. The comparison results of the above clustering algorithms and the method mentioned in this embodiment are shown in Table 1. The comparison results show that the algorithm of this embodiment has certain advantages.

[0094] Table 1 Comparison of errors of different clustering algorithms

[0095]

[0096] Embodiment 3 is an embodiment of the present application, which provides a distributed photovoltaic power prediction system based on a Gaussian mixture model, comprising:

[0097] A data acquisition module is configured to acquire historical power data of distributed photovoltaic power stations, perform data preprocessing, and divide the power data of each power station by day, and extract a time series vector representing the daily maximum power feature;

[0098] A feature extraction module is configured to perform principal component analysis on the time series vector to obtain a low-dimensional feature representation, and form a power feature vector of each photovoltaic power station;

[0099] A clustering analysis module is configured to perform clustering analysis on the power feature vector based on a probabilistic model to generate a plurality of photovoltaic power station clustering clusters;

[0100] A data input module is configured to obtain meteorological input data for each clustering cluster through a set data source priority rule and a completion mechanism;

[0101] The model training module is configured to construct a neural network power prediction model based on the accumulated power data and the corresponding meteorological features in the clustering cluster, and obtain parameter weights for prediction through training.

[0102] The power prediction module is configured to input weather forecast information into the prediction model, and output a future power generation power prediction value of the photovoltaic power station in the corresponding clustering cluster.

[0103] The embodiment also provides an electronic device suitable for the case of the distributed photovoltaic power prediction method based on the Gaussian mixture model, which comprises a memory and a processor.

[0104] The embodiment also provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the distributed photovoltaic power prediction method based on the Gaussian mixture model.

[0105] The storage medium provided in the embodiment and the distributed photovoltaic power prediction method based on the Gaussian mixture model provided in the above embodiment belong to the same inventive concept, and the technical details not described in detail in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0106] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk or an optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A distributed photovoltaic power prediction method based on a Gaussian mixture model, characterized by: include, Collect historical power data from distributed photovoltaic power stations, perform data preprocessing, divide the power data of each power station by day, and extract the time series vector representing the daily maximum power characteristics; Perform principal component analysis on the time series vector to obtain low-dimensional feature representation to form the power feature vector of each photovoltaic power station; Perform cluster analysis on the power feature vector based on a probability model to generate multiple photovoltaic power station clusters; For each cluster, meteorological input data is obtained through the set data source priority rules and completion mechanism; Based on the accumulated power data within the clusters and the corresponding meteorological characteristics, a neural network power prediction model is constructed, and the parameter weights used for prediction are obtained through training. Input weather forecast information into the prediction model and output the predicted value of future power generation of the photovoltaic power station in the corresponding cluster.

2. The distributed photovoltaic power prediction method based on a Gaussian mixture model according to claim 1, wherein: The extracting of the time series vector representing the daily maximum power feature includes normalizing the real-time power data of each power station, constructing an annual power curve based on the normalized data, and dividing the annual power curve into multiple data segments by day; For each day's data segment, the average value of several maximum power points is extracted as the representative characteristic value of the day to generate a time series vector representing the power characteristics of the whole year.

3. The distributed photovoltaic power prediction method based on a Gaussian mixture model according to claim 2, characterized in that: The principal component analysis includes calculating the covariance matrix of the time series vectors of all photovoltaic power stations, extracting the principal components by eigenvalue decomposition, and performing linear transformation on the original vectors according to the principal components to obtain a feature representation after dimensionality reduction.

4. The distributed photovoltaic power prediction method based on a Gaussian mixture model according to claim 3, characterized in that: The cluster analysis based on the probability model includes constructing a corresponding Gaussian mixture model for each candidate cluster number within a set range of candidate cluster numbers, calculating the model score based on the Bayesian Information Criterion, and selecting the cluster number with the lowest score as the optimal cluster number; Based on the selected number of clusters, the expectation maximization algorithm is used to iteratively estimate the model parameters until the preset convergence conditions are met and the model training is completed.

5. The distributed photovoltaic power prediction method based on a Gaussian mixture model according to claim 4, characterized in that: The obtaining of meteorological input data includes determining whether there is a photovoltaic power station with meteorological monitoring capability in the target cluster; If it exists, the meteorological data of the photovoltaic power station with the largest power generation capacity is selected as the meteorological input data of the cluster; If it does not exist, then based on the power station with the largest power generation in the cluster, find the external site with meteorological data that is geographically adjacent to it, and use the spatial interpolation method to estimate the meteorological input data of the cluster.

6. The distributed photovoltaic power prediction method based on a Gaussian mixture model according to claim 4, characterized in that: The constructing of the neural network power prediction model includes, for each cluster, collecting the accumulated historical power generation and corresponding meteorological characteristic data of the photovoltaic power station in the cluster to construct a training data set; A single hidden layer feedforward neural network model is established based on the training data set, wherein the model includes an input layer, a hidden layer, and an output layer, wherein the input features of the input layer include at least one meteorological element including wind speed, wind direction, temperature, humidity, air pressure, and light intensity; The output value of the output layer is the future power generation of the photovoltaic power station in the corresponding cluster; The neural network parameters are optimized by back-propagation algorithm to minimize the mean square error between the predicted output and the actual power.

7. The distributed photovoltaic power prediction method based on a Gaussian mixture model according to claim 4, characterized in that: Outputting the predicted value of future power generation of the photovoltaic power station in the corresponding cluster includes inputting weather forecast data consistent with the meteorological elements used in the model training phase into the neural network power prediction model; Outputting the prediction results of power generation in future time periods corresponding to the clusters through the model; Based on the prediction results and combined with the historical power distribution characteristics of each photovoltaic power station in the cluster, the predicted power is allocated or estimated among the power stations.

8. A distributed photovoltaic power prediction system based on a Gaussian mixture model, applying a distributed photovoltaic power prediction method based on a Gaussian mixture model as claimed in any one of claims 1 to 7, characterized in that: include: The data acquisition module is used to collect historical power data of distributed photovoltaic power stations, perform data preprocessing, divide the power data of each power station by day, and extract the time series vector representing the daily maximum power characteristics; The feature extraction module is used to perform principal component analysis on the time series vector to obtain low-dimensional feature representation and form the power feature vector of each photovoltaic power station; Cluster analysis module, used to perform cluster analysis on power feature vectors based on probability models to generate multiple photovoltaic power station clusters; The data input module is used to obtain meteorological input data for each cluster through the set data source priority rules and completion mechanism; The model training module is used to build a neural network power prediction model based on the accumulated power data in the cluster and the corresponding meteorological characteristics, and obtain the parameter weights used for prediction through training; The power prediction module is used to input weather forecast information into the prediction model and output the predicted value of future power generation of the photovoltaic power station in the corresponding cluster.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a distributed photovoltaic power prediction method based on a Gaussian mixture model according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a distributed photovoltaic power prediction method based on a Gaussian mixture model according to any one of claims 1 to 7 are implemented.

Images