CN107766990B - Method for predicting power generation power of photovoltaic power station - Google Patents

Method for predicting power generation power of photovoltaic power station Download PDF

Info

Publication number
CN107766990B
CN107766990B CN201711103682.9A CN201711103682A CN107766990B CN 107766990 B CN107766990 B CN 107766990B CN 201711103682 A CN201711103682 A CN 201711103682A CN 107766990 B CN107766990 B CN 107766990B
Authority
CN
China
Prior art keywords
data
max
clustering
meteorological
weather
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.)
Active
Application number
CN201711103682.9A
Other languages
Chinese (zh)
Other versions
CN107766990A (en
Inventor
梅飞
刘皓明
李玉杰
王力
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.)
Hohai University HHU
Original Assignee
Hohai University HHU
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 Hohai University HHU filed Critical Hohai University HHU
Priority to CN201711103682.9A priority Critical patent/CN107766990B/en
Publication of CN107766990A publication Critical patent/CN107766990A/en
Application granted granted Critical
Publication of CN107766990B publication Critical patent/CN107766990B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F18/23—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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Game Theory and Decision Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Quality & Reliability (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method for predicting the power generation power of a photovoltaic power station, which comprises the steps of extracting six meteorological features by taking the day as a unit by utilizing historical meteorological data of the photovoltaic power station, and establishing a meteorological feature library; clustering the day characteristic data in the meteorological characteristic library through a KFCM algorithm to realize weather type classification, and carrying out category marking on the power data and the meteorological data of each day; establishing an SVR sub-model according to the category labels and the power data and the meteorological data in each category; identifying the weather type of a target day by using an SVM (support vector machine) according to the weather characteristics of the target day provided by the numerical weather forecast, and selecting a corresponding SVR (support vector regression) sub-model; establishing an ARIMA model by using real-time monitoring data of a target day, and realizing real-time prediction of irradiation intensity and air temperature by using a rolling prediction mode; and inputting the predicted values of the irradiation intensity and the air temperature into the selected SVR submodel to obtain a power prediction result of the photovoltaic power station. The method improves the prediction precision of the generated power of the photovoltaic power station.

Description

Method for predicting power generation power of photovoltaic power station
Technical Field
The invention relates to a method for predicting the generated power of a photovoltaic power station, and belongs to the technical field of power system automation.
Background
Photovoltaic power generation has been rapidly developed at home and abroad in recent years as a main technical means for realizing solar energy utilization. By the end of 2015, the global photovoltaic cumulative installed capacity reached 227 GW. The accumulated installed capacity of the photovoltaic power generation in China reaches 43.18GW, wherein the photovoltaic power station is 37.12GW, and the country with the largest installed capacity of the photovoltaic power generation becomes the world. Meanwhile, 15.13GW is added in China in 2015, which is also one of countries with faster global growth. The large-scale development and application of the photovoltaic power station bring serious influence on the stability of a power grid and the quality of electric energy, and the photovoltaic consumption becomes an important problem which hinders the development of the photovoltaic industry. The photovoltaic power prediction can provide important support for coordinated control and scheduling operation of a power grid on one hand, and can improve photovoltaic consumption capacity and increase the investment return rate of a photovoltaic power station on the other hand. On a time scale, photovoltaic power prediction can be classified into ultra-short term, and medium-long term. From the grid operation point of view, the shorter the period of prediction, the more advantageous is the handling of emergency and preventive conditions. Therefore, ultra-short term power prediction for photovoltaic power plants is of high interest.
Conventional photovoltaic power prediction is divided into direct and indirect methods. The direct prediction model is a generated power regression model established by using historical data related to irradiation intensity, air temperature, humidity, wind speed and the like, the data is usually derived from numerical weather forecast (NWP), and the modeling method comprises a statistical model, a neural network, a support vector machine and the like. The indirect prediction is divided into two stages, wherein the first stage predicts the solar radiation intensity or other meteorological information, and the second stage calculates the power generation power. The irradiation intensity prediction method comprises cloud picture-based irradiation prediction, multiple regression, time series, Markov chain and the like. In the aspect of direct prediction, an adaptive photovoltaic ultra-short term output prediction model based on a Support Vector Machine (SVM) is provided by an adaptive photovoltaic ultra-short term output prediction model [ J ] based on machine learning, a power grid technology, 2015,39(2):307 and 311 ] of Gaoyang and the like (Gaoyang, Zhang Biling, Maojinli, Liuyong); d G.De Giorgi et al (De Giorgi MG, Congedo PM, Malvoni M.Photovoltic power for evaluating using static methods: impact of weather data [ J ]. IET Science Measurement & Technology,2014,8(3):90-97.) have established an Artificial Neural Network (ANN) based photovoltaic prediction output model; melitt et al (Melitt A, Pavan AM, Lughi V.short-term evaluating of power production in a large-scale photovoltaic plant [ J ]. Solar Energy,2014,105:401-413.) develop three different Artificial Neural Network (ANN) models with databases of Solar irradiance, battery temperature and power output, applicable to photovoltaic output prediction of three typical weathers (sunny, partly cloudy and cloudy); in the aspect of indirect prediction, the prediction of irradiation intensity is the most critical, and R.Arquez et al (Arquez R, Coombra CFM. intra-road DNI for detecting based on closed tracking image analysis [ J ] Solar Energy,2013,91:327 and 336.) provides a direct illumination intensity prediction method based on cloud tracking image analysis; H.D. Yang et al (Yang HD, Kurtz B, Nguyen D, et al.solar ir radial for detecting using a ground-based sky image manager leveled at UC San Diego [ J ]. Solar Energy,2014,103:502-524.) propose a Solar radiation intensity prediction method for foundation cloud map analysis; escrig et al (Escrig H, Batleles FJ, Alonso J. cloud detection, classification and motion estimation using geological information for cloud cover for Energy,2013,55: 853) use geostationary satellite images for cloud detection, classification and motion estimation to predict irradiation intensity. In recent years, hybrid prediction models are rapidly developed, and the basic process can be divided into three stages of classification, regression and prediction. H.T.Yang et al (Yang HT, Huang CM, Huang YC, et al.A Weather-Based Hybrid Method for 1-Day Ahead how Forecasting of PV Power Output [ J ]. IEEE Transactions on Sustainable Energy,2014,5(3):917-926.) propose a Weather type identification Based Hourly photovoltaic Output prediction Hybrid model; a Photovoltaic output comprehensive prediction model Based on Environmental Factors and genetic SVM is proposed by J.D.Wang et al (Wang JD, Ran R, Song ZL, et al, short-Term Photoolic Power Generation relating to basic on Environmental Factors and GA-SVM [ J ]. Journal of electric Engineering & Technology,2017,12(1): 64-71.).
The current photovoltaic power prediction algorithm has two main defects, namely, the accuracy of photovoltaic power prediction cannot be guaranteed, the output power of a photovoltaic power generation system is closely related to the solar irradiation intensity, but the irradiation intensity is influenced by the shielding of a cloud layer, so that the fluctuation is obvious, the weather conditions are different, the intensity of fluctuation is different, and the randomness of fluctuation seriously influences the accuracy of photovoltaic power prediction; secondly, the prediction time scale is long, and real-time prediction is difficult to realize.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects of the prior art, and provide a method for predicting the generated power of a photovoltaic power station.
In order to solve the problems, the technical scheme adopted by the invention is as follows: a method for predicting the generated power of a photovoltaic power station comprises the following steps:
1) collecting historical meteorological data and power data of a photovoltaic power station, extracting six meteorological features by taking a day as a unit, and establishing a meteorological feature library;
2) clustering the solar weather feature data in the weather feature library by a fuzzy C-mean kernel clustering algorithm to realize weather type classification, and performing category marking on the power data and the weather data of each day on the basis;
3) establishing a regression sub-model of a support vector machine according to the category marks of the step 2) and the power data and the meteorological data in each category;
4) identifying the weather type of the target day by using a support vector machine algorithm according to the weather characteristics of the target day provided by the numerical weather forecast data, and selecting a support vector machine regression sub-model corresponding to the target day;
5) establishing an autoregressive integral sliding average model by utilizing real-time monitoring data of a target day, and realizing real-time prediction of irradiation intensity and air temperature by using a rolling prediction mode;
6) inputting the predicted values of the irradiation intensity and the air temperature in the step 5) into the support vector machine regression sub-model selected in the step 4) to obtain a final photovoltaic power station power prediction result.
The historical meteorological data in the step 1) are from a comprehensive monitoring system in a photovoltaic power station, and the system comprises an independent meteorological station for collecting real-time meteorological data; the meteorological data comprises real-time irradiation intensity and temperature; the power data is derived from a photovoltaic power plant metering system.
The six meteorological features described above are: [ IR ]max，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]，
Wherein the content of the first and second substances,
IRmax＝max(IRi) Maximum irradiance, Tmax＝max(Ti) Is the highest air temperature，DIFFIRmax＝max(DIFFIRi) As maximum fluctuation value, STDIRFor fluctuating standard deviation, MVIRTo the mean value of fluctuation, TDIRmaxFor the third derivative of the fluctuation, IRiAnd TiIs the ith real-time irradiation intensity and temperature data in the calendar history data,
DIFFIRi＝IRi+1-IRi i＝1,2,...,s-1
s is the number of sample points.
The specific process of fuzzy C-means kernel clustering in the step 2) is as follows:
(2-1) preparing data samples for the first clustering: x is the number ofk＝[IRmax，Tmax，DIFFIRmax，TDIR]k k＝1,2,3,…,n，
n is the number of samples;
(2-2) an initial cluster number C ═ 2;
(2-3) random initialization of clustering center vi；
(2-4) calculating the membership coefficient uik；
Figure BDA0001463882380000031
Wherein, K (x)i,xj) Is a function of a gaussian kernel, which is,
Figure BDA0001463882380000032
delta is a kernel function coefficient, and m is 2;
(2-5) calculating a new cluster center:
Figure BDA0001463882380000033
Figure BDA0001463882380000034
represents uikSquare of (d);
(2-6) circulating the step (2-3) to the step (2-5), and if the termination condition is reached, terminating the circulation;
(2-7) calculating a clustering validity coefficient V when C is 2XB(C)；
Figure BDA0001463882380000035
(2-8) if C is equal to C +1, and cyclically executing the steps (2-3) to (2-7) until C is equal to CmaxTo obtain VXB(C+1)，CmaxIs the clustering times;
(2-9) judging the optimal clustering number CoptMinimum VXB(C) Corresponding C is Copt；
(2-10) with C ═ CoptRe-executing the steps (2-3) to (2-6) to obtain the optimal clustering result, and sampling the sample xkCarrying out category marking;
(2-11) quadratic clustering: on the basis of first clustering, respectively using x to cluster several kinds of datak'＝[MVIR，STDIR]k'And k ' is 1,2,3, …, n ' is a sample number, and n ' is the sample number, and then the fuzzy C-mean kernel clustering algorithm is used for secondary clustering.
In the step 3), the input of the regression submodel of the support vector machine is the irradiation intensity IRiAnd temperature TiOutput as instantaneous power Pi。
In the foregoing step 4), the numerical weather forecast data is from a weather service provider.
In the foregoing step 4), the process of identifying the weather type of the target day by using the support vector machine algorithm is as follows: training a support vector machine model, inputting training data into a meteorological feature library in the step 1), wherein the meteorological feature library comprises six features
[IRmax，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]Outputting the weather type in the step 2); in the identification process, the characteristics of the target day [ IRmax，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]tarInput to a Multi-class support vectorAnd modeling the model to obtain the weather type of the target day.
In the step 5), the rolling prediction mode is to respectively establish two autoregressive integral moving average model time sequences by using a plurality of real-time monitoring data arrays of the irradiation intensity and the temperature of the target day, and respectively predict the irradiation intensity and the temperature value at the next moment; in the next cycle, the actually measured irradiation intensity and temperature data are used for respectively correcting the two autoregressive integral moving average model time sequences, and the irradiation intensity and the temperature value at the next moment are respectively predicted; in this loop, the rolling prediction is performed.
Compared with the closest prior art, the invention has the following beneficial effects:
1. historical meteorological data and output data monitored by a meteorological station of a photovoltaic power station are close to actual conditions on site;
2. the invention adopts two-layer clustering models, and different characteristics are respectively adopted for clustering, so that the classification is more detailed and reasonable, and scientific basis is provided for the selection of the sub-models;
3. according to the invention, the SVR submodel based on the weather type is adopted for regression prediction, other interference factors are eliminated, and the prediction precision is improved;
4. the method adopts an ARIMA time sequence model, and realizes the real-time prediction of the 5-minute time interval through the real-time data provided by the monitoring system of the meteorological station of the photovoltaic power station.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a KFCM modeling flow chart;
FIG. 3 shows example VXBIndex change trend graph; FIG. 3(a) shows the index change after the first layer clustering; FIG. 3(B) shows the A + B index change after the second layer clustering; FIG. 3(C) is the C + D index change after the second level clustering;
FIG. 4 shows the prediction results of the irradiation intensity ARIMA time series in the examples; FIG. 4(a) shows the predicted results of day 19/5, FIG. 4(b) shows the predicted results of day 7/6, FIG. 4(c) shows the predicted results of day 21/7, and FIG. 4(d) shows the predicted results of day 22/8;
FIG. 5 shows the prediction results of the temperature ARIMA time series in the examples; FIG. 5(a) shows the predicted result on day 19/5, FIG. 5(b) shows the predicted result on day 7/6, FIG. 5(c) shows the predicted result on day 21/7, and FIG. 5(d) shows the predicted result on day 22/8;
FIG. 6 is a power SVR model prediction result in an embodiment; fig. 6(a) shows the prediction results for day 19/5, fig. 6(b) shows the prediction results for day 7/6, fig. 6(c) shows the prediction results for day 21/7, and fig. 6(d) shows the prediction results for day 22/8.
Detailed Description
The invention is further described below. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
The invention provides a method for predicting the generated power of a photovoltaic power station, which comprises the following steps as shown in figure 1:
the method comprises the steps of firstly, extracting six meteorological features by taking days as units by utilizing historical meteorological data of a photovoltaic power station, and establishing a meteorological feature library.
The classification regression modeling has proved to have better precision compared with the integral model, and has the advantages of eliminating the interference of unknown factors in other weather conditions on the regression model and realizing the refinement of the data model. The historical meteorological data is from a comprehensive monitoring system in the photovoltaic power station, the meteorological data collected by the independent meteorological station is included, and the real-time irradiation intensity and temperature data collected by the photovoltaic meteorological station and the power data provided by the photovoltaic power station metering system are mainly used in the invention. The irradiation intensity and temperature are the two factors that have the greatest influence on the photovoltaic output. Meanwhile, the fluctuation of the irradiation intensity is also a main factor influencing the accuracy of the prediction model. The significance of the weather type classification is that the weather types with approximate fluctuation conditions are classified into one category, so that an independent regression sub-model is established to eliminate interference and improve prediction accuracy. Based on the above, the six meteorological features selected by the invention are as follows: [ IR ]max，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]Specifically, the following are defined:
(1) maximum irradiationDegree IRmax＝max(IRi)；
(2) Maximum air temperature Tmax＝max(Ti)；
These two indices reflect the maximum instantaneous power values that the photovoltaic module can emit, where IRiAnd TiIs the ith irradiation intensity and temperature data in the calendar history data;
(3) maximum fluctuation value DIFFIRmax＝max(DIFFIRi)；
For discrete data with a constant sampling rate, a first-order difference DIFF is usually adoptedIRInstead of the first derivative;
DIFFIRi＝IRi+1-IRi(i＝1,2,...,s-1) (1)
s is the number of sampling points;
(4) standard deviation of fluctuation STDIR：DIFFIRiThe standard deviation of (a), reflecting the intensity of irradiance fluctuation;
(5) fluctuating mean MVIR：DIFFIRiThe mean value of (a), reflecting the overall amplitude of irradiance fluctuation;
(6) fluctuating third derivative TDIRmax: the third derivative is more sensitive to rapid changes in weather conditions than the other orders.
And secondly, clustering the solar weather feature data in the weather feature library by a fuzzy C-mean value kernel clustering (KFCM) algorithm to realize weather type classification, and carrying out category marking on the power data and the weather data of each day on the basis, wherein the weather data refers to the irradiation intensity and the temperature collected by the photovoltaic weather station, and the power data refers to the output power of the photovoltaic power station.
Considering practical application, the finer the model classification is, the fewer interference factors are; however, too many classifications will seriously affect the computation efficiency and are not good for real-time performance. Therefore, the invention adopts a layered KFCM clustering mode to cluster different features on different layers. Specifically, first clustering is performed according to the maximum value feature, and then secondary clustering is performed on the primary clustering result by using the remaining features respectively. As shown in fig. 2, the weather feature data in the weather feature library is clustered through a KFCM algorithm to realize weather type classification, and on the basis, the power data and weather data of each day are subjected to category marking, and two-layer clustering of KFCM clustering is as follows:
(1) according to maximum value characteristic [ IRmax，Tmax，DIFFIRmax，TDIRmax]Performing first clustering;
(2) the initial clustering result will again use the remaining features STDIR，MVIR]Secondary clustering is performed.
The KFCM two-layer clustering process comprises the following steps:
(1) preparing data samples x for first clusteringk＝[IRmax，Tmax，DIFFIRmax，TDIR]kk is 1,2,3, …, n, n is the number of samples;
(2) the initial clustering number C ═ 2;
(3) randomly initializing a clustering center vi；
(4) Calculating the membership coefficient uik；
Figure BDA0001463882380000061
K is a gaussian kernel, δ is a kernel coefficient, m is typically taken to be 2:
Figure BDA0001463882380000062
(5) calculating a new cluster center:
Figure BDA0001463882380000071
since m is taken to be 2, m,
Figure BDA0001463882380000072
represents uikSquare of (d).
(6) The step (3) to the step (5) are circulated, and the termination condition is judged, and two types are generally available: if the circulation reaches a certain number of times or the difference between the distances between the clustering centers calculated in the previous and subsequent times is smaller than a set threshold value, the circulation is terminated;
(7) calculating a cluster validity coefficient V when C is 2XB(C)；
Figure BDA0001463882380000073
(8) And C +1, and executing the steps (3) to (7) in a loop until C is equal to CmaxTo obtain VXB(C+1)，CmaxIs the clustering times;
(9) judging the optimal clustering number CoptMinimum VXB(C) Corresponding C is Copt；
(10) By C ═ CoptRe-executing the steps (3) to (6) to obtain the optimal clustering result, and sampling the sample xkCarrying out category marking;
(11) secondary clustering: on the basis of first clustering, respectively using x to cluster several kinds of datak'＝[MVIR，STDIR]k'And k ' is 1,2,3, …, n ' is a sample number, and n ' is the sample number, and the KFCM algorithm is used for carrying out second clustering.
And step three, establishing a Support Vector Regression (SVR) sub-model according to the category labels and the power data and the meteorological data in each category.
And D, according to the clustering result in the step two, dividing the historical data into a plurality of classes, wherein each class can establish an SVR sub-model. Specifically, a SVR sub-model is established by power data and meteorological data in each class, the class mark of the SVR sub-model is derived from the result of KFCM two-layer clustering, a plurality of trained SVR sub-models are input as irradiation intensity IRiAnd temperature TiOutput as instantaneous power PiAnd the established model is marked as SVR I, SVR II, … SVR N, and N is the final category number of the cluster.
SVR is a mode of SVM for regression modeling, and the core idea is to construct a classification hyperplane and ensure that the distance between a sample closest to the hyperplane and the hyperplane is maximum. SVMFor two classification problems: (x)i,yi)，i＝1,2,…,l，xi∈Rn，yiE { -1, +1}, are classified into two categories completely without error by the optimal classification hyperplane w · x + b ═ 0. Thus, constructing an optimal hyperplane problem can translate into an optimization problem:
Figure BDA0001463882380000074
the constraint conditions are as follows:
yi((w·xi)+b)≥1-ξi,i＝1,2,…,l (7)
w is the optimal hyperplane normal vector, b is the threshold, M is the penalty parameter, xiiFor the relaxation variable, l is the number of points, and l points are divided into two categories, (x)i,yi) Representing a point on a plane coordinate.
The problem can be solved using the lagrange multiplier method. If the problem is extended to the non-linearity problem, the samples in the low-dimensional space can be mapped into the high-dimensional space by using the mapping phi (x), and then the objective function is:
Figure BDA0001463882380000081
αiis a lagrange multiplier; the kernel function selects a gaussian kernel function.
And step four, identifying the weather type of the target day by using an SVM algorithm according to the weather characteristics of the target day provided by the numerical weather forecast data, and selecting a corresponding SVR sub-model.
The SVR submodel selection is carried out on the premise that the weather type of the target day needs to be judged, the weather type of the target day is identified through the weather characteristics of the target day provided by the numerical weather forecast data by using an SVM algorithm, the corresponding SVR submodel is selected, and the weather characteristics of the target day are consistent with the calculation method of the six characteristics in the step one. Numeric weather forecast data originates from a weather service provider.
The SVM modeling in the invention is also established through characteristic data and training dataThe input of (a) is a meteorological feature library in step one, which contains six features [ IRmax，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax](ii) a And outputting the category label in the step two. In the identification process, the characteristics of the target day [ IRmax，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]tarAnd inputting the multi-classification SVM model to obtain the weather type of the target day.
And fifthly, establishing an autoregressive integrated moving average model (ARIMA) by utilizing the real-time monitoring data of the target day, and realizing the real-time prediction of the irradiation intensity and the air temperature by using a rolling prediction mode.
The hybrid prediction model provided by the invention is used for predicting the temperature and the irradiation intensity in real time through an ARIMA time sequence and realizing instantaneous power regression by utilizing an SVR sub-model. Establishing an ARIMA model by utilizing real-time monitoring data of a target day, and realizing real-time prediction of the irradiation intensity and the air temperature by using a rolling prediction mode, wherein the ARIMA model is represented as ARIMA (p, q, d), p is an autoregressive order, q is a moving average order, and d is a differential processing order. The method comprises the following basic steps:
(1) differential processing, original time series [ X ]t]Stationary time series data [ XA ] obtained by d-order difference processingt]. The time series in the present invention includes irradiation intensity series [ X ]t-E]And temperature sequence [ X ]t-T]。
(2) Model identification and parameter order determination, autocorrelation and partial correlation functions of the data samples are calculated, and model types (AR, MA and ARMA) are preliminarily judged. In general, the expression of the established ARIMA model is as follows:
Figure BDA0001463882380000082
wherein, aiIs an autoregressive coefficient, bjIs a moving average coefficient. e.g. of the typet-jIs a white noise sequence, is an independent error. The values of p and q are defined using the Akage Information Criterion (AIC) information criterion.
(3) Parameter estimationMeter, using correlation moments for aiAnd bjFinally obtaining the ARIMA (p, q, d) model.
(4) And data prediction, namely realizing single-step prediction by an ARIMA model.
Further, the rolling prediction mode is to utilize a plurality of irradiation intensity and temperature real-time monitoring data arrays (IR) of the target daytAnd TtAnd t is 1,2, …, g and g are sequence sample numbers) respectively establishing two ARIMA time sequences, and respectively predicting the irradiation intensity and the temperature value (recorded as IR ') at the next moment'n+1And T'n+1) (ii) a The measured irradiation intensity and temperature data IR are reused in the next cyclen+1And Tn+1Correcting two ARIMA time sequences respectively, and predicting the irradiation intensity and temperature value (recorded as
Figure BDA0001463882380000091
And
Figure BDA0001463882380000092
). In this loop, the rolling prediction is performed.
And step six, inputting the predicted values of the irradiation intensity and the air temperature into the SVR submodel selected in the step four to obtain a final power prediction result of the photovoltaic power station.
Examples
In the invention, 31397 effective samples are adopted in a certain photovoltaic power station in Wujiang province, Suzhou city, Jiangsu province, from 2016 to 2017, within 2 months, a photovoltaic output mixed prediction model is established, and four typical meteorological states are selected: clear (7 months and 21 days), cloudy (5 months and 19 days), sleet (6 months and 7 days) and cloudy (8 months and 22 days) are used as test samples (586) to test the modeling effect, and the rest data samples are used as training samples, which are 30811 in total.
Building of photovoltaic historical meteorological feature library and reasonable clustering of KFCM weather types
The modeling dataset will extract daily meteorological features according to the method of step one, and the effective feature set in the invention comprises 261 samples, which represents 261 days.
Establishing scores according to the KFCM modeling and optimizing process in the step twoAnd (3) a layer clustering model, wherein the clustering number of the two layers of models is 2-10. VXBThe trend is shown in fig. 3. It can be seen that after each layer is clustered, VXB(2) The value of (c) is minimal. Therefore, according to the second step, the optimal cluster number of the two-layer cluster can be considered to be 2, and all samples are finally classified into 4 types, which are marked as a type a, a type B, a type C and a type D. The classification results are shown in table 1.
TABLE 1 KFCM clustering results
Figure BDA0001463882380000093
Second, SVM modeling and target solar weather type identification
By clustering, 261 samples were classified into 4 classes. In order to establish an SVM weather type recognition model and test the accuracy of the model, four types of samples in the feature set are used for establishing a multi-classification SVM model, 70% of the four types of samples are selected as training samples, the rest 30% of the four types of samples are used as test samples, and optimal parameters are obtained through a cross-validation method. The training data is 261 feature samples, and the output is a class label. The results of the weather type test are shown in table 2.
TABLE 2 weather type recognition results of SVM model
Figure BDA0001463882380000101
Of the total 78 test samples, 4 were misclassified, with a classification accuracy of 94.87%. In addition, only the B-type samples in the four types of samples are classified by mistake, which shows that the identification precision of the SVM classifier is high, and the established weather type identification model can accurately identify the weather type of the forecast day and select a proper SVR regression sub-model.
Three, ARIMA time series modeling and SVM regression prediction
As mentioned above, according to the results of KFCM, the irradiation intensity, temperature and power data samples for 261 days are also classified into four categories a, B, C and D, and are respectively given category labels. Building samples in the same classOne SVR regression submodel to obtain four submodels SUB-A, SUB-B, SUB-C, SUB-D with input of irradiation intensity IRiTemperature TiOutput as the instantaneous power P of the corresponding timei. And obtaining the optimal parameters of the SVR model by adopting a cross validation method.
Subsequently, an identification of the predicted daily weather type is made. And inputting the meteorological information of the predicted day into the SVM multi-classification model to obtain the category attribute of the predicted day. The prediction days selected in the present invention are: (7-month-21), cloudy (5-month-19), sleet (6-month-7), cloudy (8-month-22), and the classification attributes obtained by inputting the SVM models are as follows: d, B, C and B. The SUB-models selected correspondingly are SUB-D, SUB-B, SUB-C, SUB-B.
To achieve mixed real-time prediction, IR is established as described in step fivei,TiAnd performing rolling prediction on the ARIMA time series model. The predicted 20 day ahead IR is selected according to the requirements of time series modeling and engineering applicationsi,TiAnd (i is 1-20) respectively establishing two ARIMA models by data, wherein the data is derived from data provided by an online monitoring system for predicting days, and the sampling interval is 5 min. Taking day 21/7 as an example, the first monitoring data appeared from 6 o' clock 15. Collect 20 site monitoring IRi,TiThe ARIMA modeling and prediction work is started at 7 points and 55 time sharing to respectively obtain the ARIMAIRAnd ARIMAT. Subsequently, the IR 'of the next sample point (after 5min, i.e. 8 points) was predicted'i+1,T′i+1. The predicted value is input into the submodel SUB-D to obtain the predicted instantaneous power P. Third, 8 o' clock, new actual monitoring data IRi+1,Ti+1Can be obtained from a monitoring system and used for monitoring ARIMAIRAnd ARIMATAnd carrying out real-time correction. Then predict the IR of the next step (8 points and 5 points)i+2,Ti+2And P. The instantaneous power can be predicted in real time all day by the aid of the circular rolling. Final four predicted days ARIMA model IRi,TiThe prediction results of (2) are shown in FIG. 4 and FIG. 5, and the regression results of the SVR submodel are shown in FIG. 6. Using mean absolute percentage error εMAPEAnd the root mean square error εRMSEAs an index of prediction accuracy, IRi,TiAnd P prediction accuracy are shown in table 3.
Figure BDA0001463882380000111
Figure BDA0001463882380000112
TABLE 3 prediction accuracy of hybrid prediction model
Figure BDA0001463882380000113
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.

Claims (6)

1. A method for predicting the generated power of a photovoltaic power station is characterized by comprising the following steps:
1) collecting historical meteorological data and power data of a photovoltaic power station, extracting six meteorological features by taking a day as a unit, and establishing a meteorological feature library;
the six meteorological features are as follows: [ IR ]max，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]，
Wherein the content of the first and second substances,
IRmax＝max(IRi) Maximum irradiance, Tmax＝max(Ti) At the maximum air temperature, DIFFIRmax＝max(DIFFIRi) As maximum fluctuation value, STDIRFor fluctuating standard deviation, MVIRTo the mean value of fluctuation, TDIRmaxFor the third derivative of the fluctuation, IRiAnd TiIs the ith real-time irradiation intensity and temperature data in the calendar history data,
DIFFIRi＝IRi+1-IRi i＝1,2,...,s-1
s is the number of sampling points;
2) clustering the solar weather feature data in the weather feature library by a fuzzy C-mean kernel clustering algorithm to realize weather type classification, and performing category marking on the power data and the weather data of each day on the basis;
the specific flow of the fuzzy C-means kernel clustering is as follows:
(2-1) preparing data samples for the first clustering: x is the number ofk＝[IRmax，Tmax，DIFFIRmax，TDIR]kk is 1,2,3, …, n, n is the number of samples;
(2-2) an initial cluster number C ═ 2;
(2-3) random initialization of clustering center vi；
(2-4) calculating the membership coefficient uik；
Figure FDA0002945478770000011
Wherein, K (x)i,xj) Is a function of a gaussian kernel, which is,
Figure FDA0002945478770000012
delta is a kernel function coefficient, and m is 2;
(2-5) calculating a new cluster center:
Figure FDA0002945478770000013
Figure FDA0002945478770000014
represents uikSquare of (d);
(2-6) circulating the step (2-3) to the step (2-5), and if the termination condition is reached, terminating the circulation;
(2-7) calculating a clustering validity coefficient V when C is 2XB(C)；
Figure FDA0002945478770000021
(2-8) if C is equal to C +1, and cyclically executing the steps (2-3) to (2-7) until C is equal to CmaxTo obtain VXB(C+1)，CmaxIs the clustering times;
(2-9) judging the optimal clustering number CoptMinimum VXB(C) Corresponding C is Copt；
(2-10) with C ═ CoptRe-executing the steps (2-3) to (2-6) to obtain the optimal clustering result, and sampling the sample xkCarrying out category marking;
(2-11) quadratic clustering: on the basis of first clustering, respectively using x to cluster several kinds of datak'＝[MVIR，STDIR]k'k ' is 1,2,3, …, n ' is a sample, n ' is the number of samples, and then the fuzzy C-mean kernel clustering algorithm is used for secondary clustering;
3) establishing a regression sub-model of a support vector machine according to the category marks of the step 2) and the power data and the meteorological data in each category;
4) identifying the weather type of the target day by using a support vector machine algorithm according to the weather characteristics of the target day provided by the numerical weather forecast data, and selecting a support vector machine regression sub-model corresponding to the target day;
5) establishing an autoregressive integral sliding average model by utilizing real-time monitoring data of a target day, and realizing real-time prediction of irradiation intensity and air temperature by using a rolling prediction mode;
6) inputting the predicted values of the irradiation intensity and the air temperature in the step 5) into the support vector machine regression sub-model selected in the step 4) to obtain a final photovoltaic power station power prediction result.
2. The method for predicting the power generation capacity of the photovoltaic power station as claimed in claim 1, wherein the historical meteorological data in the step 1) is derived from an integrated monitoring system in the photovoltaic power station, and the method comprises the steps of collecting real-time meteorological data by an independent meteorological station; the meteorological data comprises real-time irradiation intensity and temperature; the power data is derived from a photovoltaic power plant metering system.
3. The method for predicting the generated power of the photovoltaic power plant as claimed in claim 1, wherein in the step 3), the input of the regression submodel of the support vector machine is the irradiation intensity IRiAnd temperature TiOutput as instantaneous power Pi。
4. The method for predicting the generated power of the photovoltaic power plant of claim 1, wherein in the step 4), the numerical weather forecast data is from a weather service provider.
5. The method for predicting the power generation capacity of the photovoltaic power station as claimed in claim 4, wherein in the step 4), the process of identifying the weather type of the target day by using the support vector machine algorithm comprises the following steps: training a support vector machine model, inputting training data into a meteorological feature library in the step 1), wherein the meteorological feature library comprises six features [ IR ]max，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]Outputting the weather type in the step 2); in the identification process, the characteristics of the target day [ IRmax，Tmax，DIFFIRmax，STDIR，MVIR，TDIRmax]tarAnd inputting the weather pattern into a multi-classification support vector machine model to obtain the weather type of the target day.
6. The method for predicting the generated power of the photovoltaic power station as claimed in claim 1, wherein in the step 5), the rolling prediction mode is that two autoregressive integral moving average model time sequences are respectively established by using a plurality of real-time monitoring data arrays of the irradiation intensity and the temperature of the target day, and the irradiation intensity and the temperature value at the next moment are respectively predicted; respectively correcting the time sequences of the two autoregressive integral sliding average models by using actually measured irradiation intensity and temperature data in the next cycle, and respectively predicting the irradiation intensity and the temperature value at the next moment; in this loop, the rolling prediction is performed.
CN201711103682.9A 2017-11-10 2017-11-10 Method for predicting power generation power of photovoltaic power station Active CN107766990B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711103682.9A CN107766990B (en) 2017-11-10 2017-11-10 Method for predicting power generation power of photovoltaic power station

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711103682.9A CN107766990B (en) 2017-11-10 2017-11-10 Method for predicting power generation power of photovoltaic power station

Publications (2)

Publication Number Publication Date
CN107766990A CN107766990A (en) 2018-03-06
CN107766990B true CN107766990B (en) 2021-06-08

Family

ID=61273364

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711103682.9A Active CN107766990B (en) 2017-11-10 2017-11-10 Method for predicting power generation power of photovoltaic power station

Country Status (1)

Country Link
CN (1) CN107766990B (en)

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109102102A (en) * 2018-06-25 2018-12-28 南京邮电大学 Based on the photovoltaic of multivariate phase space reconstruction and SVR power output short term prediction method
CN108985513A (en) * 2018-07-17 2018-12-11 合肥工业大学 Industrial enterprise's Electric Load Prediction System
CN109388863B (en) * 2018-09-21 2023-05-26 国网天津市电力公司 ARIMA model-based distributed photovoltaic output power prediction method
CN109685257A (en) * 2018-12-13 2019-04-26 国网青海省电力公司 A kind of photovoltaic power generation power prediction method based on Support vector regression
CN109919353B (en) * 2019-01-11 2022-07-05 杭州电子科技大学 Distributed photovoltaic prediction method of ARIMA model based on spatial correlation
CN110009385B (en) * 2019-01-11 2023-12-12 国网浙江杭州市临安区供电有限公司 Multi-factor mapping-based photovoltaic power generation user group division method
CN109871996A (en) * 2019-02-14 2019-06-11 浙江中控太阳能技术有限公司 A kind of tower photo-thermal power station mirror field cloud monitoring System and method for based on photovoltaic panel
CN110059441B (en) * 2019-04-29 2022-07-19 广东电网有限责任公司 Photovoltaic power station modeling method and photovoltaic power station model output correction method
CN110289605B (en) * 2019-06-27 2021-12-10 昆明理工大学 Intelligent direction protection method for hybrid compensation circuit based on instantaneous power curve cluster
CN111079980B (en) * 2019-11-22 2021-06-29 天合云能源互联网技术(杭州)有限公司 Optical power prediction method based on self-adaptive classification strategy and hybrid optimization SVR
CN110929953A (en) * 2019-12-04 2020-03-27 国网山东省电力公司电力科学研究院 Photovoltaic power station ultra-short term output prediction method based on cluster analysis
CN111738327B (en) * 2020-06-18 2022-09-13 河海大学常州校区 Ultrashort-term irradiation prediction method based on typical cloud shielding irradiation difference
CN111967660B (en) * 2020-08-05 2022-09-23 河海大学 Ultra-short-term photovoltaic prediction residual error correction method based on SVR
CN112633572A (en) * 2020-12-21 2021-04-09 国网甘肃省电力公司电力科学研究院 Day classification dual-model photovoltaic power generation combined prediction method based on prediction meteorological data
CN112987802B (en) * 2021-02-22 2022-08-09 山东理工大学 Photovoltaic power generation method and device, electronic equipment and storage medium
CN113240153A (en) * 2021-04-09 2021-08-10 水发兴业能源(珠海)有限公司 Photovoltaic power generation data prediction method and device, computing equipment and storage medium
CN113033910B (en) * 2021-04-09 2023-06-09 电子科技大学中山学院 Photovoltaic power generation power prediction method, storage medium and terminal equipment
CN115632396A (en) * 2022-11-02 2023-01-20 广州瑞鑫智能制造有限公司 Photovoltaic energy storage system for air compression station
CN116109007B (en) * 2023-03-23 2023-07-11 阿里巴巴达摩院(杭州)科技有限公司 Power generation power determination method, server and storage medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106649919A (en) * 2016-09-13 2017-05-10 上海电气电站环保工程有限公司 Method and system for predicting carbon content of fly ash in coal-fired power plant boiler

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106649919A (en) * 2016-09-13 2017-05-10 上海电气电站环保工程有限公司 Method and system for predicting carbon content of fly ash in coal-fired power plant boiler

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
"基于ARIMA和SVR的光伏电站超短期功率预测";赫卫国等;《广东电力》;20170825;第30卷(第8期);第32-37页 *
"基于模糊C- 均值聚类的支持向量机";彭璐等;《工业控制计算机》;20061125;第19卷(第11期);第43-44页 *

Also Published As

Publication number Publication date
CN107766990A (en) 2018-03-06

Similar Documents

Publication Publication Date Title
CN107766990B (en) Method for predicting power generation power of photovoltaic power station
CN109002915B (en) Photovoltaic power station short-term power prediction method based on Kmeans-GRA-Elman model
CN106779223B (en) Photovoltaic system power generation real-time prediction method and device
CN106529814B (en) Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain
CN110909919A (en) Photovoltaic power prediction method of depth neural network model with attention mechanism fused
CN110619360A (en) Ultra-short-term wind power prediction method considering historical sample similarity
Yang et al. Day-ahead forecasting of photovoltaic output power with similar cloud space fusion based on incomplete historical data mining
Melzi et al. Hourly solar irradiance forecasting based on machine learning models
CN112257941A (en) Photovoltaic power station short-term power prediction method based on improved Bi-LSTM
CN107169683A (en) A kind of grid-connected photovoltaic power station short term power combination forecasting method of variable weight coefficient
CN105701572B (en) Photovoltaic short-term output prediction method based on improved Gaussian process regression
CN105069521A (en) Photovoltaic power plant output power prediction method based on weighted FCM clustering algorithm
CN112186761B (en) Wind power scene generation method and system based on probability distribution
CN111626473A (en) Two-stage photovoltaic power prediction method considering error correction
CN114792156A (en) Photovoltaic output power prediction method and system based on curve characteristic index clustering
Li et al. Deep spatio-temporal wind power forecasting
CN110717610A (en) Wind power prediction method based on data mining
CN112215428A (en) Photovoltaic power generation power prediction method and system based on error correction and fuzzy logic
CN114169445A (en) Day-ahead photovoltaic power prediction method, device and system based on CAE and GAN hybrid network
CN114399081A (en) Photovoltaic power generation power prediction method based on weather classification
CN110852492A (en) Photovoltaic power ultra-short-term prediction method for finding similarity based on Mahalanobis distance
CN116167508B (en) Short-term photovoltaic output rapid prediction method and system based on meteorological factor decomposition
CN114234392B (en) Air conditioner load fine prediction method based on improved PSO-LSTM
CN110163437A (en) Day-ahead photovoltaic power generation power prediction method based on DPK-means
CN115481788A (en) Load prediction method and system for phase change energy storage system

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
GR01 Patent grant
GR01 Patent grant