CN115879602A - Ultra-short-term photovoltaic output prediction method based on transient weather - Google Patents
Ultra-short-term photovoltaic output prediction method based on transient weather Download PDFInfo
- Publication number
- CN115879602A CN115879602A CN202211429283.2A CN202211429283A CN115879602A CN 115879602 A CN115879602 A CN 115879602A CN 202211429283 A CN202211429283 A CN 202211429283A CN 115879602 A CN115879602 A CN 115879602A
- Authority
- CN
- China
- Prior art keywords
- data
- weather
- historical
- meteorological
- photovoltaic output
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to an ultra-short-term photovoltaic output prediction method based on transient weather, which comprises the following steps: acquiring historical sample data and preprocessing the historical sample data; clustering historical sample data based on fuzzy C-means clustering; performing feature dimension reduction on historical meteorological data based on a principal component analysis method to obtain meteorological data features; randomly determining a reference day sample under each weather type, and sequencing based on cosine distances between the reference day sample and similar day weather data characteristics; constructing a neural network model of the self-adaptive gating circulation unit; respectively taking historical photovoltaic output data of the adjacent sequenced similar day samples as input and output of the model, and training the model based on an uncertainty weighting method; and predicting the photovoltaic output based on the trained model. Compared with the prior art, the method fully considers the relation between the historical photovoltaic output and the meteorological information and the related information between similar day samples under the same weather type, and can effectively and accurately predict the ultra-short-term photovoltaic output.
Description
Technical Field
The invention relates to the technical field of photovoltaic output prediction, in particular to an ultra-short-term photovoltaic output prediction method based on transient weather.
Background
With the popularization of new energy welfare policies and the reduction of the cost of photovoltaic power generation equipment, and the significance of photovoltaic power prediction on power grid dispatching, the demand for photovoltaic output prediction is higher and higher. In order to improve the benefits of photovoltaic power generation, accurate photovoltaic output prediction becomes an increasingly important problem.
With the implementation of the "carbon peaking" and "carbon neutralization" targets, the installed capacity and the fraction of photovoltaics continues to increase. The photovoltaic power station comprises a centralized type and a distributed type, wherein the installed capacity ratio of the distributed photovoltaic is increased year by year and the development is rapid. Distributed photovoltaic is often installed on carriers such as administrative units, industrial and commercial roofs and residential roofs, compared with centralized photovoltaic, the distributed photovoltaic has the advantages of being environment-friendly, suitable in cost and capable of being used at any time, and residual electricity can be connected to the grid while spontaneous self-use is achieved. The accurate ultra-short-term photovoltaic power generation power prediction plays an important role in the optimized operation of a photovoltaic power station, the scheduling arrangement of a photovoltaic power system and the safe, stable and economic operation of a power grid.
At present, the main methods for realizing photovoltaic output prediction include 2 methods: the method is a statistical-based method and comprises a modeling technology represented by a time sequence; the other is a method based on an artificial intelligence algorithm, which comprises technologies such as a convolutional neural network and a long-short term neural network. The latter has good nonlinear processing capability and feature extraction capability, and is widely applied in the field of photovoltaic output prediction in recent years. However, the prior art still has the following defects:
(1) The method based on the artificial intelligence algorithm is easy to have the problems of over-fitting and under-fitting, the independent and same distribution of a test data set and a training data set needs to be met, the premise is difficult to be completely met due to a large amount of uncertainty of photovoltaic output, and the algorithm precision is low.
(2) The distributed photovoltaic installation sites are scattered, the installed capacity is small, the cost for installing the meteorological measurement device for each distributed photovoltaic system is high, the management is difficult, the time granularity of ultra-short-term load prediction is mostly 5min, 10min or 15min, auxiliary prediction is carried out extremely depending on meteorological data, if the cost is considered, the number of the installed meteorological measurement devices is small, and therefore sufficient input data are lacked to carry out auxiliary prediction on photovoltaic output, and the algorithm accuracy is poor.
Disclosure of Invention
The invention aims to provide an ultra-short-term photovoltaic output prediction method based on transient weather, which not only fully considers the relation between historical photovoltaic output and weather information, but also fully considers the related information between similar day data under the same weather type, and can effectively and accurately predict the ultra-short-term photovoltaic output.
The purpose of the invention can be realized by the following technical scheme:
an ultrashort-term photovoltaic output prediction method based on transient weather comprises the following steps:
s1: acquiring and preprocessing historical sample data, wherein the historical sample data comprises ultra-short-term historical photovoltaic output data and corresponding historical meteorological data;
s2: carrying out fuzzy C-means clustering on historical sample data by taking a statistical index of historical meteorological data as a clustering index, clustering the historical sample data to a plurality of weather types, wherein the sample data in the same weather type is a similar day sample;
s3: respectively performing feature dimensionality reduction on historical meteorological data under each weather type based on a principal component analysis method to obtain meteorological data features after dimensionality reduction;
s4: randomly determining a reference day sample under each weather type, and sequencing the similar day samples based on the cosine distance between the meteorological data characteristics of the similar day samples and the reference day samples;
s5: respectively constructing an adaptive gating circulation unit neural network model aiming at each weather type, wherein the adaptive gating circulation unit neural network model comprises a distribution identification module and an adaptive distribution matching module, the distribution identification module is used for identifying that the data distribution of similar day samples under the same weather type is different, and the adaptive distribution matching module is used for learning the related information among the similar day samples under the same weather type;
s6: respectively taking historical photovoltaic output data in the sequenced adjacent similar day samples as input and output of a neural network model of the adaptive gating circulation unit, balancing a prediction error and a related information error based on an uncertainty weighting method, and training the model;
s7: obtaining predicted weather data and determining a corresponding weather type, performing principal component analysis on the predicted weather data to obtain predicted weather data characteristics, calculating a predicted cosine distance between the predicted weather data characteristics and corresponding reference day sample weather data characteristics, determining a closest similar day sample based on the predicted cosine distance, taking historical photovoltaic output data of the closest similar day sample as model input, calling an adaptive gating cycle unit neural network model matched with the predicted weather type, and predicting to obtain the photovoltaic output of a predicted day.
The step S1 includes the steps of:
s11: acquiring historical sample data;
s12: checking and eliminating abnormal values in the historical sample data by using a 3 sigma criterion;
s13: interpolating a missing value in the historical sample data by using a Lagrange interpolation method;
s14: and carrying out normalization processing on the historical sample data.
The statistical indexes of the historical meteorological data comprise harmonic mean, geometric mean, variation coefficient, skewness and kurtosis.
The weather types are classified into 5 types, namely sunny days, cloudy days, rainy days and extreme weather.
The clustering loss function for fuzzy C-means clustering of historical sample data is as follows:
wherein J (U, V) is a clustering loss function; u. of ij Is the membership degree of the ith sample belonging to the jth class, and U is a membership degree matrix; n is the number of samples; v is the cluster center; m is a membership factor, and m is more than or equal to 1 and less than infinity; d ij Is the distance of the sample data to the cluster center.
The method for obtaining the meteorological data features by carrying out feature dimension reduction on the meteorological data based on a principal component analysis method specifically comprises the following steps: the method comprises the following steps of constructing meteorological data into a meteorological matrix, carrying out principal component analysis on the meteorological matrix to obtain the contribution rate of each principal component, and taking the principal component with the highest contribution rate as the extracted meteorological data characteristics, wherein the calculation method of the contribution rate of each principal component comprises the following steps:
wherein λ is i Is the eigenvector of the covariance matrix of the meteorological matrix, X is the meteorological matrix of Nxm, N is the number of samples, m is the initial characteristic dimension, e i Is an eigenvalue, r, of a covariance matrix of a meteorological matrix i Is the contribution rate of each principal component.
The self-adaptive gating cyclic unit neural network model consists of an input layer, two hidden layers, a full connection layer and an output layer.
The distribution identification module is constructed based on a maximum average difference method and specifically comprises the following steps:
wherein h is s And h t Two groups of different sample data, n s And n t The number of two groups of sample data is respectively, k is a kernel function, and i and j represent the ith and j data.
The adaptive distribution matching module is used for learning related information among similar day samples under the same weather type, and is specifically expressed in the form of a loss function:
wherein L (θ) represents a loss function of the adaptive distribution matching module, L pred (theta) is a prediction error loss function, L ada For the correlation information error loss function, λ is a balance term that balances the prediction error and the correlation information error, K is a parameter that avoids over-learning, D j,i Is the j, i distribution, and θ represents the model parameter.
The method for balancing the prediction error and the related information error based on the uncertainty weighting method specifically comprises the following steps:
wherein σ 1 Is L pred Standard deviation of medium output value, σ 2 Is L ada Standard deviation of the output value.
Compared with the prior art, the invention has the following beneficial effects:
(1) The method fully considers the relation between the historical photovoltaic output and the meteorological information and the related information between similar day data under the same weather type, and improves the accuracy of ultra-short-term photovoltaic output prediction.
(2) The invention clusters the samples by adopting a fuzzy C-means clustering method, and can overcome the premise that the required data meets normal distribution.
(3) According to the invention, through clustering and principal component dimension reduction analysis, redundant information of meteorological data is reduced, more deep utilization of the meteorological data is realized, effective prediction can be realized under the condition of insufficient meteorological data input, and thus the installation cost of the meteorological measuring device is also reduced.
(4) The adaptive distribution matching module of the UW-ADAGRU model can be used for mining relevant information among all similar day data to deal with possible future unseen meteorological information, further can be used for dealing with uncertainty of photovoltaic output, and realizes accurate prediction on the premise of lacking input data.
(5) The self-adaptive gated cyclic unit neural network model can well overcome the nonlinearity and the sequence of data, and the capability of avoiding over-fitting and under-fitting is stronger than that of the traditional artificial intelligence algorithm (BP, CNN and the like).
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
The embodiment provides a method for predicting ultra-short-term photovoltaic output based on transient weather, as shown in fig. 1, the method includes the following steps:
s1: and acquiring and preprocessing historical sample data, wherein the historical sample data comprises ultra-short-term historical photovoltaic output data and corresponding historical meteorological data.
S11: acquiring historical sample data;
s12: abnormal values in the historical sample data are checked and removed by using a 3 sigma criterion:
under the 3 σ principle, data does not need to follow normal distribution, and if the data abnormal value exceeds 3 times of standard deviation, it can be regarded as an abnormal value, and the probability of ± 3 σ is 98.9%, so the probability of occurrence of a value other than the average value 3 σ is P (| x-u | > 3 σ) =0.011, and belongs to a very individual small probability event. The method comprises the steps of supposing that historical meteorological data and historical photovoltaic output data which need to be detected only contain random errors, calculating and processing original data to obtain standard deviation, then determining a probability interval according to a certain probability, and considering that the error exceeding the probability interval belongs to an abnormal value. After the abnormal value is detected, the abnormal value is removed.
S13: and (3) interpolating missing values in the historical sample data by using a Lagrange interpolation method:
the original data has missing values, and the abnormal values are removed in step S12, so that the missing values are further increased, and the values at the missing positions need to be interpolated to obtain complete data; the Lagrangian interpolation comes from the number N +1 mutually different points x of a given function f (x) in numerical analysis i The corresponding function value is y i Its lagrange interpolation polynomial can be written in the form:
in the above formula, /) k (x) As a basis function, the expression is as follows:
s14: and carrying out normalization processing on the historical sample data.
S2: and carrying out fuzzy C-means clustering (FCM) on the historical sample data by taking the statistical index of the historical meteorological data as a clustering index, and clustering the historical sample data to a plurality of weather types.
S21: determining statistical indicators of historical meteorological data, comprising:
1) Harmonic mean
Where H is the harmonic mean, N is the number of samples, X i Is the ith variable.
2) Geometric mean
Wherein G is a geometric mean.
3) Coefficient of variation
Wherein cv is a coefficient of variation and σ is a standard deviation.
4) Deflection degree
Wherein s is skewness, μ 3 Is the central moment of order 3, and σ is the standard deviation.
5) Kurtosis
Wherein k is kurtosis.
S22: the weather types are classified into 5 categories: determining typical weather data under each weather type in sunny days, cloudy days, rainy days and extreme weather respectively, specifically: and marking Weather forecast information (NWP) as historical Weather day sample data of sunny days, cloudy days, rainy days and special extreme Weather, and extracting typical day Weather data by using an average value method to serve as a selection basis of a clustering center.
S23: and clustering the historical sample data to five weather types according to five statistical indexes of the historical meteorological data through fuzzy C-means clustering (FCM) and a determined clustering center, wherein the sample data under the same weather type are similar day samples.
Defining a clustering loss function for fuzzy C-means clustering on historical sample data as follows:
where J (U, V) is the clustering loss function, U ij Is the membership degree of the ith sample belonging to the jth class, and U is a membership degree matrix; v is the cluster center, i.e., the typical day data determined in S22; m is a membership factor, and m is more than or equal to 1 and less than infinity; d ij Is the distance of the sample data to the cluster center.
The strategy of FCM clustering is to continuously iterate and calculate U and V, so that the clustering loss function is minimum.
Specifically, the FCM in the ultra-short term photovoltaic prediction of the present invention comprises the following specific steps:
s231: initializing the number of clusters to be 5 weather types and membership degree matrix U (0) Let l denote the number of iterations.
S232: calculating clustering center V of the l iteration (l) ：
S233: updating membership degree matrix U (l) Computing a clustering loss function J (l) ：
Wherein the content of the first and second substances,
s234: given a membership termination threshold ε u Loss function termination threshold ε J The iteration is stopped when the threshold is reached, otherwise return to S232.
S3: and respectively performing feature dimension reduction on the historical meteorological data under each weather type based on a principal component analysis method to obtain the meteorological data features after dimension reduction.
Principal Component Analysis (PCA) is a commonly used and effective data dimension reduction method, and due to redundancy among various meteorological factors, excessive redundant information affects calculation efficiency and reduces model accuracy.
S31: determining the meteorological characteristic dimension to be analyzed, wherein the meteorological characteristic dimension comprises 7 dimensions of temperature, humidity, wind speed, oblique scattering, horizontal radiation and oblique scattering.
S32: the meteorological data are constructed into a meteorological matrix X of nxm, N is the number of samples, m is the initial characteristic dimension, and m =7 in this embodiment.
S33: calculate the average for each dimensional feature:
wherein the content of the first and second substances,
are averages.
S34: calculating a covariance matrix C:
s35: computing the eigenvectors e of C i And a characteristic value lambda i ,i＝1,2,…,m：
Ce i ＝λ i e i
Namely:
s36: determining the reduced matrix Z = XE, where E = [ E ] 1 ,e 2 ,…,e k ]And k is the dimensionality after dimensionality reduction.
S37: determining the size of k, namely the number of principal components in Z, and calculating the contribution rate of each principal component after characteristic reduction according to the following formula:
wherein e is i Is a feature vector, λ i Is a characteristic value, r i Is the contribution rate of each principal component.
In this embodiment, a principal component analysis method is used to perform comprehensive analysis on the 7 initial meteorological features to obtain the meteorological data features after dimension reduction. And (3) respectively taking different k values (from 1 to 7) to perform PCA analysis on the 7 main meteorological factors, and dividing the data into spring, summer, autumn and winter according to seasons. The calculated contribution rates of 7 groups of main components in different seasons and all the year are shown in table 1, and it can be seen that the contribution rates of the main components 1 in spring, summer and autumn reach more than 95% through dimensionality reduction of the initial meteorological features, wherein the contribution rates in summer and autumn can reach more than 97%. In the whole year, the principal component 1 can obtain a contribution rate higher than 96%, and most information in the original meteorological data is reserved, so that the principal component 1 is used as the meteorological data feature after dimension reduction.
TABLE 1 principal Components contribution ratio
S4: a reference day sample is randomly determined for each weather type, and similar day samples are ranked based on a cosine distance between weather data features of the similar day samples and the reference day samples.
The cosine distance calculation method comprises the following steps:
s5: an adaptive gated cyclic unit neural network model (UW-ADAGRU model) is constructed for each weather type and comprises a distribution identification module and an adaptive distribution matching module.
The self-adaptive gate control circulation unit neural network model consists of an input layer, two hidden layers, a full connection layer and an output layer.
The distribution identification module is constructed based on a Maximum Mean variance (MMD) method, and is used for identifying that data distribution of similar day samples under the same weather type is different, specifically:
wherein h is s And h t Two groups of different sample data, n s And n t The number of two groups of sample data is respectively, k is a kernel function, and i and j represent the ith and j data.
The Adaptive distribution matching module (Adaptive distribution matching module) is used for learning related information between similar day samples in the same weather type, so that the Adaptive distribution matching module functions in a loss function mode, and is specifically represented as:
wherein L (θ) represents a loss function of the adaptive distribution matching module; l is pred (θ) is a prediction error loss function; l is ada As a function of the error loss of the relevant information; lambda is a balance item for balancing the prediction error and the related information error, and the error is increased in order to avoid the phenomenon that the network excessively learns the shared knowledge; k is a parameter to avoid over-learning; d j,i Is the j, i distribution; θ represents the model parameters.
The former part of the L (theta) has the function of well reducing errors in the learning process of the similar day model, and the latter part of the L (theta) has the function of mining related information among similar day data samples so as to achieve self-adaption to photovoltaic output uncertainty caused by weather.
Prediction error loss function L pred (θ) is:
where x is the predicted value, y is the actual value, and M is a functional relationship relating the predicted value to the network parameter.
S6: and respectively taking historical photovoltaic output data in the sequenced adjacent similar day samples as input and output of the neural network model of the adaptive gating circulation unit, balancing prediction errors and related information errors based on an Uncertainty Weighting (UW), performing inverse normalization, and training the model.
The core idea of balancing the prediction error and the related information error based on the uncertainty weighting method is to combine L pred And L ada As two tasks, a proper balance value is obtained by searching uncertainty of the two tasks, specifically:
s61: the posterior distribution of the truth value of each task is assumed to be normal distribution taking a predicted value as a mean value, and the variance is noise and represents the difficulty degree of the task:
p(y|f θ (x))＝N(f θ (x),σ 2 )
wherein p is a likelihood function, f (X) is the output of the neural network, theta is the weight of the input X, sigma is the standard deviation, N is a normal distribution expression, and y is a given output value.
S62: determining a joint distribution of two tasks:
wherein, y 1 、y 2 Respectively, the output values of the two tasks.
S63: the optimization objective is converted into the maximum likelihood of finding the above-mentioned joint distribution, which is equivalent to minimizing the inverse thereof, thus obtaining the minimization objective as:
wherein σ 1 Is L pred Standard deviation of medium output value, σ 2 Is L ada Standard deviation of the output value.
S64: the former of historical photovoltaic output data in the similar day samples sequenced under a certain weather type is used as the input of the neural network model of the adaptive gating circulation unit corresponding to the weather type, and the latter is used as the output of the model, so that the model is trained.
S7: obtaining predicted weather data and determining a corresponding weather type, performing principal component analysis on the predicted weather data to obtain predicted weather data characteristics, calculating a predicted cosine distance between the predicted weather data characteristics and corresponding reference day sample weather data characteristics, determining a closest similar day sample based on the predicted cosine distance, taking historical photovoltaic output data of the closest similar day sample as model input, calling an adaptive gating cycle unit neural network model matched with the predicted weather type, and predicting to obtain the photovoltaic output of a predicted day.
Table 2 shows the comparison of the prediction results of the method for predicting the ultra-short term photovoltaic output from the perspective of the commonly used MAE, MAPE and RMSE evaluation indexes, wherein ARMA is an autoregressive moving average model, ARIMA is a difference integration moving average autoregressive model, SVM is a support vector machine model, CNN is a convolutional neural network, LSTM is a long-short term memory neural network, GRU is a common gated cyclic unit neural network, and FCM-UW-ADAGRU is the method of the invention.
TABLE 2 estimation of prediction results
| Method | MAE | MAPE | RMSE |
| ARMA | 17.34 | 12.03％ | 23.04 |
| ARIMA | 16.97 | 11.59％ | 20.58 |
| SVM | 13.28 | 8.45％ | 17.52 |
| CNN | 12.24 | 7.42％ | 18.34 |
| LSTM | 10.51 | 7.01％ | 15.21 |
| GRU | 10.33 | 6.45％ | 13.23 |
| FCM-UW-ADAGRU | 8.52 | 4.03％ | 10.01 |
According to the results shown in table 2, the method of the invention has smaller MAE, MAPE and RMSE, and the effectiveness and the prediction accuracy of the method are proved.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions that can be obtained by a person skilled in the art through logic analysis, reasoning or limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (10)
1. An ultrashort-term photovoltaic output prediction method based on transient weather is characterized by comprising the following steps:
s1: acquiring and preprocessing historical sample data, wherein the historical sample data comprises ultra-short-term historical photovoltaic output data and corresponding historical meteorological data;
s2: carrying out fuzzy C-means clustering on historical sample data by taking a statistical index of historical meteorological data as a clustering index, clustering the historical sample data to a plurality of weather types, wherein the sample data in the same weather type is a similar day sample;
s3: respectively performing feature dimensionality reduction on historical meteorological data under each weather type based on a principal component analysis method to obtain meteorological data features after dimensionality reduction;
s4: randomly determining a reference day sample under each weather type, and sequencing the similar day samples based on the cosine distance between the meteorological data characteristics of the similar day samples and the reference day samples;
s5: respectively constructing an adaptive gating circulation unit neural network model aiming at each weather type, wherein the adaptive gating circulation unit neural network model comprises a distribution identification module and an adaptive distribution matching module, the distribution identification module is used for identifying that the data distribution of similar day samples under the same weather type is different, and the adaptive distribution matching module is used for learning the related information among the similar day samples under the same weather type;
s6: respectively taking historical photovoltaic output data in the sequenced adjacent similar day samples as input and output of a neural network model of the self-adaptive gating circulation unit, balancing a prediction error and a related information error based on an uncertainty weighting method, and training the model;
s7: obtaining predicted weather data and determining a corresponding weather type, performing principal component analysis on the predicted weather data to obtain predicted weather data characteristics, calculating a predicted cosine distance between the predicted weather data characteristics and corresponding reference day sample weather data characteristics, determining a closest similar day sample based on the predicted cosine distance, taking historical photovoltaic output data of the closest similar day sample as model input, calling an adaptive gating cycle unit neural network model matched with the predicted weather type, and predicting to obtain the photovoltaic output of a predicted day.
2. The method of claim 1, wherein the step S1 comprises the steps of:
s11: acquiring historical sample data;
s12: checking and eliminating abnormal values in the historical sample data by using a 3 sigma criterion;
s13: interpolating a missing value in the historical sample data by using a Lagrange interpolation method;
s14: and carrying out normalization processing on the historical sample data.
3. The method of claim 1, wherein the statistical indicators of the historical meteorological data comprise harmonic means, geometric means, coefficient of variation, skewness, and kurtosis.
4. The method of claim 1, wherein the weather types are classified into 5 types, namely sunny weather, cloudy weather, rainy weather, and extreme weather.
5. The method of claim 1, wherein the clustering loss function for fuzzy C-means clustering of historical sample data is as follows:
wherein J (U, V) is a clustering loss function; u. of ij Is the membership degree of the ith sample belonging to the jth class, and U is a membership degree matrix; n is the number of samples; v is the cluster center; m is a membership factor, and m is more than or equal to 1 and less than infinity; d ij Is the distance of the sample data to the cluster center.
6. The ultra-short-term photovoltaic output prediction method based on transient weather is characterized in that the feature dimension reduction is performed on the weather data based on the principal component analysis method to obtain the weather data features specifically as follows: the method comprises the following steps of constructing meteorological data into a meteorological matrix, carrying out principal component analysis on the meteorological matrix to obtain the contribution rate of each principal component, and taking the principal component with the highest contribution rate as the extracted meteorological data characteristics, wherein the calculation method of the contribution rate of each principal component comprises the following steps:
wherein λ is i Is the eigenvector of the covariance matrix of the meteorological matrix, X is the meteorological matrix of Nxm, N is the number of samples, m is the initial characteristic dimension, e i Is an eigenvalue, r, of a covariance matrix of a meteorological matrix i Is the contribution rate of each principal component.
7. The method of claim 1, wherein the adaptive gated cyclic unit neural network model comprises an input layer, two hidden layers, a fully connected layer and an output layer.
8. The ultrashort-term photovoltaic output prediction method based on transient weather, according to claim 1, wherein the distribution identification module is constructed based on a maximum average difference method, specifically:
wherein h is s And h t Two groups of different sample data, n s And n t The numbers of two groups of sample data are respectively, k is a kernel function, and i and j represent the ith and j data.
9. The method of claim 1, wherein the adaptive distribution matching module is configured to learn the correlation information between similar day samples in the same weather type, specifically expressed as a loss function:
wherein L (θ) represents a loss function of the adaptive distribution matching module, L pred (θ) is a prediction error loss function, L ada For the correlation information error loss function, λ is a balance term that balances the prediction error and the correlation information error, K is a parameter that avoids over-learning, D j,i Is the j, i distribution, and theta represents the model parameter.
10. The ultra-short-term photovoltaic output prediction method based on transient weather, as claimed in claim 9, wherein the uncertainty weighting based method for balancing the prediction error and the related information error is specifically:
wherein σ 1 Is L pred Standard deviation of medium output value, σ 2 Is L ada Standard deviation of the output value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211429283.2A CN115879602A (en) | 2022-11-15 | 2022-11-15 | Ultra-short-term photovoltaic output prediction method based on transient weather |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211429283.2A CN115879602A (en) | 2022-11-15 | 2022-11-15 | Ultra-short-term photovoltaic output prediction method based on transient weather |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115879602A true CN115879602A (en) | 2023-03-31 |
Family
ID=85759984
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211429283.2A Pending CN115879602A (en) | 2022-11-15 | 2022-11-15 | Ultra-short-term photovoltaic output prediction method based on transient weather |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115879602A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117350549A (en) * | 2023-12-04 | 2024-01-05 | 国网江苏省电力有限公司经济技术研究院 | Distribution network voltage risk identification method, device and equipment considering output correlation |
| CN117993739A (en) * | 2024-04-03 | 2024-05-07 | 国网山西省电力公司营销服务中心 | Photovoltaic output ultra-short-term prediction method based on data denoising and point compensation correction |
-
2022
- 2022-11-15 CN CN202211429283.2A patent/CN115879602A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117350549A (en) * | 2023-12-04 | 2024-01-05 | 国网江苏省电力有限公司经济技术研究院 | Distribution network voltage risk identification method, device and equipment considering output correlation |
| CN117350549B (en) * | 2023-12-04 | 2024-02-23 | 国网江苏省电力有限公司经济技术研究院 | Distribution network voltage risk identification method, device and equipment considering output correlation |
| CN117993739A (en) * | 2024-04-03 | 2024-05-07 | 国网山西省电力公司营销服务中心 | Photovoltaic output ultra-short-term prediction method based on data denoising and point compensation correction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108564192B (en) | Short-term photovoltaic power prediction method based on meteorological factor weight similarity day | |
| CN115879602A (en) | Ultra-short-term photovoltaic output prediction method based on transient weather | |
| CN108921339B (en) | Quantile regression-based photovoltaic power interval prediction method for genetic support vector machine | |
| Lu et al. | Feature extraction of meteorological factors for wind power prediction based on variable weight combined method | |
| CN110909912A (en) | Park electric power system net load combination prediction method based on self-adaptive error feedback | |
| CN111754037B (en) | Long-term load hybrid prediction method for regional terminal integrated energy supply system | |
| CN112418496B (en) | Power distribution station energy storage configuration method based on deep learning | |
| CN109325631A (en) | Electric car charging load forecasting method and system based on data mining | |
| CN113255900A (en) | Impulse load prediction method considering improved spectral clustering and Bi-LSTM neural network | |
| CN112149890A (en) | Comprehensive energy load prediction method and system based on user energy label | |
| CN112232561A (en) | Power load probability prediction method based on constrained parallel LSTM quantile regression | |
| CN110210670A (en) | A kind of prediction technique based on power-system short-term load | |
| CN113361785A (en) | Power distribution network short-term load prediction method and device, terminal and storage medium | |
| CN113516271A (en) | Wind power cluster power day-ahead prediction method based on space-time neural network | |
| CN115099511A (en) | Photovoltaic power probability estimation method and system based on optimized copula | |
| Zeng et al. | Short-term load forecasting of smart grid systems by combination of general regression neural network and least squares-support vector machine algorithm optimized by harmony search algorithm method | |
| CN112288157A (en) | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning | |
| CN115660182A (en) | Photovoltaic output prediction method based on maximum expected sample weighted neural network model | |
| CN115204698A (en) | Real-time analysis method for power supply stability of low-voltage transformer area | |
| CN115358437A (en) | Power supply load prediction method based on convolutional neural network | |
| CN111815039A (en) | Weekly scale wind power probability prediction method and system based on weather classification | |
| CN108694475B (en) | Short-time-scale photovoltaic cell power generation capacity prediction method based on hybrid model | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| Xu et al. | Short-term electricity consumption forecasting method for residential users based on cluster classification and backpropagation neural network | |
| CN116826710A (en) | Peak clipping strategy recommendation method and device based on load prediction and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |