CN114897129A - Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM - Google Patents

Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM Download PDF

Info

Publication number
CN114897129A
CN114897129A CN202210292305.9A CN202210292305A CN114897129A CN 114897129 A CN114897129 A CN 114897129A CN 202210292305 A CN202210292305 A CN 202210292305A CN 114897129 A CN114897129 A CN 114897129A
Authority
CN
China
Prior art keywords
data
power
multivariate
meteorological factor
lstm
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
Application number
CN202210292305.9A
Other languages
Chinese (zh)
Inventor
黄从智
张昕慧
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North China Electric Power University
Original Assignee
North China Electric Power University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by North China Electric Power University filed Critical North China Electric Power University
Priority to CN202210292305.9A priority Critical patent/CN114897129A/en
Publication of CN114897129A publication Critical patent/CN114897129A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/251—Fusion techniques of input or preprocessed data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H02J3/381—Dispersed generators
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Power Engineering (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a photovoltaic power station short-term power prediction method based on day-like clustering and Kmeans-GRA-LSTM. And then extracting a power characteristic value and clustering by using a K-means clustering algorithm. And normalizing the data and extracting the characteristic vector of the multivariate meteorological factor according to the multivariate meteorological factor corresponding to the power characteristic value clustering result. Selecting a multivariate meteorological factor feature vector of a predicted day, selecting power and multivariate meteorological factor data with higher association degree with the predicted day as similar day samples, and optimizing parameters of the LSTM by using an arithmetic optimization algorithm to construct an LSTM network topology structure. And predicting the power of a grid-connected point of a certain photovoltaic power station according to the constructed LSTM. The invention provides a scheme and a thought for solving the problem of short-term photovoltaic power prediction.

Description

Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM
Technical Field
The invention relates to the technical field of short-term photovoltaic power prediction in the technical field of photovoltaic power generation and grid connection, in particular to a photovoltaic power station short-term power prediction method based on day-like clustering and Kmeans-GRA-LSTM.
Background
The increasing shortage of world energy and environmental situation brings about the problems of outstanding power supply and demand contradiction, so that the development and utilization of the traditional energy are more limited, and among new energy, solar energy is considered as an ideal renewable energy power generation source. Photovoltaic power generation is an important solar energy utilization mode, but the output of a photovoltaic power station has the characteristics of high randomness, volatility, intermittency and the like, and the access of large-scale photovoltaic power generation can bring serious challenges to the safety and stable operation of a power system and the guarantee of electric energy quality. Therefore, the method has important significance for predicting the output power of the photovoltaic power station, adjusting the scheduling plan in time for the power department, improving the operation reliability of the power system and the access level of the photovoltaic power station and reducing the rotating reserve capacity of the system.
In the field of photovoltaic power prediction, the physical time layers spanned according to prediction can be divided according to ultra-short term, short term and long term and longer time. The research of the short-term photovoltaic power prediction algorithm is mainly started from the physical time spanned by prediction, the physical time is used as a basic measurement standard, the basic construction of a prediction model is completed, and the current short-term power prediction usually uses a neural network algorithm, a classification regression algorithm, a time series algorithm, a probability prediction algorithm and a comprehensive prediction algorithm.
Disclosure of Invention
The invention aims to provide a photovoltaic power station short-term power prediction method based on day-like clustering and Kmeans-GRA-LSTM, which can be used for more accurately predicting photovoltaic power generation power and reducing possible damage of photovoltaic grid connection to a power grid. The method comprises the steps of preprocessing acquired power data of a grid-connected point and multivariate meteorological factor data according to power data of the grid-connected point of a certain photovoltaic power station and the multivariate meteorological factor data of the location of the certain photovoltaic power station, extracting a characteristic value of the acquired power of the grid-connected point of the certain photovoltaic power station, clustering by using a K-means clustering algorithm, normalizing the data and extracting a characteristic vector of the multivariate meteorological factor according to the clustering result of the power characteristic value of the grid-connected point, selecting the multivariate meteorological factor characteristic vector of a predicted day, selecting the power data of the grid-connected point and the multivariate meteorological factor data of the previous 10 days with the highest correlation degree with the predicted day as similar day samples by using a gray correlation method, using the similar day samples as a training set of an LSTM neural network, optimizing parameters of the LSTM by using an arithmetic optimization algorithm, and constructing an LSTM network topology structure. And predicting the power of a grid-connected point of a certain photovoltaic power station according to the constructed LSTM, and verifying the accuracy of the method.
The invention adopts the technology that a photovoltaic power station short-term power prediction method based on day-like clustering and Kmeans-GRA-LSTM is implemented according to the following steps:
the method comprises the steps of obtaining grid-connected power data of a certain photovoltaic power station and multivariate meteorological factor data of the location of the certain photovoltaic power station.
The multi-element meteorological factor data comprise meteorological factors such as temperature, humidity, atmospheric pressure, wind speed, wind direction and radiation degree.
And preprocessing the acquired power data of the grid-connected point and the data of the multivariate meteorological factors.
The preprocessed data are typically missing data and anomalous data.
And (3) completing abnormal data and missing data by using a mean interpolation method, wherein the mean interpolation method comprises the following steps:
Figure BDA0003561990810000021
wherein x' i As a desired target for stabilization at time t. Wherein x is i-1 For the previous data point of the data to be processed, x i+1 And for the next data point of the data to be processed, carrying out difference compensation on the mean value of the two data points in pairs before and after the data to be processed.
And extracting a characteristic value from the obtained power data of the grid-connected point of a certain photovoltaic power station.
The characteristic values are selected to be the average daily power, the standard difference daily, the variation coefficient of daily power, the deviation state of daily power, the peak state of daily power and the total daily power. The specific formula is as follows:
Average daily power:
Figure BDA0003561990810000031
where N represents the number of data points related to the radiation level duration per day. P is i Representing the power of the point of connection at each point in time.
Day standard deviation:
Figure BDA0003561990810000032
daily power coefficient of variation:
Figure BDA0003561990810000033
daily power off-normal:
Figure BDA0003561990810000034
daily power kurtosis:
Figure BDA0003561990810000035
total daily power:
Figure BDA0003561990810000036
and after extracting the daily characteristic value, forming a power characteristic vector and carrying out normalization processing. The normalization formula is as follows:
Figure BDA0003561990810000037
wherein x min ,x max Respectively represent the minimum and maximum values of the sample, and y min ＝-1,y max ＝1。
And clustering the normalized power characteristic vectors by using a K-means clustering algorithm.
The specific flow of the K-means clustering algorithm is as follows:
and randomly selecting K samples from the sample set as cluster centers.
And calculating the distances between all samples and the K cluster centers, dividing each sample into the clusters with the cluster centers closest to the sample, and calculating the new cluster centers of all clusters for the new clusters.
Repeating the steps until the cluster center is not moving.
Because the power characteristic vectors are clustered by using K-means and the number of clustering centers is uncertain, the contour coefficient of a clustering evaluation algorithm is introduced, and the contour coefficient formula is as follows:
Figure BDA0003561990810000041
wherein, the average distance a (i) from the sample i to other samples in the same cluster is calculated, and the smaller the a (i), the more the sample i should be clustered to the cluster, and the intra-cluster dissimilarity referred to as the sample i in a (i) is calculated. Computing samples i to a cluster C j Is referred to as sample i and cluster C j Of (b) in which i ＝min{b i1 ,b i2 ,....,b ik ,}。
And corresponding to the multivariate meteorological factor eigenvector according to the power eigenvector clustering result of the grid-connected point.
And determining the characteristic vector of the multivariate meteorological factor to comprise the maximum value and the minimum value of wind speed, atmospheric pressure, temperature, humidity, wind direction and radiation degree and the radiation duration.
And the power feature vector clustering result of the grid-connected point corresponds to the feature vector of the multivariate meteorological factor.
And selecting the multivariate meteorological factor characteristic vector of the predicted day, and selecting the power data of the grid-connected point and the multivariate meteorological factor data of the previous 10 days with the highest correlation degree with the predicted day as similar day samples by using a grey correlation method to serve as a training set of the LSTM neural network.
And extracting the multivariate meteorological factor feature vector of the forecast day.
Determining meteorological factor characteristic vectors, and respectively extracting the maximum and minimum values of temperature, humidity, atmospheric pressure, wind speed, wind direction and radiation degree and radiation duration to form the meteorological factor characteristic vectors.
And determining the cluster to which the meteorological factor characteristic vector of the forecast day belongs according to Euclidean measurement between the meteorological factor characteristic vector of the forecast day and the meteorological factor characteristic vector of each cluster center. The calculation formula of the Euclidean distance is as follows:
Figure BDA0003561990810000051
Wherein x o A weather factor characteristic value representing a predicted day,
Figure BDA0003561990810000052
cluster centers for each cluster are represented by a meteorological factor characteristic value.
And adopting gray correlation analysis to select the geometric similarity between the characteristic vector of the weather factor in the forecast sun and the characteristic vector of the weather factor in the cluster to obtain the correlation between the characteristic vectors.
And selecting the characteristic vector of the weather factors in the predicted weather as a reference sequence, and selecting the characteristic vector of the weather factors in the cluster as a comparison sequence.
The reference sequence is represented as: y (k) k 1,2
The comparative sequences are shown as: x i ＝Y(k)|k＝1,2...,n；i＝1,2,...m
Carrying out non-dimensionalization processing on the reference sequence and the comparison sequence:
Figure BDA0003561990810000053
i＝1,2,...m
where k corresponds to a time period and i corresponds to a row in the comparison sequence.
Calculating a correlation coefficient after dimensionless processing:
Figure BDA0003561990810000054
where ρ represents a resolution coefficient, and usually ranges between (0,1), and when p ≦ 0.5463, the resolution is best, usually p ≦ 0.5.
Calculating the relevance:
Figure BDA0003561990810000055
and (4) sorting the relevance from large to small, and selecting the power data of the grid-connected point in the first 10 days with the highest similarity in the meteorological factor feature vectors in the cluster and the multivariate meteorological factor data as similar day samples.
And selecting the power data of the grid-connected points and the multi-element meteorological factor data with the highest correlation degree with the prediction day as a verification set in the similar days, and using the rest power data of the grid-connected points and the multi-element meteorological factor data as a training set, wherein the input data of the training neural network is the power data of the grid-connected points at the t-24 moment, the multi-element meteorological factor data at the t-24 moment and the multi-element meteorological factor data at the t moment, and the output data of the training neural network is the power data of the grid-connected points at the t moment. Selecting MSE of the verification set as a fitness function of an arithmetic optimization algorithm, wherein the MSE is as follows:
Figure BDA0003561990810000061
The LSTM neural network can transmit information before a long time to cells of a later time step, has larger memory capacity and stronger generalization capability and self-adaption capability, and can solve the problem of long-time dependence. The LSTM neural network structure is as follows:
Figure BDA0003561990810000062
wherein W f 、W i 、W c 、W o Weight matrices which are all LSTMs, b f 、b i 、b c 、b o Is the bias of LSTM, and sigma represents the sigmoid activation function.
Taking the number of neural network hidden layer units in the LSTM neural network, the iteration times and the learning rate as optimization objects of an arithmetic optimization algorithm, initializing a population and a related parameter r 1 ,r 2 ,r 3 And (4) parameters.
The algorithm accelerates the function through a mathematical optimizer (MOA) selection search phase when r 1 When the average value is more than MOA, the arithmetic optimization algorithm carries out global exploration when r is greater than MOA 1 If the current time is less than MOA, the arithmetic optimization algorithm enters a local development stage.
Figure BDA0003561990810000071
Min and Max respectively represent the maximum value and the minimum value of the position of the population, and T and T respectively represent the current iteration times and the maximum iteration times.
Wherein the global search is implemented by multiplication and division, when r 2 When the value is more than 0.5, executing a division search strategy, and when r is greater than 0.5, executing a division search strategy 2 When the number is less than 0.5, executing a multiplication search strategy, wherein the population updating formula is as follows:
Figure BDA0003561990810000072
wherein
Figure BDA0003561990810000073
Wherein r is 2 ∈[0,1]U is a control parameter for adjusting the search process, the value is 0.499, ξ is a minimum value, alpha is a sensitive parameter, the local development precision in the iterative process is defined, and the value is 5.
The local search is realized by addition and subtraction, and the population update formula is as follows:
Figure BDA0003561990810000074
wherein r is 3 Is a random number between 0 and 1.
And substituting the number of neural network hidden layer units, iteration times and learning rate in the LSTM neural network obtained by optimization into the LSTM neural network.
And inputting the multivariate meteorological factor data of the predicted day into the constructed LSTM neural network, and outputting the power of the grid-connected point of the predicted day.
Drawings
FIG. 1 is a flow chart of a photovoltaic power station short-term power prediction method based on day-like clustering and Kmeans-GRA-LSTM.
FIG. 2 is an LSTM neural network topology.
Fig. 3 is a flow chart of the arithmetic optimization algorithm.
Detailed description of the invention
The present invention is further described with reference to the accompanying drawings, but the scope of the present application is not limited thereto.
In this example, referring to fig. 1, the present invention provides a photovoltaic power plant short-term power prediction method based on similar-to-day clustering and Kmeans-GRA-LSTM, including the steps of:
acquiring grid-connected power data of a certain photovoltaic power station and multivariate meteorological factor data of the place where the certain photovoltaic power station is located.
The multivariate meteorological factor data comprises meteorological factors such as temperature, humidity, atmospheric pressure, wind speed, wind direction and radiation degree.
And preprocessing the acquired power data of the grid-connected point and the data of the multivariate meteorological factors.
The preprocessed data are typically missing data and anomalous data.
And (3) completing abnormal data and missing data by using a mean interpolation method, wherein the mean interpolation method comprises the following steps:
Figure BDA0003561990810000081
wherein x' i As a desired target for stabilization at time t. Wherein x is i-1 For the previous data point of the data to be processed, x i+1 And for the next data point of the data to be processed, carrying out difference compensation on the mean value of the two data points in pairs before and after the data to be processed.
And extracting a characteristic value from the obtained power data of the grid-connected point of a certain photovoltaic power station.
The characteristic values are selected to be the average daily power, the standard difference daily, the variation coefficient of daily power, the deviation state of daily power, the peak state of daily power and the total daily power. The specific formula is as follows:
average daily power:
Figure BDA0003561990810000091
where N represents the number of data points related to the radiation level duration per day. P i Representing the power of the point of connection at each point in time.
Day standard deviation:
Figure BDA0003561990810000092
daily power coefficient of variation:
Figure BDA0003561990810000093
daily power off-normal:
Figure BDA0003561990810000094
daily power kurtosis:
Figure BDA0003561990810000095
total daily power:
Figure BDA0003561990810000096
and extracting the daily characteristic value to form a power characteristic vector.
And the power characteristic vector is [ the daily average power daily standard deviation daily power variation coefficient daily power off-state daily power peak state daily total power ].
Carrying out normalization processing, wherein the normalization formula is as follows:
Figure BDA0003561990810000097
wherein x is min ,x max Respectively represent the minimum and maximum values of the sample, and y min ＝-1,y max ＝1。
And clustering the normalized power characteristic vectors by using a K-means clustering algorithm.
Setting the values of K as 2,3,4,5 and 6, sequentially carrying out K-means clustering and solving the contour coefficient.
The specific flow of the K-means clustering algorithm is as follows:
and setting the values of K to be 2,3,4,5 and 6 respectively.
And randomly selecting K samples from the sample set as cluster centers.
And calculating the distances between all samples and the K cluster centers, dividing each sample into the clusters with the cluster centers closest to the sample, and calculating the new cluster centers of all clusters for the new clusters.
Repeating the steps until the cluster center is not moving.
Because the power characteristic vectors are clustered by using K-means and the number of clustering centers is uncertain, the contour coefficient of a clustering evaluation algorithm is introduced, and the contour coefficient formula is as follows:
Figure BDA0003561990810000101
wherein, the average distance a (i) from the sample i to other samples in the same cluster is calculated, and the smaller the a (i), the more the sample i should be clustered to the cluster, and the intra-cluster dissimilarity referred to as the sample i in a (i) is calculated. Computing samples i to a cluster C j Is referred to as sample i and cluster C j Of (b) a degree of dissimilarity, wherein i ＝min{b i1 ,b i2 ,....,b ik ,}。
And selecting the K value corresponding to the maximum contour coefficient as the cluster center number of the K-means according to the solved contour coefficient.
And determining the maximum value and the minimum value of the characteristic vectors of the multivariate meteorological factors including wind speed, atmospheric pressure, temperature, humidity, wind direction and radiation degree and radiation duration according to the multivariate meteorological factor characteristic vectors corresponding to the power characteristic vector clustering result of the grid-connected point.
Characteristic vector of multivariate meteorological factor (wind speed maximum value, wind speed minimum value, atmospheric pressure maximum value, temperature maximum value, humidity maximum value, wind direction minimum value, radiation degree maximum value, radiation degree minimum value radiation duration)
And the power feature vector clustering result of the grid-connected point corresponds to the feature vector of the multivariate meteorological factor.
And selecting the multivariate meteorological factor characteristic vector of the predicted day, and selecting the power data of the grid-connected point and the multivariate meteorological factor data of the previous 10 days with the highest correlation degree with the predicted day as similar day samples by using a grey correlation method to serve as a training set of the LSTM neural network.
And extracting the multivariate meteorological factor feature vector of the forecast day.
Determining meteorological factor characteristic vectors, and respectively extracting the maximum and minimum values of temperature, humidity, atmospheric pressure, wind speed, wind direction and radiation degree and radiation duration to form the meteorological factor characteristic vectors.
And determining the cluster to which the meteorological factor characteristic vector of the forecast day belongs according to Euclidean measurement between the meteorological factor characteristic vector of the forecast day and the meteorological factor characteristic vector of each cluster center. The calculation formula of the Euclidean distance is as follows:
Figure BDA0003561990810000111
wherein x is o A weather factor characteristic value representing a predicted day,
Figure BDA0003561990810000112
cluster centers for each cluster are represented by a meteorological factor characteristic value.
And selecting the geometric similarity between the characteristic vector of the weather factor in the forecast sun and the characteristic vector of the weather factor in the cluster by adopting gray correlation analysis to obtain the correlation between the characteristic vectors.
And selecting the characteristic vector of the weather factors in the predicted weather as a reference sequence, and selecting the characteristic vector of the weather factors in the cluster as a comparison sequence.
The reference sequence is represented as: y (k) k 1,2
The comparative sequences are shown as: x i ＝Y(k)|k＝1,2...,n；i＝1,2,...m
Carrying out non-dimensionalization processing on the reference sequence and the comparison sequence:
Figure BDA0003561990810000113
i＝1,2,...m
where k corresponds to a time period and i corresponds to a row in the comparison sequence.
Calculating a correlation coefficient after dimensionless processing:
Figure BDA0003561990810000114
where ρ represents a resolution coefficient, and usually ranges between (0,1), and when p ≦ 0.5463, the resolution is best, usually p ≦ 0.5.
Calculating the relevance:
Figure BDA0003561990810000121
and (4) sorting the relevance from large to small, and selecting the power data of the grid-connected point in the first 10 days with the highest similarity in the meteorological factor feature vectors in the cluster and the multivariate meteorological factor data as similar day samples.
And selecting the power data of the grid-connected point and the multivariate meteorological factor data with the highest correlation degree with the predicted day as a verification set in the similar day, and using the rest power data of the grid-connected point and the multivariate meteorological factor data as a training set, wherein the input data of the training neural network is the power data of the grid-connected point at the time t-24, the multivariate meteorological factor data at the time t-24 and the multivariate meteorological factor data at the time t, and the output data of the training neural network is the power data of the grid-connected point at the time t. Selecting MSE of the verification set as a fitness function of an arithmetic optimization algorithm, wherein the MSE is as follows:
Figure BDA0003561990810000122
the LSTM neural network can transmit information before a long time to cells of a later time step, has larger memory capacity and stronger generalization capability and self-adaption capability, and can solve the problem of long-time dependence. The LSTM neural network structure is as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
Figure BDA0003561990810000123
Figure BDA0003561990810000124
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
wherein W f 、W i 、W c 、W o Weight matrices which are LSTM, b f 、b i 、b c 、b o Are both LSTM biases, and σ denotes the sigmoid activation function.
Taking the number of neural network hidden layer units in the LSTM neural network, the iteration times and the learning rate as optimization objects of an arithmetic optimization algorithm, initializing a population and a related parameter r 1 ,r 2 ,r 3 And (4) parameters.
The algorithm selects a search stage through a mathematical optimizer acceleration function (MOA) when r is 1 When the arithmetic optimization algorithm is more than MOA, the arithmetic optimization algorithm carries out global exploration when r is greater than MOA 1 If the current time is less than MOA, the arithmetic optimization algorithm enters a local development stage.
Figure BDA0003561990810000131
Min and Max respectively represent the maximum value and the minimum value of the position of the population, and T and T respectively represent the current iteration times and the maximum iteration times.
Wherein the global search is implemented by multiplication and division, when r 2 When the value is more than 0.5, executing a division search strategy, and when r is greater than 0.5, executing a division search strategy 2 When the number is less than 0.5, executing a multiplication search strategy, wherein the population updating formula is as follows:
Figure BDA0003561990810000132
wherein
Figure BDA0003561990810000133
Wherein r is 2 ∈[0,1]U is a control parameter for adjusting the search process, the value is 0.499, ξ is a minimum value, alpha is a sensitive parameter, the local development precision in the iterative process is defined, and the value is 5.
The local search is realized by addition and subtraction, and the population updating formula is as follows:
Figure BDA0003561990810000134
wherein r is 3 Is a random number between 0 and 1.
And substituting the number of neural network hidden layer units, iteration times and learning rate in the LSTM neural network obtained by optimization into the LSTM neural network.
And inputting the multivariate meteorological factor data of the predicted day into the constructed LSTM neural network, and outputting the power of the grid-connected point of the predicted day.
While the applicant has described particular embodiments of the invention in conjunction with the drawings herein, it will be understood by those skilled in the art that the foregoing description and description are only illustrative of the principles of the invention, and that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined by the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (8)

1. A photovoltaic power station short-term power prediction method based on day-like clustering and Kmeans-GRA-LSTM is characterized in that a photovoltaic power station short-term prediction model is established by using an LSTM network obtained through optimization of an intelligent optimization algorithm, and the method is implemented according to the following steps:
step 1: acquiring grid-connected power data of a certain photovoltaic power station and multivariate meteorological factor data of the place where the certain photovoltaic power station is located.
And 2, step: and preprocessing the acquired power data of the grid-connected point and the data of the multivariate meteorological factors.
And step 3: and extracting a characteristic value of the obtained power of the grid-connected point of a certain photovoltaic power station, and clustering by using a K-means clustering algorithm.
And 4, step 4: and normalizing the multivariate meteorological factor data and extracting the eigenvectors of the multivariate meteorological factor according to the clustering result of the power eigenvalue of the grid-connected point, which corresponds to the multivariate meteorological factor data.
And 5: and selecting the feature vector of the multivariate meteorological factor of the predicted day, and selecting the power of the grid-connected point and the multivariate meteorological factor data of the previous 10 days with the highest correlation degree with the predicted day as a similar day sample by using a grey correlation method to serve as a training set of the LSTM neural network.
Step 6: and optimizing the parameters of the LSTM by selecting an arithmetic optimization algorithm to construct an LSTM network topology structure.
And 7: and predicting the short-term grid-connected point power of a certain photovoltaic power station by using the finished LSTM neural network.
2. The method as claimed in claim 1, wherein the step 1 is specifically configured to obtain grid-connected point power data of a certain photovoltaic power station and multivariate meteorological factor data of a location of the certain photovoltaic power station within one year, wherein the multivariate meteorological factor data includes meteorological factors such as temperature, humidity, atmospheric pressure, wind speed, wind direction and radiance.
3. The photovoltaic power plant short-term power prediction method based on similar-day clustering and Kmeans-GRA-LSTM as claimed in claim 1, wherein said step 1 specifically comprises preprocessing the obtained grid-connected point power data and multivariate meteorological factor data. The specific process is as follows:
step 3.1, the preprocessed data are usually missing data and abnormal data.
And 3.2, screening data to make a specific judgment criterion according to the relevant criterion.
And 3.3, completing abnormal data and missing data by using a mean interpolation method.
Step 3.4, the mean interpolation method is as follows:
Figure FDA0003561990800000021
wherein x is i-1 For the previous data point, x, of the data to be processed i+1 And for the next data point of the data to be processed, carrying out difference compensation on the mean value of the two data points in pairs before and after the data to be processed.
4. The photovoltaic power plant short-term power prediction method based on day-like clustering and Kmeans-GRA-LSTM as claimed in claim 1, wherein said step 3 is specifically to extract a characteristic value from the obtained grid-connected point power data of a certain photovoltaic power plant and perform clustering by using a K-means clustering algorithm. The method comprises the following specific steps:
and 4.1, extracting a characteristic value from the acquired power data of the grid-connected point of a certain photovoltaic power station.
And 4.2, selecting characteristic values which are a daily power average value, a daily standard difference, a daily power variation coefficient, a daily power deviation, a daily power peak value and a daily total power respectively.
And 4.3, extracting the daily characteristic values to form power characteristic vectors, and performing normalization processing. The normalization formula is as follows:
Figure FDA0003561990800000022
wherein x min ,x max Respectively represent the minimum and maximum values of the sample, and y min ＝-1,y max ＝1。
And 4.4, clustering the normalized power characteristic vectors by using a K-means clustering algorithm.
5. The photovoltaic power plant short-term power prediction method based on similar-day clustering and Kmeans-GRA-LSTM as claimed in claim 1, wherein in said step 4, according to the multivariate meteorological factor eigenvector corresponding to the power eigenvector clustering result of the grid-connected point, the multivariate meteorological factor data is normalized and the eigenvector of the multivariate meteorological factor is extracted. The specific process is as follows:
And 5.1, determining the characteristic vectors of the multivariate meteorological factors, wherein the characteristic vectors comprise the maximum value and the minimum value of wind speed, atmospheric pressure, temperature, humidity, wind direction and radiation degree and radiation duration.
And 5.2, enabling the power feature vector clustering result of the grid-connected point to correspond to the multi-element meteorological factor feature vector.
6. The method for predicting the short-term power of the photovoltaic power station based on the similar daily clustering and the Kmeans-GRA-LSTM as claimed in claim 1, wherein the step 5 specifically comprises selecting a multivariate meteorological factor feature vector of a predicted day, and selecting the power data of the grid-connected point and the multivariate meteorological factor data of the previous 10 days with the highest correlation with the predicted day as a similar day sample by using a gray correlation method to serve as a training set of the LSTM neural network. The method comprises the following specific steps:
and 6.1, extracting the multivariate meteorological factor characteristic vector of the forecast day.
And 6.2, determining the meteorological factor characteristic vector, and respectively extracting the maximum and minimum values of the temperature, the humidity, the atmospheric pressure, the wind speed, the wind direction and the radiation degree and the radiation duration to form the meteorological factor characteristic vector.
And 6.3, determining the cluster to which the meteorological factor characteristic vector of the forecast day belongs according to Euclidean measurement between the meteorological factor characteristic vector of the forecast day and the meteorological factor characteristic vector of each cluster center.
And 6.4, selecting the geometric similarity between the characteristic vector of the weather factor in the forecast weather and the characteristic vector of the weather factor in the cluster by adopting gray correlation analysis to obtain the correlation between the characteristic vectors.
Step 6.5, the formula for calculating the correlation coefficient is as follows:
Figure FDA0003561990800000031
wherein y (k) and x i (k) Reference sequence and comparison sequence are indicated separately, where ρ represents resolution and ρ is 0.5.
6.6, calculating the association degree, wherein the association degree is defined as follows:
Figure FDA0003561990800000041
and 6.7, selecting the power data of the grid-connected point and the multivariate meteorological factor data of the previous 10 days with the highest similarity in the meteorological factor characteristic vectors in the cluster as similar day samples.
7. The photovoltaic power plant short-term power prediction method based on similar-day clustering and Kmeans-GRA-LSTM as claimed in claim 1, wherein in said step 6 specifically, an arithmetic optimization algorithm is selected to optimize LSTM parameters to construct LSTM network topology. The method comprises the following specific steps:
and 7.1, the LSTM neural network can transmit information before a long time to cells of a later time step, has larger memory capacity and stronger generalization capability and self-adaption capability, and can solve the problem of long-time dependence. The LSTM neural network structure is as follows:
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
Figure FDA0003561990800000042
Figure FDA0003561990800000043
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t *tanh(C t )
And 7.3, selecting the power data of the grid-connected point and the multivariate meteorological factor data with the highest correlation degree with the predicted day as a verification set in the similar day, and using the remaining power data of the grid-connected point and the multivariate meteorological factor data as a training set, wherein the input data of the training neural network is the power data of the grid-connected point at the time t-24, the multivariate meteorological factor data at the time t-24 and the multivariate meteorological factor data at the time t, and the output data of the training neural network is the power data of the grid-connected point at the time t. Selecting MSE of the verification set as a fitness function of an arithmetic optimization algorithm, wherein the MSE is as follows:
Figure FDA0003561990800000051
y represents the merged dot power of the authentication set,
Figure FDA0003561990800000052
representing the predicted validation set grid-connected point power.
Step 7.2, taking the number of the hidden layer units of the neural network in the LSTM neural network, the iteration times and the learning rate as the optimization objects of the arithmetic optimization algorithm, initializing the population and the related parameter r 1 ,r 2 ,r 3 And (4) parameters.
Step 7.3, selecting a search stage by an algorithm through a mathematical optimizer acceleration function (MOA) when r is 1 When the average value is more than MOA, the arithmetic optimization algorithm carries out global exploration when r is greater than MOA 1 If the current time is less than MOA, the arithmetic optimization algorithm enters a local development stage.
Figure FDA0003561990800000053
Step 7.4, realizing global search through multiplication and division operation, when r is 2 When the value is more than 0.5, a division search strategy is executed, and when r is greater than 0.5 2 When the number is less than 0.5, executing a multiplication search strategy, wherein the population updating formula is as follows:
Figure FDA0003561990800000054
wherein the MOP is as follows:
Figure FDA0003561990800000055
the local search is realized by utilizing addition operation and subtraction operation, and the population position updating formula is as follows:
Figure FDA0003561990800000056
and 7.5, substituting the number of the neural network hidden layer units in the LSTM neural network obtained by optimization, the iteration times and the learning rate into the LSTM neural network for training.
And 7.6, completing the construction of the LSTM neural network based on the prediction day.
8. The photovoltaic power station short-term power prediction method based on similar-day clustering and Kmeans-GRA-LSTM as claimed in claim 1, wherein in said step 7, the short-term power of the grid-connected point of a certain photovoltaic power station is predicted according to the constructed LSTM neural network. The specific process is as follows:
and 8.1, introducing the constructed LSTM neural network.
And 8.2, inputting the multivariate meteorological factor data of the predicted day into the constructed LSTM neural network, and outputting the power of the grid-connected point of the predicted day.
CN202210292305.9A 2022-03-24 2022-03-24 Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM Pending CN114897129A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210292305.9A CN114897129A (en) 2022-03-24 2022-03-24 Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210292305.9A CN114897129A (en) 2022-03-24 2022-03-24 Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM

Publications (1)

Publication Number Publication Date
CN114897129A true CN114897129A (en) 2022-08-12

Family

ID=82716122

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210292305.9A Pending CN114897129A (en) 2022-03-24 2022-03-24 Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM

Country Status (1)

Country Link
CN (1) CN114897129A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116307287A (en) * 2023-05-19 2023-06-23 国网信息通信产业集团有限公司 Prediction method, system and prediction terminal for effective period of photovoltaic power generation
CN117117859A (en) * 2023-10-20 2023-11-24 华能新能源股份有限公司山西分公司 Photovoltaic power generation power prediction method and system based on neural network
CN117250869A (en) * 2023-11-15 2023-12-19 成都态坦测试科技有限公司 Control method and device of temperature control equipment, readable storage medium and electronic equipment

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116307287A (en) * 2023-05-19 2023-06-23 国网信息通信产业集团有限公司 Prediction method, system and prediction terminal for effective period of photovoltaic power generation
CN116307287B (en) * 2023-05-19 2023-08-01 国网信息通信产业集团有限公司 Prediction method, system and prediction terminal for effective period of photovoltaic power generation
CN117117859A (en) * 2023-10-20 2023-11-24 华能新能源股份有限公司山西分公司 Photovoltaic power generation power prediction method and system based on neural network
CN117117859B (en) * 2023-10-20 2024-01-30 华能新能源股份有限公司山西分公司 Photovoltaic power generation power prediction method and system based on neural network
CN117250869A (en) * 2023-11-15 2023-12-19 成都态坦测试科技有限公司 Control method and device of temperature control equipment, readable storage medium and electronic equipment

Similar Documents

Publication Publication Date Title
CN107766990B (en) Method for predicting power generation power of photovoltaic power station
CN114897129A (en) Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM
JP5888640B2 (en) Photovoltaic power generation prediction apparatus, solar power generation prediction method, and solar power generation prediction program
CN114462718A (en) CNN-GRU wind power prediction method based on time sliding window
CN114792156A (en) Photovoltaic output power prediction method and system based on curve characteristic index clustering
CN116128150B (en) Photovoltaic intelligent output prediction method based on two-stage optimization
CN114757427A (en) Autoregressive corrected LSTM intelligent wind power plant ultra-short term power prediction method
CN111428766B (en) Power consumption mode classification method for high-dimensional mass measurement data
CN111222689A (en) LSTM load prediction method, medium, and electronic device based on multi-scale temporal features
CN114330100A (en) Short-term photovoltaic power probability interval prediction method
CN116345555A (en) CNN-ISCA-LSTM model-based short-term photovoltaic power generation power prediction method
CN115759389A (en) Day-ahead photovoltaic power prediction method based on weather type similar day combination strategy
CN115907195A (en) Photovoltaic power generation power prediction method, system, electronic device and medium
CN115912502A (en) Intelligent power grid operation optimization method and device
CN117318036A (en) Ultra-short-term photovoltaic power prediction method based on improved whale optimization algorithm
CN116187540B (en) Wind power station ultra-short-term power prediction method based on space-time deviation correction
CN116227677A (en) Power prediction correction method considering photovoltaic power climbing characteristics
CN115392387B (en) Low-voltage distributed photovoltaic power generation output prediction method
CN115481788B (en) Phase change energy storage system load prediction method and system
CN115496153A (en) Multi-head self-attention deep convolution embedded clustering wind-light-load combined scene method
CN114234392B (en) Air conditioner load fine prediction method based on improved PSO-LSTM
CN115456286A (en) Short-term photovoltaic power prediction method
CN113836814B (en) Solar energy prediction method based on multi-flow neural network
CN117117859B (en) Photovoltaic power generation power prediction method and system based on neural network
CN113297791B (en) Wind power combination prediction method based on improved DBN

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