CN104915727B - Various dimensions allomer BP neural network optical power ultra-short term prediction method - Google Patents

Various dimensions allomer BP neural network optical power ultra-short term prediction method Download PDF

Info

Publication number
CN104915727B
CN104915727B CN201510268305.5A CN201510268305A CN104915727B CN 104915727 B CN104915727 B CN 104915727B CN 201510268305 A CN201510268305 A CN 201510268305A CN 104915727 B CN104915727 B CN 104915727B
Authority
CN
China
Prior art keywords
data
neural network
meteorological
output
prediction
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
CN201510268305.5A
Other languages
Chinese (zh)
Other versions
CN104915727A (en
Inventor
吴世伟
李靖霞
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.)
Nanjing SAC Automation Co Ltd
Original Assignee
Nanjing SAC Automation Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing SAC Automation Co Ltd filed Critical Nanjing SAC Automation Co Ltd
Priority to CN201510268305.5A priority Critical patent/CN104915727B/en
Publication of CN104915727A publication Critical patent/CN104915727A/en
Application granted granted Critical
Publication of CN104915727B publication Critical patent/CN104915727B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

The invention discloses a kind of various dimensions allomer BP neural network optical power ultra-short term prediction methods, specifically comprise the following steps: that SS1 is comprehensive using grid-connected active metric data, meteorological substation metric data, grid-connected active historical data, meteorological substation historical data and data of weather forecast, the data being daily segmented are analyzed, by meteorological condition, close, active output is similar calculates index of similarity for condition, and sorts out history of forming data sample with this;SS2 is modified according to data of weather forecast, meteorological substation metric data logarithm Weather information;Historical data sample of the SS3 according to revised Numerical Weather information and after sorting out, is matched by the close condition of meteorological condition, selects input training sample of the close sample as artificial neural network;SS4 carries out input data normalization, training sample screening, prediction output for BP neural network;When the prediction of SS5 next period starts, the process of step SS1 to step SS4 is repeated.

Description

Various dimensions allomer BP neural network optical power ultra-short term prediction method
Technical field
The present invention relates to a kind of various dimensions allomer BP neural network optical power ultra-short term prediction methods, belong to photovoltaic hair Electrical power electric powder prediction.
Background technique
Country greatly develops clean energy technology at present, and photovoltaic power generation is important component.Photovoltaic power generation feature is can It regenerates, is pollution-free, being limited in power output and fluctuated with weather conditions, impacting big, especially extensive concentration after grid-connected for power grid Formula photovoltaic plant is more obvious.If energy look-ahead photovoltaic power generation power output, is convenient for dispatching of power netwoks, reasonable arrangement formulates power generation Plan adjusts power output distribution, economic load dispatching etc., therefore accurate forecast photovoltaic generation power is significant.
Optical power prediction technique is divided into mathematical model method, statistical model method, model of mind method etc., respectively there is length.Artificial mind There is good performance in the complicated nonlinear problem of processing through network, as one kind of model of mind method, is predicted in optical power In be widely used.Discovery could obtain good prediction result after weather information is added in current research practice.Light Volt power generation is influenced greatly by environment, and in the season environment that spring, summer, autumn, the four seasons in winter are different, fine, negative, mist, rain, snow etc. are different meteorological Under the conditions of generated output situation difference it is obvious.Due to seasonal variations, the feature of meteorological condition complexity, research and propose accordingly a variety of The Artificial Neural Network of model, and adapting to IFR conditions is the outstanding problem that optical power prediction needs to solve.
Summary of the invention
To solve the above problems, existing document is used by season classification and by the method for weather classification of type, adopt respectively With different neural network models.Various dimensions allomer BP neural network optical power ultra-short term prediction side proposed by the present invention Method takes the method to seasonal factor and the classification of weather environment united analysis, and the training of matched sample library is selected from database, from And avoid and establish multiple neural network models, and transduction factors by meteorological data training obtain, solve generated output with The problem of meteorological condition changes.
The present invention adopts the following technical scheme: a kind of various dimensions allomer BP neural network optical power ultra-short term prediction side Method, which is characterized in that specifically comprise the following steps:
Step SS1 is comprehensive to use grid-connected active metric data, meteorological substation metric data, grid-connected active historical data, gas As substation historical data and data of weather forecast, the data being daily segmented are analyzed, close, the active output by meteorological condition It is similar to calculate index of similarity for condition, and history of forming data sample is sorted out with this;
Step SS2 is modified according to data of weather forecast, meteorological substation metric data logarithm Weather information;
Historical data sample of the step SS3 according to revised Numerical Weather information and after sorting out, is connect by meteorological condition Close condition is matched, and input training sample of the close sample as artificial neural network is selected;
Step SS4 carries out input data normalization, training sample screening, prediction output for BP neural network;
When the prediction of step SS5 next period starts, the process of step SS1 to step SS4 is repeated.
Preferably, step SS1 includes: whether (1) initial data needs to rationalize before use processing, including verification most Big minimum value range, quality of data situation, data integrity;(2) using comprehensive index of similarity, to meteorological condition, active Output assigns different weight factors respectively, and considers most value, mean value, amplitude of variation, growing direction to single condition.
Preferably, step SS2 includes: to be calculated using exponential smoothing, using following formula
Or(t=1,2 ...).
Preferably, step SS3 includes: to calculate characteristic index, and with the meteorological data in 4 hours futures with the index It is matched with sample data, finds out close sample.
Preferably, the construction method of BP neural network described in step SS4 are as follows: in the prediction of a period, selection has Function output, irradiation level, temperature, humidity are as input;The BP neural network structure includes input layer, middle layer, output layer, institute State input layer by 3 similar days, the same day is meteorological, 1 hour data was constituted before the same day, set as unit of hour, be spaced 15 minutes Acquisition, the active output of each similar day, irradiation level, temperature, humidity have the 4X4=16 factor altogether, the irradiation level of same day meteorology, Temperature, humidity have the 3X4=12 factor altogether, and active output in 1 hour, irradiation level, temperature, humidity had 4X4=16 altogether before the same day A factor, above-mentioned 76 factors constitute input layer;The middle layer is hidden layer;The output layer is used to export prediction 1 small When active power, including 4 points.
Preferably, normalization described in step SS4 uses following steps: input value must be normalized, and input is turned It is changed between [0,1], power input conversion formula is
The active output conversion formula is accordingly
Pout=(Pmax-Pmin)p+Pmin
The irradiation level is converted in the same manner;
The temperature is according to section band normalized;
Percentage of the humidity value between 0-100, can be converted directly into decimal.
Preferably, the screening of training sample described in step SS4 uses following steps: measuring sample using overall target Difference, then Screening Samples on this basis;Measure the index of power, P=k1*X1+k2*X2+k3*X3, wherein k1, k2, k3 For weight coefficient,
Cardinal direction marker X1, for indicating that data trend is ascending, descending or turn;
Maximin index X2, for the maximin absolute difference with reference day;
Absolute error mean value specification X3, for the absolute error mean value with reference day;
It is similar to measure meteorological index;
Using the numerical value after normalization in calculating.
Preferably, the output of prediction described in step SS4 includes the following steps: positive calculating and backpropagation,
The forward direction calculating process are as follows:
Node j is inputted(j=1,2 ..., M)
Node j reality output is Oj=fj(NETj)
The back-propagation process are as follows:
The output layer σj=Oj(1-Oj)(Tj-Oj), wherein T is desired output;
The hidden layer
The variation delta W of weightij=η σjOi, (adjacent node that i is j)
Wherein, error functionUsing steepest descent method, it is desirable that set up Δ Wij=η σjOi, warp It crosses and repeatedly cycles to reach stop condition.
One, advantageous effects of the invention: (1) solves the problems, such as that forecasted variances are big under a variety of meteorological conditions, and Avoid the process that multiple models are established by season and weather condition;(2) BP neural network used is suitable for more using extensively The optical power forecasting problem of dimension and complex nonlinear is added many factors such as weather information in input and considers Various Complex The combined influence of combination condition;(3) accuracy of weather forecast is corrected, improved to logarithm Weather information;And to a number of days According to segment processing, large-scale fluctuation is decomposed into localised waving, is conducive to the convergence speed for improving artificial neural network; (4) accurate forecast photovoltaic plant changed power can provide the economy that operation of power networks is improved with reference to scientific basis for dispatching of power netwoks And safety.
Detailed description of the invention
Fig. 1 is the flow chart of various dimensions allomer BP neural network optical power ultra-short term prediction method of the invention.
Fig. 2 is BP neural network structure chart of the invention.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.Following embodiment is only used for clearly illustrating the present invention Technical solution, and not intended to limit the protection scope of the present invention.
Fig. 1 is the flow chart of various dimensions allomer BP neural network optical power ultra-short term prediction method of the invention, this Invention proposes a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method, which is characterized in that specifically includes Following 5 steps.
Step SS1 is comprehensive to use grid-connected active metric data, meteorological substation metric data, grid-connected active historical data, gas As substation historical data and data of weather forecast, the data being daily segmented are analyzed, close, the active output by meteorological condition It is similar to calculate index of similarity for condition, and history of forming data sample is sorted out with this;Step SS1 includes: that (1) initial data exists Whether need to rationalize processing, including verification before use in maximin range, quality of data situation, data integrity;(2) Using comprehensive index of similarity, different weight factors is assigned respectively to meteorological condition, active output, and to single condition Consider most value, mean value, amplitude of variation, growing direction.
Step SS2 is modified according to data of weather forecast, meteorological substation metric data logarithm Weather information;Using referring to Number exponential smoothing is calculated, using following formula
Or(t=1,2 ...).
Historical data sample of the step SS3 according to revised Numerical Weather information and after sorting out, is connect by meteorological condition Close condition is matched, and input training sample of the close sample as artificial neural network is selected;In 4 hours following Meteorological data is calculated characteristic index, and is matched with the index with sample data, and close sample is found out.
Step SS4 carries out input data normalization, training sample screening, prediction output for BP neural network:
(1) construction method of BP neural network are as follows: in the prediction of a period, choose active output, irradiation level, temperature Degree, humidity are as input;The BP neural network structure includes input layer, middle layer, output layer, and the input layer is by 3 phases Like day, same day meteorology, 1 hour data was constituted before the same day, was set as unit of hour, is spaced 15 minutes and is acquired, each similar day Active output, irradiation level, temperature, humidity have the 4X4=16 factor altogether, the irradiation level of same day meteorology, temperature, humidity have altogether The 3X4=12 factor, active output in 1 hour before the same day, irradiation level, temperature, humidity have the 4X4=16 factor altogether, and above-mentioned 76 A factor constitutes input layer;The middle layer is hidden layer;The output layer is used to export the active power of prediction 1 hour, Including 4 points, as shown in Fig. 2 BP neural network structure chart of the invention.
(2) normalization uses following steps: input value must be normalized, and convert the input between [0,1], power Inputting conversion formula is
The active output conversion formula is accordingly
Pout=(Pmax-Pmin)p+Pmin
The irradiation level is converted in the same manner;
The temperature is according to section band normalized;
Percentage of the humidity value between 0-100, can be converted directly into decimal.
(3) training sample screening use following steps: the difference of sample is measured using overall target, then as according to According to Screening Samples;Measuring the index of power, P=k1*X1+k2*X2+k3*X3, wherein k1, k2, k3 are weight coefficient,
Cardinal direction marker X1, for indicating that data trend is ascending, descending or turn;
Maximin index X2, for the maximin absolute difference with reference day;
Absolute error mean value specification X3, for the absolute error mean value with reference day;
It is similar to measure meteorological index;
Using the numerical value after normalization in calculating.
(4) prediction output includes the following steps: positive calculating and backpropagation,
The forward direction calculating process are as follows:
Node j is inputted(j=1,2 ..., M)
Node j reality output is Oj=fj(NETj)
The back-propagation process are as follows:
The output layer σj=Oj(1-Oj)(Tj-Oj), wherein T is desired output;
The hidden layer
The variation delta W of weightij=η σjOi, (adjacent node that i is j)
Wherein, error functionUsing steepest descent method, it is desirable that set up Δ Wij=η σjOi, warp It crosses and repeatedly cycles to reach stop condition.
When the prediction of step SS5 next period starts, the process of step SS1 to step SS4 is repeated.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art For member, without departing from the technical principles of the invention, several improvement and deformations can also be made, these improvement and deformations Also it should be regarded as protection scope of the present invention.

Claims (8)

1. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method, which is characterized in that specifically include as Lower step:
Step SS1 is comprehensive to use grid-connected active metric data, meteorological substation metric data, grid-connected active historical data, meteorological son Historical data of standing and data of weather forecast analyze the data being daily segmented, and by meteorological condition, close, active output is similar Index of similarity is calculated for condition, and history of forming data sample is sorted out with this;
Step SS2 is modified according to data of weather forecast, meteorological substation metric data logarithm Weather information;
Historical data sample of the step SS3 according to revised Numerical Weather information and after sorting out, it is close by meteorological condition Condition is matched, and input training sample of the close sample as artificial neural network is selected;
Step SS4 carries out input data normalization, training sample screening, prediction output for BP neural network;
When the prediction of step SS5 next period starts, the process of step SS1 to step SS4 is repeated.
2. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method according to claim 1, It is characterized in that, the step SS1 includes: whether (1) initial data needs to rationalize before use processing, including verification in maximum Minimum value range, quality of data situation, data integrity;(2) using comprehensive index of similarity, to meteorological condition, active defeated It assigns different weight factors respectively out, and most value, mean value, amplitude of variation, growing direction is considered to single condition.
3. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method according to claim 1, It is characterized in that, the step SS2 includes: to be calculated using exponential smoothing, using following formula
Or
4. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method according to claim 1, Be characterized in that, the step SS3 include: characteristic index is calculated with the meteorological data in following 4 hours, and with the index with Sample data is matched, and close sample is found out.
5. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method according to claim 1, It is characterized in that, the construction method of BP neural network described in the step SS4 are as follows: in the prediction of a period, choose active Output, irradiation level, temperature, humidity are as input;The BP neural network structure includes input layer, middle layer, output layer, described Input layer is by 3 similar days, the same day is meteorological, 1 hour data was constituted before the same day, sets as unit of hour, interval is adopted for 15 minutes Collection, the active output of each similar day, irradiation level, temperature, humidity have the 4X4=16 factor, the irradiation level of same day meteorology, temperature altogether Degree, humidity have the 3X4=12 factor altogether, and active output in 1 hour, irradiation level, temperature, humidity had 4X4=16 altogether before the same day The factor, above-mentioned 76 factors constitute input layer;The middle layer is hidden layer;The output layer is used to export prediction 1 hour Active power, including 4 points.
6. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method according to claim 5, It is characterized in that, normalization described in the step SS4 uses following steps: input value must be normalized, and input is converted Between [0,1], power input conversion formula is
The active output conversion formula is accordingly
Pout=(Pmax-Pmin)p+Pmin
The irradiation level is converted in the same manner;
The temperature is according to section band normalized;
Percentage of the humidity value between 0-100, can be converted directly into decimal.
7. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method according to claim 1, It is characterized in that, the screening of training sample described in the step SS4 uses following steps: the difference of sample is measured using overall target It is different, then Screening Samples on this basis;Measure the index of power, P=k1*X1+k2*X2+k3*X3, wherein k1, k2, k3 are Weight coefficient,
Cardinal direction marker X1, for indicating that data trend is ascending, descending or turn;
Maximin index X2, for the maximin absolute difference with reference day;
Absolute error mean value specification X3, for the absolute error mean value with reference day;
Using the numerical value after normalization in calculating.
8. a kind of various dimensions allomer BP neural network optical power ultra-short term prediction method according to claim 5, It being characterized in that, the output of prediction described in the step SS4 includes the following steps: positive calculating and backpropagation,
The forward direction calculating process are as follows:
Node j is inputted
Node j reality output is Oj=fj(NETj)
The back-propagation process are as follows:
The output layer σj=Oj(1-Oj)(Tj-Oj), wherein T is desired output;
The hidden layer
The variation delta W of weightij=η σjOi, wherein i is the adjacent node of j;
Wherein, error functionUsing steepest descent method, it is desirable that set up Δ Wij=η σjOi, through excessive It is secondary to cycle to reach stop condition.
CN201510268305.5A 2015-05-22 2015-05-22 Various dimensions allomer BP neural network optical power ultra-short term prediction method Active CN104915727B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510268305.5A CN104915727B (en) 2015-05-22 2015-05-22 Various dimensions allomer BP neural network optical power ultra-short term prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510268305.5A CN104915727B (en) 2015-05-22 2015-05-22 Various dimensions allomer BP neural network optical power ultra-short term prediction method

Publications (2)

Publication Number Publication Date
CN104915727A CN104915727A (en) 2015-09-16
CN104915727B true CN104915727B (en) 2019-04-16

Family

ID=54084774

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510268305.5A Active CN104915727B (en) 2015-05-22 2015-05-22 Various dimensions allomer BP neural network optical power ultra-short term prediction method

Country Status (1)

Country Link
CN (1) CN104915727B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106934482A (en) * 2016-09-20 2017-07-07 威德大学产学协力团 Using 24 hours electric power demand forecasting devices and electric power demand forecasting method of the temperature for temporally dividing
CN106779175B (en) * 2016-11-28 2020-01-21 西安交通大学 Time-interval-considered nonlinear photovoltaic output prediction method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103218673A (en) * 2013-03-27 2013-07-24 河海大学 Method for predicating short-period output power of photovoltaic power generation based on BP (Back Propagation) neural network
CN104573879A (en) * 2015-01-30 2015-04-29 河海大学 Photovoltaic power station output predicting method based on optimal similar day set
CN104616085A (en) * 2015-02-16 2015-05-13 河海大学常州校区 Photovoltaic generating capacity predicting method based on BP neural network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103218673A (en) * 2013-03-27 2013-07-24 河海大学 Method for predicating short-period output power of photovoltaic power generation based on BP (Back Propagation) neural network
CN104573879A (en) * 2015-01-30 2015-04-29 河海大学 Photovoltaic power station output predicting method based on optimal similar day set
CN104616085A (en) * 2015-02-16 2015-05-13 河海大学常州校区 Photovoltaic generating capacity predicting method based on BP neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
计及天气类型指数的光伏发电短期出力预测;袁晓玲 等;《中国电机工程学报》;20131205;第33卷(第34期);第57-64页

Also Published As

Publication number Publication date
CN104915727A (en) 2015-09-16

Similar Documents

Publication Publication Date Title
CN104573879B (en) Photovoltaic plant based on optimal similar day collection goes out force prediction method
CN105046374B (en) A kind of power interval prediction technique based on core extreme learning machine model
CN105184678A (en) Method for constructing photovoltaic power station generation capacity short-term prediction model based on multiple neural network combinational algorithms
CN103106544B (en) A kind of photovoltaic generation prognoses system based on T-S Fuzzy neutral net
CN107516170B (en) Difference self-healing control method based on equipment failure probability and power grid operation risk
CN106251001A (en) A kind of based on the photovoltaic power Forecasting Methodology improving fuzzy clustering algorithm
CN105069521A (en) Photovoltaic power plant output power prediction method based on weighted FCM clustering algorithm
CN107194495B (en) Photovoltaic power longitudinal prediction method based on historical data mining
CN106022528A (en) Short-term power prediction method of photovoltaic power plant based on density peak hierarchical clustering
CN107516145A (en) A kind of multichannel photovoltaic power generation output forecasting method based on weighted euclidean distance pattern classification
CN103473621A (en) Wind power station short-term power prediction method
CN108376262B (en) Analytical model construction method for typical characteristics of wind power output
CN104915727B (en) Various dimensions allomer BP neural network optical power ultra-short term prediction method
CN111753893A (en) Wind turbine generator power cluster prediction method based on clustering and deep learning
CN106485093A (en) Based on the solar irradiance time series synthetic method for improving Markov chain
CN104376371B (en) A kind of distribution based on topology is layered load forecasting method
CN107944622A (en) Wind power forecasting method based on continuous time cluster
CN106022538B (en) Photovoltaic power generation prediction method based on K-means clustering improved generalized weather
CN109376863A (en) Photovoltaic power based on MIV-BP neural network is classified prediction technique in short term
Meng et al. A new PV generation power prediction model based on GA-BP neural network with artificial classification of history day
Zhang et al. Power forecasting of solar photovoltaic power systems based on similar day and M5'model trees
CN112418346A (en) Numerical weather forecast total radiation system error classification calculation method
CN110110912B (en) Photovoltaic power multi-model interval prediction method
CN106600060A (en) Method for predicting amount of solar radiation based on similar day sunny coefficient correction
CN110991786A (en) 10kV static load model parameter identification method based on similar daily load curve

Legal Events

Date Code Title Description
PB01 Publication
C06 Publication
SE01 Entry into force of request for substantive examination
C10 Entry into substantive examination
GR01 Patent grant
GR01 Patent grant