CN114492941A - Whole-county photovoltaic prediction method based on cluster division and data enhancement - Google Patents

Whole-county photovoltaic prediction method based on cluster division and data enhancement Download PDF

Info

Publication number
CN114492941A
CN114492941A CN202111646119.2A CN202111646119A CN114492941A CN 114492941 A CN114492941 A CN 114492941A CN 202111646119 A CN202111646119 A CN 202111646119A CN 114492941 A CN114492941 A CN 114492941A
Authority
CN
China
Prior art keywords
photovoltaic
data
county
cluster
prediction method
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
CN202111646119.2A
Other languages
Chinese (zh)
Inventor
赵家庆
陈中
赵奇
王彪
吕洋
徐春雷
庄卫金
闪鑫
王毅
吴海伟
丁宏恩
田江
俞瑜
徐秀之
孟雨庭
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.)
Southeast University
China Electric Power Research Institute Co Ltd CEPRI
State Grid Jiangsu Electric Power Co Ltd
State Grid Electric Power Research Institute
Suzhou Power Supply Co of State Grid Jiangsu Electric Power Co Ltd
Original Assignee
Southeast University
China Electric Power Research Institute Co Ltd CEPRI
State Grid Jiangsu Electric Power Co Ltd
State Grid Electric Power Research Institute
Suzhou Power Supply Co of State Grid Jiangsu Electric Power Co Ltd
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 Southeast University, China Electric Power Research Institute Co Ltd CEPRI, State Grid Jiangsu Electric Power Co Ltd, State Grid Electric Power Research Institute, Suzhou Power Supply Co of State Grid Jiangsu Electric Power Co Ltd filed Critical Southeast University
Priority to CN202111646119.2A priority Critical patent/CN114492941A/en
Publication of CN114492941A publication Critical patent/CN114492941A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • 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)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Operations Research (AREA)
  • Development Economics (AREA)
  • Biomedical Technology (AREA)
  • Game Theory and Decision Science (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A whole-county distributed photovoltaic prediction method based on cluster division and data enhancement specifically comprises the following steps: selecting a typical power curve on a sunny day from a photovoltaic output historical database of the whole county, and performing per unit on the output by using the maximum power of a single station; calculating a Pearson correlation coefficient as a distance measurement, Clustering photovoltaic stations by using a Density-Based Spatial Clustering of Applications with Noise, DBSCAN (direct-base Clustering of Applications with Noise) to form cluster division, and for abnormal points, dividing the abnormal points into a nearest cluster by using k nearest neighbor search; performing data expansion on historical data through a generative antagonistic neural network in the cluster; and (4) graphing the original data and the generated data to jointly train a deep convolutional network prediction model. According to the prediction method, the original data distribution is learned through the improved dynamic game process training process of the GAN, then the data distributed correspondingly is generated, the whole county photovoltaic historical database is supplemented, the deep convolutional neural network is trained through the enhanced training set, and the model prediction accuracy is improved.

Description

Whole-county photovoltaic prediction method based on cluster division and data enhancement
Technical Field
The invention belongs to the field of photovoltaic power generation power prediction, and particularly relates to a whole county photovoltaic prediction method based on cluster division and data enhancement.
Background
The climate change is a global problem facing human beings, with the rapid development of productivity, the amount of carbon dioxide emitted by each country is increased rapidly, the greenhouse effect is serious day by day, and in order to cope with climate warming, the carbon emission reduction target is given by each country in the world in a global contractual manner. The photovoltaic power generation does not need the combustion of fossil fuel for energy conversion, and is a green clean energy. Centralized photovoltaic needs to occupy a large amount of land resources, most of the centralized photovoltaic needs to be concentrated in northwest areas with extensive and rare land, and electric power resources need to be transmitted to a load center through an extra-high voltage transmission line. The roof distributed photovoltaic does not occupy special land resources, and is close to a load center, so that the power generation potential is huge. The permeability of the photovoltaic is continuously improved, and the randomness and the fluctuation of the output of the photovoltaic cause pressure on the safe and stable operation of a power grid, so that the accuracy of the photovoltaic power generation power prediction in the whole county is more and more important.
The distributed photovoltaic system in the whole county mainly has two problems at present, namely, the distributed photovoltaic system starts late compared with a centralized photovoltaic system, historical meteorological data and power generation data are deficient, and a high-precision data driving model is difficult to train; secondly, distributed photovoltaic is numerous, the geographical distribution is wide, and the one-station one-prediction similar to a centralized station is difficult to perform. Therefore, the problem of model training of poor data needs to be solved, and the problem of whole county photovoltaic power prediction of massive sites needs to be processed at the same time.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide a whole county photovoltaic prediction method based on cluster division and data enhancement.
The invention adopts the following technical scheme:
a whole county photovoltaic prediction method based on cluster division and data enhancement comprises the following steps:
s1, collecting historical sunny photovoltaic output data and meteorological data of each photovoltaic output station in a set region and within a set time;
s2, obtaining the maximum power of each photovoltaic output station according to the collected sunny photovoltaic output data of S1, performing per unit, and then performing photovoltaic cluster division;
s3, constructing a photovoltaic data enhancement neural network model;
s4, performing data enhancement in the photovoltaic cluster through the photovoltaic data enhancement neural network model constructed in the S3, inputting the enhanced data into the CNN neural network, and training to obtain a photovoltaic prediction neural network model;
s5: and inputting the weather forecast data into the photovoltaic prediction neural network model of S4 for photovoltaic prediction.
In S1, the photovoltaic output data is photovoltaic output data of the whole county on historical sunny days;
meteorological data includes month, day, time, minutes, direct irradiance, diffuse horizontal irradiance, total horizontal irradiance, ambient temperature, barometric pressure, relative humidity, wind direction, wind speed, ground surface reflectance, and generated power.
In S2, the historical sunny photovoltaic output curve is a per unit output curve, the maximum power of a single station is calculated from historical output data, and then a per unit value of the curve is calculated;
the formula for data per unit is as follows:
Figure BDA0003443913200000021
in the above formula, x is the original data of the sample, xmaxAnd z is the per unit data.
In S2, the photovoltaic clusters are divided by using a DBSCAN method, the distance measurement of the DBSCAN uses pearson correlation distance, the DBSCAN performs cluster analysis to generate abnormal sites, k-nearest neighbor algorithm is used to search the nearest k sites by using the pearson correlation distance as the measurement, the cluster numbers of the k sites are counted, and the abnormal sites are divided into the clusters with the largest number of numbers.
At S3, the alternative neural network model is GAN.
The optional neural network model can also be an improved GAN, and the specific algorithm is as follows:
s3.1, setting a learning rate alpha, a truncation parameter c, a batch training sample number m and a number of times n of one discriminator iteration of each iteration of a generatorcriticInitializing arbiter network parameter wtGenerator network parameter θt；
S3.2, checking the Generator parameter θtWhether convergence is carried out or not, if so, the iteration is ended; if not, entering S3.3;
s3.3, checking whether the current iteration number reaches an iteration number threshold ncriticIf not, updating the network parameter w of the discriminatortUpdating the current iteration times, and repeating the step; if a threshold value n of the number of iterations is reachedcriticStep S3.4 is entered;
step 3.4, calculate Generator loss function
Figure BDA0003443913200000031
And updates the generator parameter thetatAnd returning to the step 3.2;
updating discriminator network parameter wtThe method comprises the following steps:
Figure BDA0003443913200000032
wherein clip (.) represents a clipping function in the deep learning, RMSProp (.) represents a learning rate adaptive optimizer in the deep learning,
Figure BDA0003443913200000033
represents the discriminator lossThe specific calculation method of the gradient of the loss function is as follows:
Figure BDA0003443913200000034
wherein x is(i)Is the ith training batch in the input parameters; z is a radical of(i)Sampling a batch in a sample distribution generated from a generator for an ith training;
Figure BDA0003443913200000035
is a network of discriminators.
The gradient of the generator loss function is:
Figure BDA0003443913200000036
wherein,
Figure BDA0003443913200000037
as a network of discriminators, z(i)The batch in the sample distribution produced from the generator is sampled for the ith training.
Updating the generator parameter θtThe method comprises the following steps:
Figure BDA0003443913200000038
RMSProp (.) denotes a learning rate adaptive optimizer in deep learning.
The data enhancement of S4 means that the data collected in S1 is firstly picturized, the original data of one day is divided into 48 points every day, and a picture with the size of 48 × 15 pixels is formed; and then, performing countermeasure learning through the photovoltaic data enhancement neural network model to generate new data with the same distribution.
In S4, the CNN neural network has a 5-layer structure, each of the first four layers includes a convolutional layer, a batch normalization layer, and a linear rectification layer, and the last layer includes a fully-connected layer, a batch normalization layer, and a ReLU layer.
The invention has the advantages that compared with the prior art,
1. according to the prediction method, the cluster division is performed on the whole county distributed photovoltaic, the distributed sites with strong correlation are divided into one cluster, and the overall output prediction is performed on the photovoltaic cluster, so that the distribution of numerical weather forecast can be reduced, the calculation pressure is effectively reduced, and the overall prediction precision is improved;
2. the prediction method provided by the invention generates new similar data by countertraining and learning the interrelation and distribution characteristics of the original data, supplements a poor training data set and improves the model prediction accuracy.
Drawings
FIG. 1 is a flow chart of a prediction method of the present invention;
FIG. 2 is a DBSCAN clustering histogram of the present invention;
FIG. 3 is a timing diagram of the DBSCAN clustering results of the present invention;
FIG. 4 is the results of the k-nearest neighbor search of the present invention;
FIG. 5 is an imaging of raw data;
FIG. 6 is data and training process for improved GAN generation;
figure 7 is a comparison graph of CNN and improved GAN-CNN predicted timing.
Detailed Description
The present application is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present application is not limited thereby.
A county photovoltaic prediction method based on cluster division and data enhancement is disclosed, wherein the prediction method flow is shown in FIG. 1, and the method specifically comprises the following steps:
s1: collecting historical sunny photovoltaic output data and meteorological data of each photovoltaic output station in a set area and within a set time;
in the invention, the collected photovoltaic output data is photovoltaic output data in the whole county historical sunny days;
environmental data includes month, day, time, minutes, direct irradiance, diffuse horizontal irradiance, total horizontal irradiance, ambient temperature, air pressure, relative humidity, wind direction, wind speed, ground surface reflectance, and generated power.
S2: obtaining the maximum power of each photovoltaic output station according to the collected sunny photovoltaic output data of S1, performing per unit, and then performing cluster division;
further, the historical sunny photovoltaic output curve in S2 is a per unit output curve, the maximum power per station is calculated from the historical output data, and then a per unit value of the curve is calculated;
the formula for data per unit is as follows:
Figure BDA0003443913200000051
in the above formula, x is the original data of the sample, xmaxAnd z is the per unit data.
According to a typical historical sunny photovoltaic output curve of the station, performing clustering division on the whole county photovoltaic by adopting DBSCAN to form a highly-correlated photovoltaic cluster;
a person skilled in the art can select a method for cluster division according to actual situations, and the method selected by the present invention is only a preferred embodiment and is not necessarily limited by the present invention.
Further, in S2, a DBSCAN is used for cluster division, and a pearson correlation distance is used for distance measurement of the DBSCAN;
the formula for Pearson correlation distance is as follows:
Figure BDA0003443913200000052
wherein,
Figure BDA0003443913200000053
xs，xtfor randomising the power sequences of two stations, xsjDenotes xsJ-th data value of (1), xtiRepresents xtN represents xsjTotal number of data in (1), m represents xtiTotal number of data in (1);
further, for abnormal sites generated by clustering analysis of the DBSCAN, using pearson correlation distance as a measure, searching the nearest k sites by using a k-nearest neighbor algorithm, counting cluster numbers of the k sites, and dividing the abnormal sites into clusters with the largest number of numbers.
S3: building a photovoltaic data enhancement neural network model;
the person skilled in the art can select the neural network according to the actual situation, and the method proposed by the present invention is only a preferred embodiment and is not necessarily limited by the present invention.
Preferably, the selectable neural network model is GAN;
in this embodiment, the photovoltaic data enhanced neural network model is an improved GAN, and the algorithm of the improved GAN specifically includes:
s3.1, setting a learning rate alpha, a truncation parameter c, a batch training sample number m and a number of times n of one discriminator iteration of each iteration of a generatorcriticInitializing arbiter network parameter wtSum generator network parameter θt，
S3.2, checking the Generator parameter θtWhether convergence is carried out or not, if so, the iteration is ended; if not, entering S3.3;
s3.3, checking whether the current iteration number reaches an iteration number threshold ncriticIf not, updating the network parameter w of the discriminatortUpdating the current iteration times, and repeating the step; if the threshold value n of the iteration times is reached, the step S3.4 is carried out;
those skilled in the art can set the network parameter w of the discriminator according to the actual situationtThe method of the present invention is only a preferred embodiment, and is not necessarily limited to the present invention.
Updating discriminator network parameter wtThe method comprises the following steps:
Figure BDA0003443913200000061
wherein clip (.) represents a clipping function in the deep learning, RMSProp (.) represents a learning rate adaptive optimizer in the deep learning,
Figure BDA0003443913200000062
the gradient function representing the loss function of the discriminator comprises the following specific calculation methods:
Figure BDA0003443913200000063
wherein x is(i)Is the ith training batch in the input parameters; z is a radical of(i)Sampling a batch in a sample distribution generated from a generator for an ith training;
Figure BDA0003443913200000064
a discriminator network, a person skilled in the art can select the discriminator network according to actual conditions;
step 3.4, calculate Generator loss function
Figure BDA0003443913200000065
And updates the generator parameter thetatAnd returning to the step 3.2;
one of ordinary skill in the art can set the generator loss function according to actual conditions
Figure BDA0003443913200000066
And a generator parameter θtThe method of the present invention is only a preferred embodiment, and is not necessarily limited to the present invention.
The gradient of the generator loss function is:
Figure BDA0003443913200000067
updating the generator parameter θtThe method comprises the following steps:
Figure BDA0003443913200000071
s4: and performing data enhancement in the photovoltaic cluster through the photovoltaic data enhancement neural network model constructed in S3, inputting the enhanced data into the CNN neural network, and training to obtain the photovoltaic prediction neural network model.
Further, the data collected in the step S1 are graphed, and the data in one day are divided into 48 points per day to form a picture with the size of 48 × 15 pixels; and then, performing countermeasure learning through the photovoltaic data enhancement neural network model to generate new data with the same distribution.
Further, the CNN neural network has a 5-layer structure, each of the first four layers includes a convolutional layer, a batch normalization layer, and a Linear rectification (Rectified Linear Unit) layer, and the last layer includes a full connection layer, a batch normalization layer, and a ReLU layer. And training the deep convolution network by adopting the original data and the new data to obtain a final prediction model.
S5: and inputting the weather forecast data into the photovoltaic prediction neural network model of S4 for photovoltaic prediction. Preferably, the photovoltaic prediction time ranges from 1 to 3 days.
An example of step S1 in the photovoltaic power prediction method of the present invention is as follows:
explaining the photovoltaic data of the households in 2014 published by an Australian power grid, selecting historical photovoltaic data of 105 households, counting the maximum photovoltaic output of each household in one year, performing per unit according to the formula of claim 2, and selecting a typical sunny output sequence. Clustering is carried out by taking pearson correlation distance as measurement, and partial parameters of the DBSCAN are finally determined through parameter adjustment, wherein the field radius is 0.04, and the minimum point number is 4. Clustering results as shown in fig. 2, it can be seen from the histogram that 105 distributed sites are divided into two clusters, and 9 sites are determined as outliers. The clustering timing chart is shown in fig. 3, the force curve can be roughly divided into two strips with higher overlapping degree from the timing chart, and the clustering result is basically consistent with the sense. And (3) searching the neighbor sites for 9 abnormal sites by adopting k-neighbor search, taking k as 5, classifying the abnormal sites into the clusters with the largest number, and finally classifying the clusters as shown in the figure 4.
The power output of each station in the cluster is added to obtain the total cluster output, the output curve and the corresponding meteorological data are divided according to the day, and the total time points are 15 characteristic 48 time points, so that a 48-by-15 data matrix is formed and is further imaged, as shown in fig. 5. The new data is generated by the built data-enhanced neural network for resisting learning, as shown in fig. 6, it can be seen that the newly generated data has similar data characteristics with the original data, and the improved GAN tends to be stable in the training process.
The original data and the generated data jointly form a training set for training the deep convolution network, the training set is compared with the deep convolution network which adopts the same structure and parameters and only uses the original data, the prediction indexes on the test set are shown in table 1, and the time sequence diagram of the prediction result is shown in fig. 7.
TABLE 1 error statistics for two methods comparison
Figure BDA0003443913200000081
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The present applicant has described and illustrated embodiments of the present invention in detail with reference to the accompanying drawings, but it should be understood by those skilled in the art that the above embodiments are merely preferred embodiments of the present invention, and the detailed description is only for the purpose of helping the reader to better understand the spirit of the present invention, and not for limiting the scope of the present invention, and on the contrary, any improvement or modification made based on the spirit of the present invention should fall within the scope of the present invention.

Claims (11)

1. The whole county photovoltaic prediction method based on cluster division and data enhancement is characterized by comprising the following steps of:
s1, collecting historical sunny photovoltaic output data and meteorological data of each photovoltaic output station in a set region and within a set time;
s2, obtaining the maximum power of each photovoltaic output station according to the collected sunny photovoltaic output data of S1, performing per unit, and then performing photovoltaic cluster division;
s3, constructing a photovoltaic data enhancement neural network model;
s4, performing data enhancement in the photovoltaic cluster through the photovoltaic data enhancement neural network model constructed in the S3, inputting the enhanced data to a CNN neural network, and training to obtain a photovoltaic prediction neural network model;
s5: and inputting the weather forecast data into the photovoltaic prediction neural network model of S4 for photovoltaic prediction.
2. The county photovoltaic prediction method based on cluster division and data enhancement of claim 1,
in S1, the photovoltaic output data is photovoltaic output data of the entire county on historical sunny days;
the meteorological data includes month, day, time, minute, direct irradiance, diffuse horizontal irradiance, total horizontal irradiance, ambient temperature, barometric pressure, relative humidity, wind direction, wind speed, ground surface reflectance, and generated power.
3. The integer county photovoltaic prediction method based on cluster partitioning and data enhancement according to claim 1 or 2,
in the step S2, the historical sunny photovoltaic output curve is a per unit output curve, the maximum power per station is calculated from the historical output data, and then a per unit value of the curve is calculated;
the formula for data per unit is as follows:
Figure FDA0003443913190000011
in the above formula, x is the original data of the sample, xmaxAnd z is the per unit data.
4. The integer county photovoltaic prediction method based on cluster division and data enhancement according to claim 1 or 3,
in the step S2, dividing the photovoltaic clusters by using a DBSCAN method, where a pearson correlation distance is used for distance measurement of the DBSCAN; and (4) carrying out clustering analysis on abnormal sites generated by the DBSCAN, searching the nearest k sites by using a k-nearest neighbor algorithm with pearson correlation distance as a measure, counting cluster numbers of the k sites, and dividing the abnormal sites into clusters with the largest number of numbers.
5. The county photovoltaic prediction method based on cluster division and data enhancement of claim 1,
in S3, the selectable neural network model is GAN.
6. The integer county photovoltaic prediction method based on cluster partitioning and data enhancement according to claim 1 or 5,
the optional neural network model is an improved GAN, and the specific algorithm is as follows:
s3.1, setting a learning rate alpha, a truncation parameter c, a batch training sample number m and a number of times n of one discriminator iteration of each iteration of a generatorcriticInitializing arbiter network parameter wtAnd generateDevice network parameter thetat；
S3.2, checking the Generator parameter θtWhether convergence is carried out or not, if so, the iteration is ended; if not, entering S3.3;
s3.3, checking whether the current iteration number reaches an iteration number threshold ncriticIf not, updating the network parameter w of the discriminatortUpdating the current iteration times, and repeating the step; if a threshold value n of the number of iterations is reachedcriticStep S3.4 is entered;
step 3.4, calculate Generator loss function
Figure FDA0003443913190000021
And updates the generator parameter thetatAnd returns to step 3.2.
7. The county photovoltaic prediction method based on cluster division and data enhancement of claim 6,
updating arbiter network parameter wtThe method comprises the following steps:
Figure FDA0003443913190000022
wherein clip (.) represents a clipping function in the deep learning, RMSProp (.) represents a learning rate adaptive optimizer in the deep learning,
Figure FDA0003443913190000023
the gradient of the loss function of the discriminator is represented, and the specific calculation method is as follows:
Figure FDA0003443913190000024
wherein x is(i)Is the ith training batch in the input parameters; z is a radical of(i)Sampling a batch in a sample distribution generated from a generator for an ith training;
Figure FDA0003443913190000025
is a network of discriminators.
8. The integer county photovoltaic prediction method based on cluster partitioning and data enhancement according to claim 6 or 7,
the gradient of the generator loss function is:
Figure FDA0003443913190000031
wherein,
Figure FDA0003443913190000032
as a network of discriminators, z(i)The batch in the sample distribution produced from the generator is sampled for the ith training.
9. The county photovoltaic prediction method based on cluster division and data enhancement of claim 8,
updating generator parameter QtThe method comprises the following steps:
Figure FDA0003443913190000033
RMSProp (.) denotes a learning rate adaptive optimizer in deep learning.
10. The county photovoltaic prediction method based on cluster division and data enhancement of claim 1,
the data enhancement of the S4 means that the data collected in the S1 is firstly picturized, the original data of one day is divided into 48 points every day, and a picture with the size of 48 x 15 pixels is formed; and then, performing countermeasure learning through the photovoltaic data enhancement neural network model to generate new data with the same distribution.
11. The integer county photovoltaic prediction method based on cluster division and data enhancement according to any one of claims 1 to 10,
in S4, the CNN neural network has a 5-layer structure, each of the first four layers includes a convolutional layer, a batch normalization layer, and a linear rectification layer, and the last layer includes a fully-connected layer, a batch normalization layer, and a ReLU layer.
CN202111646119.2A 2021-12-29 2021-12-29 Whole-county photovoltaic prediction method based on cluster division and data enhancement Pending CN114492941A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111646119.2A CN114492941A (en) 2021-12-29 2021-12-29 Whole-county photovoltaic prediction method based on cluster division and data enhancement

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111646119.2A CN114492941A (en) 2021-12-29 2021-12-29 Whole-county photovoltaic prediction method based on cluster division and data enhancement

Publications (1)

Publication Number Publication Date
CN114492941A true CN114492941A (en) 2022-05-13

Family

ID=81508943

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111646119.2A Pending CN114492941A (en) 2021-12-29 2021-12-29 Whole-county photovoltaic prediction method based on cluster division and data enhancement

Country Status (1)

Country Link
CN (1) CN114492941A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116365591A (en) * 2023-05-25 2023-06-30 北京智盟信通科技有限公司 Distributed light Fu Qun group control method, device and storage medium
CN116885715A (en) * 2023-09-05 2023-10-13 国网山西省电力公司太原供电公司 Distributed photovoltaic cluster power short-term prediction method and device
WO2023201552A1 (en) * 2022-04-20 2023-10-26 Southeast University County-wide photovoltaic prediction method based on cluster division and data enhancement

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110414748A (en) * 2019-08-12 2019-11-05 合肥阳光新能源科技有限公司 Photovoltaic power prediction technique
CN110705760A (en) * 2019-09-19 2020-01-17 广东工业大学 Photovoltaic power generation power prediction method based on deep belief network
WO2021007812A1 (en) * 2019-07-17 2021-01-21 深圳大学 Deep neural network hyperparameter optimization method, electronic device and storage medium
WO2021204226A1 (en) * 2020-04-09 2021-10-14 国网江苏省电力有限公司苏州供电分公司 Photovoltaic power prediction method under lightning condition based on thundercloud trajectory tracking

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021007812A1 (en) * 2019-07-17 2021-01-21 深圳大学 Deep neural network hyperparameter optimization method, electronic device and storage medium
CN110414748A (en) * 2019-08-12 2019-11-05 合肥阳光新能源科技有限公司 Photovoltaic power prediction technique
CN110705760A (en) * 2019-09-19 2020-01-17 广东工业大学 Photovoltaic power generation power prediction method based on deep belief network
WO2021204226A1 (en) * 2020-04-09 2021-10-14 国网江苏省电力有限公司苏州供电分公司 Photovoltaic power prediction method under lightning condition based on thundercloud trajectory tracking

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023201552A1 (en) * 2022-04-20 2023-10-26 Southeast University County-wide photovoltaic prediction method based on cluster division and data enhancement
CN116365591A (en) * 2023-05-25 2023-06-30 北京智盟信通科技有限公司 Distributed light Fu Qun group control method, device and storage medium
CN116365591B (en) * 2023-05-25 2023-08-25 北京智盟信通科技有限公司 Distributed light Fu Qun group control method, device and storage medium
CN116885715A (en) * 2023-09-05 2023-10-13 国网山西省电力公司太原供电公司 Distributed photovoltaic cluster power short-term prediction method and device
CN116885715B (en) * 2023-09-05 2023-11-14 国网山西省电力公司太原供电公司 Distributed photovoltaic cluster power short-term prediction method and device

Similar Documents

Publication Publication Date Title
CN114492941A (en) Whole-county photovoltaic prediction method based on cluster division and data enhancement
CN109902863B (en) Wind speed prediction method and device based on multi-factor time-space correlation
WO2023201552A1 (en) County-wide photovoltaic prediction method based on cluster division and data enhancement
CN110942205B (en) Short-term photovoltaic power generation power prediction method based on HIMVO-SVM
CN111210082A (en) Optimized BP neural network algorithm-based precipitation prediction method
CN109978284B (en) Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model
CN109086928A (en) Photovoltaic plant realtime power prediction technique based on SAGA-FCM-LSSVM model
CN105260800A (en) Photovoltaic module temperature prediction method and device
CN113657662B (en) Downscaling wind power prediction method based on data fusion
CN112990558B (en) Meteorological temperature and illumination prediction method based on deep migration learning
CN114240000A (en) Air quality prediction method based on space-time graph convolution network
CN105574615A (en) Spatial correlation and genetic algorithm (GA) based wind power forecast method for wavelet-BP neural network
CN117394306B (en) Wind power prediction model establishment method based on new energy grid connection and application thereof
CN111242355A (en) Photovoltaic probability prediction method and system based on Bayesian neural network
CN114418174A (en) Electric vehicle charging load prediction method
CN114882373A (en) Multi-feature fusion sandstorm prediction method based on deep neural network
CN111488974A (en) Deep learning neural network-based ocean wind energy downscaling method
CN117060407B (en) Wind power cluster power prediction method and system based on similar day division
CN112688306B (en) Photovoltaic weekly output scene prediction method based on weather forecast and historical power data
CN116885703B (en) Short-term wind-solar power prediction method for high-dimensional multi-element meteorological data fusion
CN116525135B (en) Method for predicting epidemic situation development situation by space-time model based on meteorological factors
CN115510767B (en) Regional air temperature prediction method based on depth space-time network
CN112712213B (en) Method and system for predicting deep migration learning energy consumption of concentrated air conditioning house
CN114626622A (en) Photovoltaic power prediction method, system, device and storage medium
CN112561181A (en) Photovoltaic power generation prediction system based on Unet network and foundation cloud picture

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