CN117634701A - Solar radiation short-term forecasting method and system based on regional photovoltaic set forecasting - Google Patents
Solar radiation short-term forecasting method and system based on regional photovoltaic set forecasting Download PDFInfo
- Publication number
- CN117634701A CN117634701A CN202311711210.7A CN202311711210A CN117634701A CN 117634701 A CN117634701 A CN 117634701A CN 202311711210 A CN202311711210 A CN 202311711210A CN 117634701 A CN117634701 A CN 117634701A
- Authority
- CN
- China
- Prior art keywords
- matrix
- data
- forecast
- forecasting
- solar radiation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000005855 radiation Effects 0.000 title claims abstract description 71
- 238000013277 forecasting method Methods 0.000 title claims abstract description 14
- 239000011159 matrix material Substances 0.000 claims abstract description 113
- 238000000034 method Methods 0.000 claims abstract description 26
- 238000010801 machine learning Methods 0.000 claims abstract description 18
- 230000004927 fusion Effects 0.000 claims abstract description 8
- 238000005259 measurement Methods 0.000 claims description 20
- 238000012549 training Methods 0.000 claims description 19
- 238000012795 verification Methods 0.000 claims description 9
- 230000004907 flux Effects 0.000 claims description 6
- 238000012163 sequencing technique Methods 0.000 claims description 6
- 230000002776 aggregation Effects 0.000 claims description 4
- 238000004220 aggregation Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000007619 statistical method Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 241000219094 Vitaceae Species 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 235000021021 grapes Nutrition 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 230000003094 perturbing effect Effects 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000002759 z-score normalization Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010972 statistical evaluation Methods 0.000 description 1
Classifications
-
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a solar radiation short-term forecasting method and a system based on regional photovoltaic set forecasting, which are used for producing a solar radiation short-term data set under multi-source photovoltaic mode set forecasting; producing solar radiation weight similarity set forecast data and constructing an optimal irradiance prediction matrix; producing multi-source meteorological region mode irradiance prediction data based on machine learning; and carrying out linear weighted data fusion on the weight similarity set forecast data and the multi-source meteorological region mode irradiance forecast data to obtain a final irradiance forecast result. The invention combines the traditional set forecasting method and the similar set method, thereby improving the solar radiation forecasting precision.
Description
Technical Field
The invention belongs to the technical field of new energy, and particularly relates to a solar radiation short-term forecasting method and system based on regional photovoltaic mode set forecasting.
Background
Currently, the installed amount of new energy stations is increasing, wherein concentrated photovoltaic and distributed photovoltaic have considerable installed amount demands in the future. The main factors influencing the solar photovoltaic power prediction are on the prediction of solar radiation related meteorological elements. The intermittence, volatility and randomness of solar radiation data greatly influence the short-term prediction accuracy of solar radiation, thereby causing uncertainty to photovoltaic output and grid-connected scheduling and profound influence to the safe and steady-state operation of a power grid.
In recent years, various short-term solar irradiance prediction methods are gradually developed at home and abroad in view of the self characteristic of solar energy, and two main flow directions of an NWP method and machine learning are formed. The NWP method codes the basic equation of motion of the atmosphere, and explains the occurrence and development of the atmospheric process from the aspects of atmospheric power and heat. Machine learning methods learn the intrinsic law from historical data of solar radiation and use it in future predictions. However, both methods have respective limitations, and the NWP method has larger uncertainty in the prediction result due to errors of initial values, parameterization schemes and space-time resolutions, and the machine learning method is limited by model selection, model super-parameter setting, feature engineering rationality and the like.
One of the methods to reduce NWP method uncertainty is aggregate forecasting. The aggregate forecast forms a group of aggregate which describes the future atmospheric condition as accurately as possible by perturbing the initial value or parameterized scheme parameters of the NWP, and can greatly reduce the uncertainty of the numerical mode. The more collection members of the collection forecast, the more accurate the characterization of future solar radiation. However, the collection forecasting requires more calculation resources, and usually only a small number of collection members are adopted, so that the solar photovoltaic collection forecasting precision is greatly affected. Meanwhile, the set forecasting member is only a forecasting mode of solar radiation at the current moment through a certain initial field, and no rule existing in the historical mapping relation of numerical mode data and measured data is additionally considered.
Disclosure of Invention
The invention provides a solar radiation short-term forecasting method and a system based on regional photovoltaic set forecasting, which combine a traditional set forecasting method and a similar set method, so that the solar radiation forecasting precision is improved.
In order to achieve the above purpose, the technical scheme of the invention is realized as follows:
a solar radiation short-term forecasting method based on regional photovoltaic set forecasting comprises the following steps:
s1, producing a solar radiation short-term data set under the forecast of a multisource photovoltaic mode set;
s2, producing solar radiation weight similarity set forecast data and constructing an optimal irradiance prediction matrix;
s3, producing multi-source meteorological region mode irradiance prediction data based on machine learning;
and S4, carrying out linear weighted data fusion on the weight similarity set forecast data and the multi-source meteorological region mode irradiance forecast data to obtain a final irradiance forecast result.
Further, the step S1 specifically includes:
s11: acquiring multi-source meteorological data; the multi-source meteorological data are meteorological data of m different meteorological sources;
s12: deploying a regional photovoltaic mode set forecasting system;
s13: respectively producing meshed solar radiation short-term forecast data sets of the plurality of different meteorological sources under regional photovoltaic mode set forecast; a short-term solar radiation forecast set is formed.
Further, the step S2 specifically includes:
s21, acquiring a solar radiation short-term forecast set for d days according to the step S1;
s22, collecting a historical actual measurement data matrix O of solar downward irradiance observed by a ground station irradiator of a target area, wherein the dimension of the matrix O is (q, S); wherein the number of ground stations in the target area is s, q=d×96;
s23, generating a multi-source region photovoltaic mode set forecasting classification result based on a matrix O;
s24, constructing an optimal irradiance prediction matrix based on the multi-source area photovoltaic mode set prediction classification result.
Further, the step S23 specifically includes:
s231: extracting a regional photovoltaic mode data matrix ST (ST 1, ST2, STm) under multi-source drive in the solar radiation short-term prediction set of d days in the step S21 based on the observation point position of the target regional ground station irradiator, wherein the dimension of the matrix ST is (m, q, S, e, n); wherein m represents the number of meteorological sources, n represents the number of meteorological variables related to solar radiation forecast, e represents the number of members of the set forecast, and the ground-down short-wave radiation fluxes in the matrix ST form a matrix SW (SW 1, SW., SWm) alone; the dimensions of the matrix SW are (m, q, s, e);
s232: calculating a site set average matrix EM (EM 1, EM2,.. EMm) under multi-source driving in the target area according to the matrices SW (SW 1, SW2,..swm) respectively, the dimensions of the matrix EM being (m, q, s);
s233: carrying out statistical analysis on a site set average matrix EM and the historical actual measurement data matrix O to respectively calculate RMSE and a correlation coefficient; calculating RMSE and correlation coefficient for each site; averaging RMSE and correlation coefficients for all sites within the target area to finally obtain a matrix RMSE-EM (RMSE-EM 1, RMSE-EM2,..rmse-EMm) and a matrix CORR-EM (CORR-EM 1, CORR-EM2,..corr-EMm);
s234: taking the inverse of each RMSE result in the matrix RMSE-EM and normalizing the data to obtain a matrix rmsir-EM (rmsir-EM 1, rmsir-EM 2., rmsir-EMm); adding the matrix rmseg-EM and the matrix CORR-EM corresponding position elements to obtain a matrix STAT-EM (STAT-EM 1, STAT-EM2,..the., STAT-EMm); sorting elements of the matrix STAT-EM from large to small; dividing m meteorological sources into 3 classes according to STAT-EM sequencing results, wherein the numbers of class 1 sources, class 2 sources and class 3 sources are respectively int (m/3), int (m/3) and m-2 x int (m/3), and the int represents downward rounding; the number of similar sets for the class 1 source, the class 2 source, and the class 3 source are respectively assigned to f1, f2, and f3, and f1> f2> f3.
Still further, step S24 specifically includes:
s241: calculating the Euclidean distance; taking one site data of one set of forecast members of one data source of the ST, namely eliminating dimensions m, e and s, wherein the ST is specially used as a matrix X, and the dimension of the matrix X is (q, n); carrying out data normalization processing on all n meteorological variables in the matrix X to form a matrix Y; dividing matrix Y into a matrix Y with a time dimension of a history period i And a matrix Y with a time dimension as a forecast moment t ；
Y i ＝(Y i,1 ,Y i,2 ,...,Y i,n )；
Y t ＝(Y t,1 ,Y t,2 ,...,Y t,n )；
The traversal history period calculates the euclidean distance dis,
sequencing the obtained Euclidean distance from small to large;
s242: s241, performing operation of step S241 on each station under each meteorological source driven by each set forecasting member of the solar radiation short-term forecasting set to obtain a distance sequence S;
s243: distributing the multisource similarity set membership; for the meteorological sources belonging to the 1 st, 2 nd and 3 rd sources, respectively taking the site irradiance actual measurement data of the meteorological sources at the first f1 time, the first f2 time and the first f3 time, namely taking the site irradiance actual measurement data of the meteorological source belonging to the 1 st source at the first f1 time of the historical actual measurement data matrix O, taking the site irradiance actual measurement data of the meteorological source belonging to the 2 nd source at the first f2 time of the historical actual measurement data matrix O, taking the site irradiance actual measurement data of the meteorological source belonging to the 3 rd source at the first f3 time of the historical actual measurement data matrix O, performing a step 242 operation on the solar radiation short-term forecast set of each meteorological source, and obtaining e (int (m/3)) f1+ int (m/3) f2+ int (m/3)) members of each site in the ordered target area at each pre-time according to the obtained distance sequence matrix S; wherein int represents rounding down; e is the number of collection forecast members, m is the number of weather sources;
s244: according to the result of step S243, carrying out set average on each site in the target area to obtain an optimal irradiance prediction matrix F1 containing each site under the weight similarity set method; f1 has dimensions (s, 1).
Still further, the step S3 specifically includes:
s31: preparing training data; respectively selecting a matrix ST in the step S231 and a matrix O in the step S22 as model characteristic columns and actual measurement data;
s32: determining a machine learning model;
s33: model training and super-parameter optimization; dividing training data into two sections q1 and q2, wherein q1 comprises a training set and a verification set, and q2 is a test set; dividing q1 into k groups of training sets and verification sets by using a k-fold cross verification method, and performing super-parameter tuning on each group of training set data through grid search; training to obtain a prediction model;
s34: model prediction; and (3) transmitting corresponding features to the prediction model trained in the step S33 to obtain an irradiance prediction matrix F2, wherein the dimension of the irradiance prediction matrix F2 is (S, 1).
Further, step S4 includes: linear weighted data fusion is carried out on the matrix F1 and the matrix F2 to obtain a final irradiance prediction result F3:
F3＝p1·F1+p2·F2；
where p1 and p2 are weight coefficients, p1+p2=1.
The invention also provides a solar radiation short-term forecasting system based on regional photovoltaic set forecasting, which comprises the following steps:
the multi-source data module is used for producing a solar radiation short-term data set under the forecast of a multi-source photovoltaic mode set;
weight similarity data module for producing solar radiation weight similarity set forecast data and constructing optimal irradiance prediction matrix
The machine learning data module is used for producing multi-source meteorological area mode irradiance prediction data based on machine learning;
and a prediction result module: and carrying out linear weighted data fusion on the weight similarity set forecast data and the multi-source meteorological region mode irradiance forecast data to obtain a final irradiance forecast result.
Further, the multi-source data module specifically includes:
weather data unit: acquiring multi-source meteorological data; the multi-source meteorological data are meteorological data of m different meteorological sources;
a system deployment unit: deploying a regional photovoltaic mode set forecasting system;
a forecast aggregation unit: respectively producing meshed solar radiation short-term forecast data sets of the plurality of different meteorological sources under regional photovoltaic mode set forecast; a short-term solar radiation forecast set is formed.
Further, the weight similarity data module specifically includes:
aggregation unit: s1, acquiring a solar radiation short-term forecasting set of d days;
actual measurement data unit: collecting a historical actual measurement data matrix O of solar downward irradiance observed by a ground station irradiator in a target area, wherein the dimension of the matrix O is (q, s); wherein the number of ground stations in the target area is s, q=d×96;
classification unit: generating a multi-source region photovoltaic mode set forecasting classification result based on a matrix O;
prediction matrix unit: and constructing an optimal irradiance prediction matrix based on the multi-source region photovoltaic mode set prediction classification result.
Compared with the prior art, the invention has the following beneficial effects:
under the condition of not increasing a large amount of computing resources, the invention combines the traditional set forecasting method with the similar set method, on one hand, the number of set forecasting members is increased, on the other hand, the error rule of solar radiation set forecasting and history measured data is considered in irradiance forecasting (namely, the advantages of the traditional set forecasting and machine learning algorithm are combined), and the solar radiation forecasting precision is improved.
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the present invention;
FIG. 2 is a schematic flow chart of step S1 according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart of step S2 according to an embodiment of the present invention;
FIG. 4 is a flow chart of step S24 according to an embodiment of the present invention;
fig. 5 is a schematic flow chart of step S3 in the embodiment of the present invention.
Detailed Description
It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other.
First, the technical terms in this embodiment will be explained:
multisource meteorological data: weather sources for the multi-source weather data include, but are not limited to, china weather office CMA (GRAPES), european mid-term weather forecast ECMWF (HRES), united states national environmental forecast center NCEP (GFS), canadian weather center CMC (GDPS), german weather office DWD (ICON), japan weather office JMA (GSM), french weather office MF (ARPEGE), australian weather office BOM (ACCESS-G);
irradiance of: i.e. solar radiant flux, in W/m 2 ；
Aggregate forecast: with respect to deterministic forecasting, aggregate forecasting characterizes the atmosphere evolution process by perturbing initial values or parameterized scheme parameters to form multiple groups of aggregate members;
aggregate average: averaging the forecast of a certain meteorological element by a plurality of collection members;
regional photovoltaic mode aggregate forecasting system: the system comprises, but is not limited to, a set forecasting mode system such as WRF-Solar-EPS and the like which can disturb key parameters affecting Solar radiation simulation;
z-score normalization: the data were converted to a normal distribution with a mean of 0 and standard deviation of 1. The formula is as follows: x '= (x-mean)/std, where x is the raw data, x' is the normalized data, mean is the mean of the dataset, and std is the standard deviation of the dataset.
Machine learning algorithm: including but not limited to linear regression, ridge regression, random forests, support vector machines, lightGBM, XGBoost, adaBoost, GBRT, etc. regression algorithms.
NWP method: the numerical weather forecast method is characterized in that according to the actual condition of the atmosphere, under the condition of a certain initial value and a certain boundary value, a large-scale computer is used for numerical calculation, a hydrodynamic and thermodynamic equation set representing the weather evolution process is solved, and the aerodynamic state and the weather phenomenon of a certain period of time in the future are predicted.
RMSE: root mean square error, characterizing absolute deviation (o is measured data, s is pattern data, i is time dimension):
correlation coefficient: characterization of the linear trend (o is measured data, s is pattern data, i is time dimension):
the following describes the embodiments of the present invention with reference to the drawings.
As shown in fig. 1, the solar radiation short-term forecasting method based on regional photovoltaic mode set forecasting of the invention comprises the following steps:
s1: and producing a solar radiation data set under the forecast of the multi-source photovoltaic mode set.
The step specifically includes, as shown in fig. 2:
s11: and acquiring multi-source meteorological data. The global weather forecast grid point data set used in the present invention is provided by m weather centers (G1, G2,..gm) including, but not limited to, chinese weather office CMA (GRAPES), mid-european weather forecast ECMWF (HRES), united states national environmental forecast center NCEP (GFS), canadian weather center CMC (GDPS), german weather office DWD (ICON), japanese weather office JMA (GSM), french weather office MF (ARPEGE), australian weather office BOM (ACCESS-G). The reason for using multi-source weather data is that weather forecast accuracy varies from weather source to weather source in different spaces and at different times.
S12: a regional photovoltaic mode set forecasting system is deployed. The method comprises the step of installing a relevant dependency library and a regional photovoltaic mode set forecasting system in a linux environment.
S13: taking a meteorological source GFS as an example, a solar radiation short-term forecasting set under photovoltaic mode set forecasting is produced.
S131: a simulation target area is defined and geographic information data is interpolated into the simulation area.
S132: and extracting GFS meteorological source data stored in a grib format and interpolating the GFS meteorological source data into a selected simulation area.
S133: and setting the disturbance parameter of the aggregate forecast, and setting the number of the aggregate forecast members as e.
S134: the operating area photovoltaic mode set forecasting system obtains a meshed meteorological set data set D1 under the driving of GFS, wherein the meshed meteorological set data set D1 comprises D1, D2. The step can also be understood as downscaling the global meshed weather data by using the photovoltaic regional set forecasting system.
S14: repeating S13 steps (D2, D3,..dm, where m is the number of meteorological sources) with the data of the remaining meteorological sources.
S2: and producing solar radiation weight similarity set forecast data.
The step specifically includes, as shown in fig. 3:
s21: and (3) operating the regional photovoltaic mode set forecasting system according to the steps included in the step S1 and producing data over d days, wherein m.e sets of grid solar radiation related meteorological set data sets are shared. q=d×96, where 96 represents the time resolution of the data within a day, i.e. one data sample every 15 minutes. The data set (D1, D2,..dm) contains solar radiation related meteorological variables such as ground down short wave radiant flux, ground down short wave normal direct irradiance, ground down short wave scattered irradiance, solar zenith angle, 2m temperature, dew point temperature, ground air pressure, 10m wind, etc.
S22: and collecting solar downward irradiance data observed by a ground station irradiator in a target area to form a historical actual measurement data matrix O. The number of ground stations in the target area is s. The history time of the observed data is d days or more. The dimensions of matrix O are (q, s).
S23: and generating a multi-source region photovoltaic mode set forecast classification result based on irradiance history actual measurement data statistical evaluation.
S231: regional photovoltaic mode data under multi-source drive in S21 is extracted based on ground-illuminator viewpoint in the target region (ST 1, ST2,..stm). The dimensions of the ST matrix are (m, q, s, e, n). Wherein n represents the number of weather variables related to solar radiation forecast. The ground down short wave radiant flux in ST data alone forms SW (SW 1, SW., SWm). The dimensions of the SW matrix are (m, q, s, e).
S232: the ground down short wave radiant flux data sets (SW 1, SW2,..swm) forecast for the multisource photovoltaic pattern set in S231 are calculated as site set averages (EM 1, EM2,.. EMm) within the target area, respectively. The dimensions of the EM matrix are (m, q, s).
S233: and (3) carrying out statistical analysis on the site set average matrix EM (with the dimension of (m, q, S)) and the site observation matrix O (with the dimension of (q, S)) acquired in the step S22 to respectively calculate the RMSE and the correlation coefficient. The specific flow is as follows: 1) Calculating s RMSE and correlation coefficients for each site (s sites total); 2) The RMSE and correlation coefficients for s sites within the target area are averaged to finally yield RMSE-EM (RMSE-EM 1, RMSE-EM2,..rmse-EMm) and CORR-EM (CORR-EM 1, CORR-EM2,..corr-EMm). The RMSE-EM and CORR-EM matrix dimensions are (m, 1).
S234: the m RMSE results (i.e., each element of the RMSE-EM matrix) were reciprocal and data normalized to yield rmseg-EM (rmseg-EM 1, rmseg-EM 2.). Adding the rmseg-EM and CORR-EM corresponding positional elements yields STAT-EM (STAT-EM 1, STAT-EM 2.,. The term STAT-EMm). The STAT-EM matrix elements with dimension m are ordered from large to small. The m meteorological sources are equally divided into 3 classes according to STAT-EM sequencing results. The numbers of the class 1 source, the class 2 source and the class 3 source are respectively int (m/3), int (m/3) and m-2 x int (m/3). Where int denotes rounding down. The number of similar sets for the class 1 source, class 2 source, and class 3 source are assigned to f1, f2, and f3, respectively (f 1> f2> f3, e.g., f1=50, f2=20, f3=5). In order to balance the timeliness of the preferred weather source with the consumption of computing resources, step S23 computes f1, f2, and f3 once every d1 days. Whereas fixed f1, f2, and f3 are read directly over the d1 day time frame.
S24: and constructing a solar radiation weight similarity set forecast data set based on the mode and the observed solar radiation history data. The specific scheme of this step is shown in fig. 4:
s241: and calculating the Euclidean distance. Taking as an example the data of a certain site (dimension s) of a certain member (dimension e) of the short-term collection forecast of solar radiation driven by GFS (dimension m is eliminated), ST of the original dimension (m, q, s, e, n) is specified as a matrix X, and the dimension of X is (q, n)). First, data Z-score normalization processing is performed on each variable (n in total) in X to form a matrix Y. The time dimension (q=h+1) is divided into a history period (time dimension is denoted by i, i=1, 2,..h, h is the total number of samples of history data) and a forecast moment (time dimension is denoted by t):
Y i ＝(Y i,1 ,Y i,2 ,...,Y i,n )
Y t ＝(Y t,1 ,Y t,2 ,...,Y t,n )
secondly, the Euclidean distance d is calculated by traversing the history period, and the calculation formula is as follows:
subsequently, the h Euclidean distances are sorted from small to large.
S242: and S241, performing operation of step S241 on each station (dimension S) under each set member (dimension e) driven by each meteorological source (dimension m) to obtain m.times.e.s distance sequence matrixes S with the length of h. The dimensions of the matrix S are (m, e, S, h). The meaning of the distance sequence matrix S is that the similarity degree of the forecast results at the current forecast moment and all data of the historical period are stored.
S243: multiple source affinity set membership is assigned. Taking GFS as an example, if GFS belongs to a class 1 source after S234 calculation, taking site irradiance measured data at the first f1 time points of the history measured data matrix O, and if GFS belongs to a class 2 source or a class 3 source, the same applies. Step S242 is performed on the photovoltaic short-term set forecast driven by each global weather source, and finally, e (int (m/3) ×f1+int (m/3) ×f2+ (m-2×int (m/3))×f3 set members of S sites in the target area at t time are obtained. Where int denotes rounding down. To further specifically illustrate the calculation process, EC, GFS, and MF (i.e., m=3) are taken as examples. Assuming that the set membership e is 20, and EC, GFS and MF belong to a class 1 source, a class 2 source and a class 3 source respectively, the similarity set numbers are f1, f2 and f3 respectively; let f1, f2 and f3 be 50, 20 and 5, respectively. The weighted similarity set membership of each station driven by EC, GFS and MF at time t obtained by the calculation in the step is respectively 1000 (20×50), 400 (20×20) and 100 (20×5);
s244: and carrying out set average (namely eliminating the dimension e, the dimension h and the dimension m) on each site in the target area to obtain the optimal irradiance prediction matrix F1, F1 of each site under the weight similarity set method, wherein the dimension is (s, 1).
S3: machine learning based multi-source weather zone mode irradiance prediction is produced.
The specific scheme of this step is shown in fig. 5:
s31: training data is prepared. And respectively selecting the ST matrix in the step S231 and the O matrix in the step S22 as model characteristic columns and measured data.
S32: a machine learning model is determined. The machine learning algorithm involved in this step includes, but is not limited to, linear regression, ridge regression, random forest, support vector machine, lightGBM, XGBoost, adaBoost, GBRT, etc. regression algorithms.
S33: model training and super-parameter optimization. The historical data period is divided into two sections q1 and q2, q1 comprises a training set and a verification set, and q2 is a test set. And dividing q1 into k groups of training sets and verification sets by using a k-fold cross verification method, and performing super-parameter tuning on data of each group of training sets through grid search.
S34: model prediction. And (3) transmitting corresponding features to the model trained in the step S33 to obtain an irradiance prediction matrix F2, wherein the dimension of the irradiance prediction matrix F2 is (S, 1).
S4: and fusing the weight similarity set forecast data and the machine learning forecast. And (3) carrying out linear weighted data fusion on the F1 calculated in the S2 and the F2 calculated in the S3 to obtain a final irradiance prediction result F3:
F3＝p1·F1+p2·F2；
where p1 and p2 are weight coefficients, p1+p2=1.
In the invention, more meteorological sources other than the meteorological sources can be selected, the Euclidean distance can be changed into the rest distances such as the Markov distance or the cosine distance, and the selected 8 solar radiation related meteorological variables can be changed into other or more meteorological variables.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (10)
1. A solar radiation short-term forecasting method based on regional photovoltaic set forecasting is characterized by comprising the following steps:
s1, producing a solar radiation short-term data set under the forecast of a multisource photovoltaic mode set;
s2, producing solar radiation weight similarity set forecast data and constructing an optimal irradiance prediction matrix;
s3, producing multi-source meteorological region mode irradiance prediction data based on machine learning;
and S4, carrying out linear weighted data fusion on the weight similarity set forecast data and the multi-source meteorological region mode irradiance forecast data to obtain a final irradiance forecast result.
2. The short-term solar radiation forecasting method based on regional photovoltaic aggregate forecasting of claim 1, wherein step S1 specifically comprises:
s11: acquiring multi-source meteorological data; the multi-source meteorological data are meteorological data of m different meteorological sources;
s12: deploying a regional photovoltaic mode set forecasting system;
s13: respectively producing meshed solar radiation short-term forecast data sets of the plurality of different meteorological sources under regional photovoltaic mode set forecast; a short-term solar radiation forecast set is formed.
3. The short-term solar radiation forecasting method based on regional photovoltaic aggregate forecasting of claim 1, wherein step S2 specifically comprises:
s21, acquiring a solar radiation short-term forecast set for d days according to the step S1;
s22, collecting a historical actual measurement data matrix O of solar downward irradiance observed by a ground station irradiator of a target area, wherein the dimension of the matrix O is (q, S); wherein the number of ground stations in the target area is s, q=d×96;
s23, generating a multi-source region photovoltaic mode set forecasting classification result based on a matrix O;
s24, constructing an optimal irradiance prediction matrix based on the multi-source area photovoltaic mode set prediction classification result.
4. A method for short-term forecasting of solar radiation based on regional photovoltaic aggregate forecasting according to claim 3, characterized in that step S23 specifically comprises:
s231: extracting a regional photovoltaic mode data matrix ST (ST 1, ST2, STm) under multi-source drive in the solar radiation short-term prediction set of d days in the step S21 based on the observation point position of the target regional ground station irradiator, wherein the dimension of the matrix ST is (m, q, S, e, n); wherein m represents the number of meteorological sources, n represents the number of meteorological variables related to solar radiation forecast, e represents the number of members of the set forecast, and the ground-down short-wave radiation fluxes in the matrix ST form a matrix SW (SW 1, SW., SWm) alone; the dimensions of the matrix SW are (m, q, s, e);
s232: calculating a site set average matrix EM (EM 1, EM2,.. EMm) under multi-source driving in the target area according to the matrices SW (SW 1, SW2,..swm) respectively, the dimensions of the matrix EM being (m, q, s);
s233: carrying out statistical analysis on a site set average matrix EM and the historical actual measurement data matrix O to respectively calculate RMSE and a correlation coefficient; calculating RMSE and correlation coefficient for each site; averaging RMSE and correlation coefficients for all sites within the target area to finally obtain a matrix RMSE-EM (RMSE-EM 1, RMSE-EM2,..rmse-EMm) and a matrix CORR-EM (CORR-EM 1, CORR-EM2,..corr-EMm);
s234: taking the inverse of each RMSE result in the matrix RMSE-EM and normalizing the data to obtain a matrix rmsir-EM (rmsir-EM 1, rmsir-EM 2., rmsir-EMm); adding the matrix rmseg-EM and the matrix CORR-EM corresponding position elements to obtain a matrix STAT-EM (STAT-EM 1, STAT-EM2,..the., STAT-EMm); sorting elements of the matrix STAT-EM from large to small; dividing m meteorological sources into 3 classes according to STAT-EM sequencing results, wherein the numbers of class 1 sources, class 2 sources and class 3 sources are respectively int (m/3), int (m/3) and m-2 x int (m/3), and the int represents downward rounding; the number of similar sets for the class 1 source, the class 2 source, and the class 3 source are respectively assigned to f1, f2, and f3, and f1> f2> f3.
5. The method for short-term forecasting of solar radiation based on regional photovoltaic aggregate forecasting of claim 4, wherein step S24 specifically comprises:
s241: calculating the Euclidean distance; taking one site data of one set of forecast members of one data source of the matrix ST, namely eliminating the dimensions m, e and s, wherein the matrix ST is specially used as a matrix X, and the dimension of the matrix X is (q, n); carrying out data normalization processing on all n meteorological variables in the matrix X to form a matrix Y; dividing matrix Y into a matrix Y with a time dimension of a history period i And a matrix Y with a time dimension as a forecast moment t ；
Y i ＝(Y i，1 ，Y i,2 ，...,Y i,n )；
Y t ＝(Y t,1 ，Y t,2 ，...,Y t,n )；
The traversal history period calculates the euclidean distance dis,
sequencing the obtained Euclidean distance from small to large;
s242: s241, performing operation of step S241 on each station under each meteorological source driven by each set forecasting member of the solar radiation short-term forecasting set to obtain a distance sequence matrix S;
s243: distributing the multisource similarity set membership; for meteorological sources belonging to the 1 st, 2 nd and 3 rd sources, respectively taking site irradiance actual measurement data of the first f1 times, the first f2 times and the first f3 times, performing step 242 operation on a solar radiation short-term forecast set of each meteorological source, and obtaining e (int (m/3) f1+int (m/3) f2+ (m-2) f 3) set members of each site in the target area after sequencing at each forecast time according to the obtained distance sequence matrix S; wherein int represents rounding down; e is the number of collection forecast members, m is the number of weather sources;
s244: according to the result of step S243, carrying out set average on each site in the target area to obtain an optimal irradiance prediction matrix F1 containing each site under the weight similarity set method; f1 has dimensions (s, 1).
6. The method for short-term forecasting of solar radiation based on regional photovoltaic aggregate forecasting of claim 5, wherein step S3 specifically comprises:
s31: preparing training data; respectively selecting a matrix ST in the step S231 and a matrix O in the step S22 as model characteristic columns and actual measurement data;
s32: determining a machine learning model;
s33: model training and super-parameter optimization; dividing training data into two sections q1 and q2, wherein q1 comprises a training set and a verification set, and q2 is a test set; dividing q1 into k groups of training sets and verification sets by using a k-fold cross verification method, and performing super-parameter tuning on each group of training set data through grid search; training to obtain a prediction model;
s34: model prediction; and (3) transmitting corresponding features to the prediction model trained in the step S33 to obtain an irradiance prediction matrix F2, wherein the dimension of the irradiance prediction matrix F2 is (S, 1).
7. The short-term solar radiation forecasting method based on regional photovoltaic aggregate forecasting of claim 6, wherein step S4 comprises: linear weighted data fusion is carried out on the matrix F1 and the matrix F2 to obtain a final irradiance prediction result F3:
F3＝p1·F1+p2·F2；
where p1 and p2 are weight coefficients, p1+p2=1.
8. A short-term solar radiation forecasting system based on regional photovoltaic aggregate forecasting, comprising:
the multi-source data module is used for producing a solar radiation short-term data set under the forecast of a multi-source photovoltaic mode set;
weight similarity data module for producing solar radiation weight similarity set forecast data and constructing optimal irradiance prediction matrix
The machine learning data module is used for producing multi-source meteorological area mode irradiance prediction data based on machine learning;
and a prediction result module: and carrying out linear weighted data fusion on the weight similarity set forecast data and the multi-source meteorological region mode irradiance forecast data to obtain a final irradiance forecast result.
9. The regional photovoltaic aggregate forecast-based short-term solar radiation forecast system of claim 8, wherein the multi-source data module specifically comprises:
weather data unit: acquiring multi-source meteorological data; the multi-source meteorological data are meteorological data of m different meteorological sources;
a system deployment unit: deploying a regional photovoltaic mode set forecasting system;
a forecast aggregation unit: respectively producing meshed solar radiation short-term forecast data sets of the plurality of different meteorological sources under regional photovoltaic mode set forecast; a short-term solar radiation forecast set is formed.
10. The regional photovoltaic aggregate forecast-based short-term solar radiation forecast system of claim 8, wherein the weight similarity data module specifically comprises:
aggregation unit: s1, acquiring a solar radiation short-term forecasting set of d days;
actual measurement data unit: collecting a historical actual measurement data matrix O of solar downward irradiance observed by a ground station irradiator in a target area, wherein the dimension of the matrix O is (q, s); wherein the number of ground stations in the target area is s, q=d×96;
classification unit: generating a multi-source region photovoltaic mode set forecasting classification result based on a matrix O;
prediction matrix unit: and constructing an optimal irradiance prediction matrix based on the multi-source region photovoltaic mode set prediction classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311711210.7A CN117634701A (en) | 2023-12-13 | 2023-12-13 | Solar radiation short-term forecasting method and system based on regional photovoltaic set forecasting |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311711210.7A CN117634701A (en) | 2023-12-13 | 2023-12-13 | Solar radiation short-term forecasting method and system based on regional photovoltaic set forecasting |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117634701A true CN117634701A (en) | 2024-03-01 |
Family
ID=90033877
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311711210.7A Pending CN117634701A (en) | 2023-12-13 | 2023-12-13 | Solar radiation short-term forecasting method and system based on regional photovoltaic set forecasting |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117634701A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801427A (en) * | 2021-04-08 | 2021-05-14 | 国能日新科技股份有限公司 | XGboost-based photovoltaic irradiance forecasting method and system |
| CN115189356A (en) * | 2022-09-13 | 2022-10-14 | 国能日新科技股份有限公司 | Wind and light power prediction correction method and device based on atmospheric circulation stable state |
| WO2023216576A1 (en) * | 2022-05-13 | 2023-11-16 | 中国电力科学研究院有限公司 | Photovoltaic power generation short-term power prediction method and system |
| CN117194957A (en) * | 2023-10-08 | 2023-12-08 | 贵州北盘江电力股份有限公司马马崖光伏分公司 | Ultra-short-term prediction method based on satellite inversion radiation data technology |
-
2023
- 2023-12-13 CN CN202311711210.7A patent/CN117634701A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801427A (en) * | 2021-04-08 | 2021-05-14 | 国能日新科技股份有限公司 | XGboost-based photovoltaic irradiance forecasting method and system |
| WO2023216576A1 (en) * | 2022-05-13 | 2023-11-16 | 中国电力科学研究院有限公司 | Photovoltaic power generation short-term power prediction method and system |
| CN115189356A (en) * | 2022-09-13 | 2022-10-14 | 国能日新科技股份有限公司 | Wind and light power prediction correction method and device based on atmospheric circulation stable state |
| CN117194957A (en) * | 2023-10-08 | 2023-12-08 | 贵州北盘江电力股份有限公司马马崖光伏分公司 | Ultra-short-term prediction method based on satellite inversion radiation data technology |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919353B (en) | Distributed photovoltaic prediction method of ARIMA model based on spatial correlation | |
| CN107766990B (en) | Method for predicting power generation power of photovoltaic power station | |
| CN108197744B (en) | Method and system for determining photovoltaic power generation power | |
| CN111369070B (en) | Multimode fusion photovoltaic power prediction method based on envelope clustering | |
| CN109002915B (en) | Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model | |
| CN105373857A (en) | Photovoltaic power station irradiance prediction method | |
| CN109558975B (en) | Integration method for multiple prediction results of power load probability density | |
| CN105868853B (en) | Method for predicting short-term wind power combination probability | |
| CN107944622A (en) | Wind power forecasting method based on continuous time cluster | |
| CN114444378A (en) | Short-term power prediction method for regional wind power cluster | |
| CN115049171B (en) | Photovoltaic power prediction method and system based on feature migration | |
| CN112418346B (en) | Numerical weather forecast total radiation system error classification calculation method | |
| CN108319772A (en) | A kind of analysis method again of wave long term data | |
| CN112307672A (en) | BP neural network short-term wind power prediction method based on cuckoo algorithm optimization | |
| CN115759389A (en) | Day-ahead photovoltaic power prediction method based on weather type similar day combination strategy | |
| CN116451821A (en) | Deep learning-based large-area new energy generation power prediction method and system | |
| CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
| CN117277372A (en) | Multi-time-scale joint scheduling method and system for optical storage station and electronic equipment | |
| CN114483485A (en) | Method for improving wind speed prediction of Nudging wind power plant observation data | |
| CN116885703B (en) | Short-term wind-solar power prediction method for high-dimensional multi-element meteorological data fusion | |
| CN116484998A (en) | Distributed photovoltaic power station power prediction method and system based on meteorological similar day | |
| CN116050636A (en) | Output prediction method, device, equipment and medium of photovoltaic power station | |
| CN115907228A (en) | Short-term power load prediction analysis method based on PSO-LSSVM | |
| CN117634701A (en) | Solar radiation short-term forecasting method and system based on regional photovoltaic set forecasting | |
| CN112116127B (en) | Photovoltaic power prediction method based on association of meteorological process and power fluctuation |
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 |