CN104200274A - Power prediction method for photovoltaic devices - Google Patents
Power prediction method for photovoltaic devices Download PDFInfo
- Publication number
- CN104200274A CN104200274A CN201410118453.4A CN201410118453A CN104200274A CN 104200274 A CN104200274 A CN 104200274A CN 201410118453 A CN201410118453 A CN 201410118453A CN 104200274 A CN104200274 A CN 104200274A
- Authority
- CN
- China
- Prior art keywords
- prediction
- photovoltaic
- data
- day
- neural network
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention relates to an output power prediction method for photovoltaic generating devices; the method comprises: predicting by using a physical model prediction, and a neural network prediction of environmental factors and a rolling a neural network prediction; setting the physical model prediction as P0, the neural network prediction of the environmental factors as P1, the rolling a neural network prediction as P2, and final power prediction output value as p; if P1 is more than P0 and P2 is more than P0, P is equal to P0; if the P1 is more than P0 that is more than P2, P is equal to P2; if the P2 is more than P0 that is more than P1, P is equal to P1; if P1 is less than P0 and P2 is less than P0, P is equal to (P1+P2)/2. According to the output power prediction method for photovoltaic generating devices provided by the invention, relatively complex mathematic relation is changed to be simple and reasonable mathematic relation, thus being beneficial to applying to actual engineering, improving predicting speed, reducing predicting difficulty, and being capable of significantly improving accuracy of prediction value of photovoltaic module output power, controlling error range and eliminating outlier.
Description
Technical field
The present invention relates to a kind of output power predicting method of photovoltaic power generation equipment.
Background technology
Solar energy power generating is utilized by people to extensive grid-connected direction from autonomous system just gradually gradually as a kind of important distributed power source.But because photovoltaic generation is subject to the impact of intensity of solar radiation, battery component, temperature, weather cloud layer and the factor that some are random, system operational process is a nonequilibrium stochastic process, its generated energy and electromotive power output randomness is strong, it is large and wayward to fluctuate, and shows particularly outstandingly in the time of change in weather.This generation mode must bring a series of problem to the safety of electrical network and management after access electrical network.So can make prediction and seem particularly important the generating efficiency of photovoltaic system in advance comparatively accurately, simultaneously also for scheduling and the safe operation of electrical network provide foundation.
At present solar electrical energy generation is had to the forecasting techniques research of randomness seldom, existing forecast model comprises neural network model, radial basis function model and Multilayer apperceive model etc.Neural network model algorithm carries out short-term load forecasting, conventional is simple three layers of ANN model, main thought is that several factors of having the greatest impact using historical data and to electric load are as input quantity artificial neural network, finally generate output quantity through various neuronic effects in input layer, hidden layer and output layer, taking error as objective function, network weight is constantly revised until error reaches requirement, the network after training just can be predicted the output power of photovoltaic module again.Therefore the corresponding relation that it can solve the factor such as weather and temperature and load is preferably current comparatively common a kind of Forecasting Methodology.
But current neural network model can only be controlled at predicated error 20% left and right, predict the outcome and carry out dispatching of power netwoks according to this, still there is very large power wastage.And in a predetermined period, inevitably there will be several outliers during by Neural Network model predictive, depart from greatly actual value, this has just brought potential safety hazard to power grid security.
Given this, need one can reduce error, improve prediction accuracy and can eliminate the Forecasting Methodology of the output power of photovoltaic module of outlier.
Summary of the invention
The object of the invention is to improve a kind of accuracy that can significantly improve output power of photovoltaic module predicted value, departure scope, and eliminate the Forecasting Methodology of outlier.
To achieve these goals, the technical solution adopted in the present invention is:
The present invention adopts neural network prediction and the rolling neural network prediction of physical model prediction, environmental factor to predict respectively, the power stage value of performance number taking the performance number of physical model prediction as the neural network prediction of P0, environmental factor as the performance number of P1, rolling neural network prediction as P2, final prediction is as P:
If P1>P0 and P2>P0, P=P0;
If P1>P0>P2, P=P2;
If P2>P0>P1, P=P1;
If P1<P0 and P2<P0, P=(P1+P2)/2.
Further, in physical model prediction of the present invention, according to photovoltaic module performance parameter and monitored parameters, adopt following formula to calculate the current value (I) of prediction:
（1-16）
（1-17）
（1-18）
（1-19）
（1-20）
（1-21）
（1-22）
（1-23）
（1-24）
（1-25）
（1-26）
（1-27）
Wherein
t *for photovoltaic battery temperature, unit is K;
t tairfor the environment temperature detecting, unit is K;
kphotovoltaic battery temperature coefficient while variation for intensity of illumination;
t qfor weather conditions index;
Z is air quality index;
t reffor reference temperature, unit is DEG C;
sfor intensity of illumination index, unit is
;
s reffor intensity of illumination under the status of criterion, be 1000
;
A is temperature compensation coefficient;
B is intensity of illumination penalty coefficient;
C is temperature compensation coefficient;
D
t *, d
s *, d
i *, d
u *be respectively the correction for temperature, intensity of illumination, electric current, voltage;
i sCfor short-circuit current, unit is A;
u oCfor open-circuit voltage, unit is V;
i mfor maximum power point output current, unit is A;
u mthe output voltage of maximum power point, unit is V;
i sC *,
i m *,
u oC *,
u m *, be respectively revised electric current, voltage;
I is the photovoltaic cell output current of prediction, and unit is A;
c 1,
c 2for intermediate parameters;
Described monitored parameters is selected from environment temperature, intensity of illumination, weather conditions and air quality.
Further, the neural network prediction of environmental factor of the present invention comprises the steps:
Step 1: data acquisition: determine monitoring number of days, determine the monitoring time section of every day, determine the monitoring time interval of every day, select applicable condition element as monitored parameters image data; Gather the photovoltaic module real output of every day;
Described monitoring number of days was 1 ~ 10 week, described monitoring time section be early 5 point ~ 8 to late 16 point ~ 20 points, described monitoring time is spaced apart 1 ~ 3 hour, and described condition element is selected from one or more in weather, temperature, pollution index, intensity of illumination or photovoltaic module clean-up performance;
Step 2: the normalized of data: reject after the obvious outlier in the data of above-mentioned collection, adopt following formula to be normalized data:
（2-1）
Wherein
xfor the data after normalized,
xfor the data before normalized,
x min for variable
xminimum value,
x maxfor variable
xmaximal value;
Step 3: the selection of input layer: taking the photovoltaic output power under the time interval of every day monitoring in step 1 as input layer;
Step 4: the selection of output layer node: using the data after normalization in step 2 as output layer node;
Step 5: the selection of hidden layer and hidden nodes: select according to following formula:
(2-2)
Wherein
mfor the number of hidden nodes,
nfor input layer number,
lfor output node number,
Step 6: adopt the data neural network training model in step 3 ~ five, after having trained for predicting photovoltaic power value.
Further, the training method of neural network training model of the present invention is trainlm training method.
Further, rolling neural network prediction of the present invention comprises the steps:
Step 1: data acquisition: determine monitoring number of days, determine the monitoring time section of every day, determine the monitoring time interval of every day, select applicable condition element as monitored parameters image data; Gather the photovoltaic module real output of every day;
Described monitoring number of days is 20 ~ 50 days, described monitoring time section be early 5 point ~ 8 to late 16 point ~ 20 points; Described monitoring time is spaced apart 1 ~ 3 hour, and described condition element is selected from one or more in weather, temperature, pollution index, intensity of illumination or photovoltaic module clean-up performance;
Step 2: Data classification: reject after the obvious outlier in the data of above-mentioned collection, select a kind of monitored parameters as investigating variables A, according to investigating variables A, step 1 the data obtained is classified;
Described investigation variables A is selected from the one in weather, temperature, pollution index or intensity of illumination or photovoltaic module clean-up performance;
Step 3: the normalized of data:
Adopt following formula to be normalized data:
（2-1）
Wherein
xfor the data after normalized,
xfor the data before normalized,
x min for variable
xminimum value,
x maxfor variable
xmaximal value;
Step 4: adopt the photovoltaic output power of predicting day at three days before prediction day in the same photovoltaic output power prediction of investigating under variables A condition, until complete the prediction to whole training set;
Step 5: input layer: to predict that the monitored parameters except investigation variables A under day first three day each time interval is as input layer;
Step 6: output layer node: to predict day that photovoltaic output power under each time interval of prediction is as output layer node;
Step 7: hidden layer and hidden nodes: select according to following formula:
(2-2)
Wherein
mfor the number of hidden nodes,
nfor input layer number,
lfor output node number,
Step 8: adopt the data neural network training model in step 5 ~ seven, after having trained for predicting photovoltaic power value.
Further, the training method of neural network training model of the present invention is trainlm training method.
Physical model prediction of the present invention mainly refers to the Forecasting Methodology based on photovoltaic module electricity generating principle and physical model, the method has been considered weather conditions, temperature, intensity of illumination, the impact of pollution index (as PM2.5 concentration) on photovoltaic generation power, existing photovoltaic cell mathematical model is revised, set up the mathematical model of new photovoltaic module, by setting up physical simulation model prediction photovoltaic power size.The existing physical model Forecasting Methodology of use that those skilled in the art can select, or the physical model Forecasting Methodology of the multifactor correction of disclosed process in use the present invention, used disclosed method in the present invention can further improve the accuracy of prediction.
The neural network prediction of environmental factor of the present invention is mainly, based on environmental factor, the generated output of photovoltaic module is had to larger impact.The method has been considered the environmental factor of Various Complex and the impact of the cleanliness of photovoltaic module on photovoltaic generation power, input using weather conditions, temperature, intensity of illumination, pollution index (as PM2.5 concentration), photovoltaic module cleanliness as neural network model, photovoltaic power is exported as it, obtain photovoltaic power forecast model by training, thereby using the prediction weather conditions of day, temperature, intensity of illumination, pollution index (as PM2.5 concentration), photovoltaic module cleanliness as input, thereby obtain corresponding photovoltaic power predicted value.What those skilled in the art can select arranges applicable environmental factor according to this area situation, or uses disclosed environmental factor in the present invention to predict.
Rolling neural network prediction of the present invention is mainly the prediction based on photovoltaic power historical data, use the training set of photovoltaic power historical data as neural network model, first group of training set is output as the 4th day data of photovoltaic power historical data, first group of training set be input as first three day photovoltaic power, intensity of illumination, pollution index (as PM2.5 concentration), second group is output as the photovoltaic power historical data of the 5th day, second group is input as photovoltaic power, intensity of illumination, the historical data of second day to the 4th day of pollution index (as PM2.5 concentration), by that analogy, train the data of all training sets, obtain on line rolling prediction model, to predict that day first three daylight volt power is as prediction input data input on line rolling prediction model, thereby obtain this prediction day photovoltaic output power.The existing rolling neural net prediction method of use that those skilled in the art can select, or use the rolling neural net prediction method of the present invention to disclosed multifactor investigation, use disclosed method in the present invention can further improve the accuracy of prediction.
The beneficial effect that adopts technique scheme to produce is:
Method of the present invention is considered the impact of multiple environmental factor, more approaches environmental baseline when photovoltaic module is actual to be used, and can improve the accuracy of prediction; The present invention simultaneously, by adopting physical method, considers the impact of many factors on photovoltaic generation power, can more approach actual conditions, thereby the photovoltaic generation power that can more calculate to a nicety; Moreover the present invention is by adopting statistical method, physical method and statistical method combination, have complementary advantages, more accurately effective prediction photovoltaic generation power; Combine by said method, the error range of prediction can be contracted to 15%, and this be not attainable in prior art, by dwindling error range, can carry out output power of photovoltaic module more accurately compensatory, greatly reduce loss; And method of the present invention almost can be eliminated outlier completely, further promote the accuracy of prediction.
The mathematical relation of method relative complex of the present invention is converted into the mathematical relation of advantages of simple simultaneously, is conducive to the application of Practical Project, has improved the speed of predicting, has reduced the difficulty of predicting.
Brief description of the drawings
Fig. 1 is the process flow diagram of output power of photovoltaic module Forecasting Methodology of the present invention;
Fig. 2 is the electricity generating principle figure of photovoltaic module.
Embodiment
Embodiment 1
As shown in Figure 1, first the present embodiment predicts photovoltaic power by physical method, by analyzing physical arrangement and the electricity generating principle of photovoltaic cell, determine the various factors that affects photovoltaic generation power, comprise weather conditions, temperature, intensity of illumination, pollution index (as PM2.5 concentration), analyze the impact of these factors on photovoltaic generation power, by corresponding mathematical relation, determine the mathematical formulae that calculates photovoltaic cell capable of generating power power, and mathematical model is converted into realistic model, after the parameter of solar cell is determined, by the weather conditions of certain period, temperature, intensity of illumination, the change curve of pollution index (as PM2.5 concentration) is as input, operation realistic model, obtain the physical model predicted power change curve of photovoltaic cell capable of generating power power.
Secondly, the present embodiment adopts statistical method, because photovoltaic generation power is subject to the impact of many factors, the prediction of photovoltaic power is belonged to complicated nonlinear problem, and application neural network predicts it is a kind of effective method to photovoltaic power.
The present embodiment adopts two kinds of BP neural networks that neural net prediction method combines.First method is the neural network prediction of environmental factor, consider the impact of various environmental factors on photovoltaic generation power, with the weather conditions of a period of time, temperature, intensity of illumination, pollution index (as PM2.5 concentration), photovoltaic module cleanliness are as the input of training set, output using the actual count data of the photovoltaic power output of the solar cell of this period as the training set of neural network, select rational neural metwork training method, arithmetic accuracy and neural network structure, train this group data, obtain the neural network model of photovoltaic cell capable of generating power power by training.Using the weather conditions of certain period of needs prediction, temperature, intensity of illumination, pollution index (as PM2.5 concentration), photovoltaic module cleanliness as prediction input data, be input to the neural network model training, move this model, obtain the neural network prediction performance number of photovoltaic cell capable of generating power power.
The second method of neural network prediction adopts rolling forecast method, and the photovoltaic generation power before predict day by employings under same weather conditions in three days is inputted as the training set of neural network model, predicts that the photovoltaic power of this prediction day is exported.In the present embodiment, choose before prediction day not photovoltaic power real output value in the same time 30 day every day, weather conditions, intensity of illumination, the data such as pollution index (as PM2.5 concentration) are as the raw data of power prediction, Shijiazhuang District feature is classified according to weather conditions, as: form respectively fine day, cloudy day, haze, cloudy weather, under these conditions according to the photovoltaic power that contains of chronological order arrangement, illumination, the tables of data of pollution index (as PM2.5 concentration), select corresponding the 4th day not photovoltaic power real output value in the same time to prediction day the previous day for the output of training set, when prediction photovoltaic power, first detect the weather conditions on the same day, use the data of four forms of above-mentioned correspondence to train, obtain the neural network model of photovoltaic power rolling forecast, by the prediction day identical with the prediction day weather conditions photovoltaic power of 3 days before, intensity of illumination, pollution index (as PM2.5 concentration) is as the input of this neural network power prediction, move this model, obtain the prediction photovoltaic predicted power value of day.
By the combination application of physics Forecasting Methodology and statistical prediction methods, the output power curve of three kinds of Forecasting Methodologies is compared and revised, the output valve of predicted power is limited in certain error range, can improve the accuracy of predicted data.
Weather conditions described in the present embodiment are the weather conditions that are applicable to this area, in a data acquisition and prediction, all use identical predetermined value.As the weather conditions of selecting are fine, the moon, cloudy, sleet, haze, and the value of default fine day is 1 respectively, and the value at cloudy day is 0, and cloudy value is 0.5, and the value of sleet is 0, and the value of haze is 0.2.Those skilled in the art, in the time using method of the present invention to predict, can select the weather conditions and the preset value that are suitable for this area, all uses identical value as long as ensure in a data acquisition and prediction.
Pollution index described in the present embodiment can be air quality index (AQI), air quality subindex (IAQI), PM2.5 concentration or inspirable particle concentration, it is excellent that the data of announcing according to local weather bureau are defined as one-level, secondary is good, three grades of slight pollutions, level Four intermediate pollution, Pyatyi serious pollution, six ranks of six grades of severe contaminations.In a data acquisition and prediction, all use identical predetermined value, if default good value is 0, good value is 0.2, and the value of slight pollution is 0.3, and the value of intermediate pollution is 0.5, and the value of serious pollution is 0.7, severe contamination value be 0.9.Those skilled in the art, in the time using method of the present invention to predict, can select and be suitable for and this area pollution index and preset value, all uses identical value as long as ensure in a data acquisition and prediction.
(1) physical model prediction
The electricity generating principle figure of photovoltaic module as shown in Figure 2.Producer and the model of the photovoltaic module that the present embodiment is investigated are: prosperous JT240Ple is dredged in Jiangsu, and the data parameters that manufacturer provides is: short-circuit current
value be 8.55A, open-circuit voltage
value be 37.2V, maximum power point output current
value be 7.8A, peak power output
value be 240W, and the output voltage of maximum power point
value be 30.8V.
The formula of the mathematical model that existing photovoltaic cell parameter is set up is as follows:
（1-1）
（1-2）
（1-3）
Wherein ISC is short-circuit current, and unit is A;
UOC is open-circuit voltage, and unit is V;
Im is maximum power point output current, and unit is A;
The output voltage of Um maximum power point, unit is V;
I is the photovoltaic cell output current of prediction, and unit is A;
C1, C2 is intermediate parameters.
According to investigation factor, above-mentioned formula is revised, wherein the investigation factor of the present embodiment comprises: temperature T, and intensity of illumination S, PM2.5 concentration, weather conditions TQ(is fine, the moon is cloudy, sleet, haze).Weather conditions are fine, and the moon is cloudy, sleet, and haze, and the value of default fine day is 1 respectively, and the value at cloudy day is 0, and cloudy value is 0.5, and the value of sleet is 0, and the value of haze is 0.2.PM2.5 concentration is good 0 with air quality index Z(on the impact of photovoltaic generation power, good 0.2, slight pollution 0.3, intermediate pollution 0.5, serious pollution 0.7, severe contamination 0.9) represent.
Bring revised formula into:
（1-16）
（1-17）
（1-18）
（1-19）
（1-20）
（1-21）
（1-22）
（1-23）
（1-24）
（1-25）
（1-26）
（1-27）
Wherein T* is photovoltaic battery temperature, and unit is K;
Ttair is the environment temperature detecting, and unit is K;
Photovoltaic battery temperature coefficient when K is intensity of illumination variation;
Tq is weather conditions index;
Z is air quality index;
Tref is reference temperature, and unit is DEG C;
S is intensity of illumination index, and unit is
;
Sref is intensity of illumination under the status of criterion, is 1000
;
A is temperature compensation coefficient;
B is intensity of illumination penalty coefficient;
C is temperature compensation coefficient;
D T*, dS*, dI*, dU* is respectively the correction into temperature, intensity of illumination, electric current, voltage;
ISC is short-circuit current, and unit is A;
UOC is open-circuit voltage, and unit is V;
Im is maximum power point output current, and unit is A;
The output voltage of Um maximum power point, unit is V;
ISC**, Im**, UOC**, Um**, is respectively revised electric current, voltage;
I is the photovoltaic cell output current of prediction, and unit is A;
C1, C2 is intermediate parameters.
The current value of the photovoltaic module obtaining, according to P=UI, finally calculates generated output P0.Wherein P is power, and U is voltage, and I is electric current.
(2) neural network prediction of environmental factor
Step 1: data acquisition: determine that monitoring number of days is 30 days, the monitoring time section of determining every day is to amount to 16 hours at 8 in evening 5 of mornings, the monitoring time interval of determining every day is 1 hour, determines that according to condition element monitored parameters is weather, temperature, pollution index, intensity of illumination and photovoltaic module clean-up performance; Gather the photovoltaic module real output of every day.
Step 2: the normalized of data: reject after the obvious outlier in the data of above-mentioned collection, adopt following formula to be normalized data:
（2-1）
Be wherein the data after normalized,
for the data before normalized,
for variable
minimum value,
for variable
maximal value;
Step 3: the selection of input layer,
The present embodiment adopts 61 input variables, wherein 5 of 1-15 input mornings on daytime, (fine day represented with 1 to the weather conditions of each hour between at 8 in evening, cloudy day represents with 0, cloudy use 0.5 represents, haze represents with 0.2), 16-30 is input as 5 of mornings on daytime to the temperature value of each hour between at 8 in evening, 31-45 is input as 5 of mornings on daytime to the PM2.5 concentration value of each hour between at 8 in evening, 46-60 is input as 5 of mornings on daytime to the illumination intensity value of each hour between at 8 in evening, (cleanliness well represent with 1 the 61st clean-up performance that is input as photovoltaic cell component, cleanliness generally represent with 0.6, the very poor use 0.1 of cleanliness represents).
Step 4: the selection of output layer node: output layer node determines by target of prediction, photovoltaic generation statistical model prediction output layer node is 15 nodes, represents that 5 of mornings on daytime are to the photovoltaic output power of each hour between at 8 in evening.
Step 5: the selection of hidden layer and hidden nodes: calculate according to following formula:
Wherein
for the number of hidden nodes,
for input layer number,
for output node number,
Step 6: the present embodiment adopts the hidden layer configuration of 3 layers, and input quantity is 61, and output quantity is 15, and calculating the number of hidden nodes is 30.Select trainlm training method, neural network training model.
Obtain the neural network model net of photovoltaic cell capable of generating power power by training.Using the prediction weather conditions of day, temperature, intensity of illumination, PM2.5 concentration, photovoltaic module clean-up performance data as the input data of predicting, input neural network model net, moves this model, obtains the neural network prediction value of the photovoltaic power of this day.
(3) rolling neural network prediction
Step 1: data acquisition: the monitoring time section of determining every day is to amount to 16 hours at 8 in evening 5 of mornings, the monitoring time interval of determining every day is 1 hour, determines that according to condition element monitored parameters is weather, temperature, pollution index, intensity of illumination and photovoltaic module clean-up performance; Gather the photovoltaic module real output of every day.
Step 2: Data classification: reject after the obvious outlier in the data of above-mentioned collection, determine that investigating variables A is weather conditions, then classify according to weather conditions, form respectively the tables of data that contains photovoltaic power, illumination, PM2.5 concentration of arranging according to chronological order under fine day, cloudy day, haze, cloudy weather condition.
Step 3: the normalized of data:
Adopt following formula to be normalized data:
（2-1）
Wherein
for the data after normalized,
for the data before normalized,
for variable
minimum value,
for variable
maximal value;
Step 4: the photovoltaic output power of day is predicted in the photovoltaic output power prediction of first three day under same investigation variables A condition before of employing prediction day, until complete the prediction to whole training set.When the present embodiment prediction photovoltaic power, first detect the weather conditions on the same day, use the data of four forms of above-mentioned correspondence to predict, that fine day is as example taking prediction day weather conditions, choose not photovoltaic power, intensity of illumination, the PM2.5 concentration in the same time of the 1st day to the 3rd day of carrying out under above-mentioned fine day situation after data processing and input as first group of training set, not photovoltaic power output in the same time that corresponding training set is output as under fine day situation the 4th day; Under second group of the training set above-mentioned fine day situation of input data decimation, carry out not photovoltaic power, intensity of illumination, the PM2.5 concentration value in the same time of after data processing the 2nd day to the 4th day, the not photovoltaic power output in the same time that is output as under fine day situation the 5th day; By that analogy, always with the 4th day photovoltaic power of first three day photovoltaic power prediction under same weather conditions, until be output as last group data of training set.
Step 5: input layer: the present embodiment adopts 135 input variables, wherein 1-45 is input as 5 of the 1st day to the 3rd day every mornings to the photovoltaic output power of each hour between at 8 in evening, 46-90 is input as 5 of the 1st day to the 3rd day mornings to each little intensity of illumination temperature value between at 8 in evening, and 91-135 is input as 5 of the 1st day to the 3rd day mornings to the PM2.5 concentration value of each hour between at 8 in evening.
Step 6: output layer node: output layer node determines by target of prediction, photovoltaic generation statistical model prediction output layer node is 15 nodes, represents that 5 of mornings on daytime are to the photovoltaic output power of each hour between at 8 in evening.
Step 7: hidden layer and hidden nodes: select according to following formula:
Wherein
for the number of hidden nodes,
for input layer number,
for output node number,
Step 8: the BP neural network of this forecast model design adopts 3-tier architecture, and input quantity is 135, and output quantity is 15, and the number of hidden nodes of calculating is 45, selects trainlm training method, neural network training model.
Obtain the neural network model net1 of photovoltaic cell capable of generating power power by training.Before the prediction day identical with prediction day weather conditions, the photovoltaic power of 3 days, intensity of illumination, PM2.5 concentration are as the input of this neural network power prediction, and input neural network model net1, moves this model, obtain the photovoltaic predicted power of prediction day.
(4) power prediction
According to formula:
If P1>P0 and P2>P0, P=P0;
If P1>P0>P2, P=P2;
If P2>P0>P1, P=P1;
If P1<P0 and P2<P0, P=(P1+P2)/2.
Bring P0, P1 and the P2 of above-mentioned model prediction gained into, by said method, the predict 1 day photovoltaic output power of (8: 15～16: 45, interval 5min), obtains final predicted value P as table 1.According to the data of table 1, the performance number P of prediction approaches the actual performance number P reality recording, and the error range of predicted value is less than 15%.
The photovoltaic power predicted value of table 1 embodiment 1
(unit is: kw)
Table 1
Table 1 continuous 1
Table 1 continuous 2
Claims (6)
1. a photovoltaic power Forecasting Methodology, it is characterized in that it adopts neural network prediction and the rolling neural network prediction of physical model prediction, environmental factor to predict respectively, the power stage value of performance number taking the performance number of physical model prediction as the neural network prediction of P0, environmental factor as the performance number of P1, rolling neural network prediction as P2, final prediction is as P:
If P1>P0 and P2>P0, P=P0;
If P1>P0>P2, P=P2;
If P2>P0>P1, P=P1;
If P1<P0 and P2<P0, P=(P1+P2)/2.
2. photovoltaic power Forecasting Methodology according to claim 1, according to photovoltaic module performance parameter and monitored parameters, adopts following formula to calculate the current value (I) of prediction in the physical model prediction described in it is characterized in that:
（1-16）
（1-17）
（1-18）
（1-19）
（1-20）
（1-21）
（1-22）
（1-23）
（1-24）
（1-25）
（1-26）
（1-27）
Wherein
t *for photovoltaic battery temperature, unit is K;
t tairfor the environment temperature detecting, unit is K;
kphotovoltaic battery temperature coefficient while variation for intensity of illumination;
t qfor weather conditions index;
Z is air quality index;
t reffor reference temperature, unit is DEG C;
sfor intensity of illumination index, unit is
;
s reffor intensity of illumination under the status of criterion, be 1000
;
A is temperature compensation coefficient;
B is intensity of illumination penalty coefficient;
C is temperature compensation coefficient;
D
t *, d
s *, d
i *, d
u *be respectively the correction for temperature, intensity of illumination, electric current, voltage;
i sCfor short-circuit current, unit is A;
u oCfor open-circuit voltage, unit is V;
i mfor maximum power point output current, unit is A;
u mthe output voltage of maximum power point, unit is V;
i sC *,
i m *,
u oC *,
u m *, be respectively revised electric current, voltage;
I is the photovoltaic cell output current of prediction, and unit is A;
c 1,
c 2for intermediate parameters;
Described monitored parameters is selected from environment temperature, intensity of illumination, weather conditions and air quality.
3. photovoltaic power Forecasting Methodology according to claim 1, is characterized in that the neural network prediction of described environmental factor comprises the steps:
Step 1: data acquisition: determine monitoring number of days, determine the monitoring time section of every day, determine the monitoring time interval of every day, select applicable condition element as monitored parameters image data; Gather the photovoltaic module real output of every day;
Described monitoring number of days was 1 ~ 10 week, described monitoring time section be early 5 point ~ 8 to late 16 point ~ 20 points, described monitoring time is spaced apart 1 ~ 3 hour, and described condition element is selected from one or more in weather, temperature, pollution index, intensity of illumination or photovoltaic module clean-up performance;
Step 2: the normalized of data: reject after the obvious outlier in the data of above-mentioned collection, adopt following formula to be normalized data:
（2-1）
Wherein
xfor the data after normalized,
xfor the data before normalized,
x min for variable
xminimum value,
x maxfor variable
xmaximal value;
Step 3: the selection of input layer: taking the photovoltaic output power under the time interval of every day monitoring in step 1 as input layer;
Step 4: the selection of output layer node: using the data after normalization in step 2 as output layer node;
Step 5: the selection of hidden layer and hidden nodes: select according to following formula:
(2-2)
Wherein
mfor the number of hidden nodes,
nfor input layer number,
lfor output node number,
Step 6: adopt the data neural network training model in step 3 ~ five, after having trained for predicting photovoltaic power value.
4. photovoltaic power Forecasting Methodology according to claim 3, is characterized in that the training method of described neural network training model is trainlm training method.
5. photovoltaic power Forecasting Methodology according to claim 1, is characterized in that described rolling neural network prediction comprises the steps:
Step 1: data acquisition: determine monitoring number of days, determine the monitoring time section of every day, determine the monitoring time interval of every day, select applicable condition element as monitored parameters image data; Gather the photovoltaic module real output of every day;
Described monitoring number of days is 20 ~ 50 days, described monitoring time section be early 5 point ~ 8 to late 16 point ~ 20 points; Described monitoring time is spaced apart 1 ~ 3 hour, and described condition element is selected from one or more in weather, temperature, pollution index, intensity of illumination or photovoltaic module clean-up performance;
Step 2: Data classification: reject after the obvious outlier in the data of above-mentioned collection, select a kind of monitored parameters as investigating variables A, according to investigating variables A, step 1 the data obtained is classified;
Described investigation variables A is selected from the one in weather, temperature, pollution index or intensity of illumination or photovoltaic module clean-up performance;
Step 3: the normalized of data:
Adopt following formula to be normalized data:
（2-1）
Wherein
xfor the data after normalized,
xfor the data before normalized,
x min for variable
xminimum value,
x maxfor variable
xmaximal value;
Step 4: adopt the photovoltaic output power of predicting day at three days before prediction day in the same photovoltaic output power prediction of investigating under variables A condition, until complete the prediction to whole training set;
Step 5: input layer: to predict that the monitored parameters except investigation variables A under day first three day each time interval is as input layer;
Step 6: output layer node: to predict day that photovoltaic output power under each time interval of prediction is as output layer node;
Step 7: hidden layer and hidden nodes: select according to following formula:
(2-2)
Wherein
mfor the number of hidden nodes,
nfor input layer number,
lfor output node number,
Step 8: adopt the data neural network training model in step 5 ~ seven, after having trained for predicting photovoltaic power value.
6. photovoltaic power Forecasting Methodology according to claim 5, is characterized in that the training method of described neural network training model is trainlm training method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410118453.4A CN104200274A (en) | 2014-03-27 | 2014-03-27 | Power prediction method for photovoltaic devices |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410118453.4A CN104200274A (en) | 2014-03-27 | 2014-03-27 | Power prediction method for photovoltaic devices |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104200274A true CN104200274A (en) | 2014-12-10 |
Family
ID=52085563
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410118453.4A Pending CN104200274A (en) | 2014-03-27 | 2014-03-27 | Power prediction method for photovoltaic devices |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104200274A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104965558A (en) * | 2015-05-27 | 2015-10-07 | 华北电力大学（保定） | Photovoltaic power generation system maximum power tracking method and apparatus considering the factor of haze |
| CN109978280A (en) * | 2019-04-19 | 2019-07-05 | 上海交通大学 | A kind of generalization photovoltaic cell operating temperature prediction technique and device |
| CN110675278A (en) * | 2019-09-18 | 2020-01-10 | 上海电机学院 | Photovoltaic power short-term prediction method based on RBF neural network |
| CN110703847A (en) * | 2019-10-29 | 2020-01-17 | 特变电工西安电气科技有限公司 | Photovoltaic global maximum power point tracking method of improved particle swarm-disturbance observation method |
| CN110796292A (en) * | 2019-10-14 | 2020-02-14 | 国网辽宁省电力有限公司盘锦供电公司 | Photovoltaic power short-term prediction method considering haze influence |
| CN113031110A (en) * | 2019-12-25 | 2021-06-25 | 国创新能源汽车智慧能源装备创新中心(江苏)有限公司 | Cloud layer monitoring method and device and photovoltaic power prediction method and device |
| CN113326658A (en) * | 2021-06-03 | 2021-08-31 | 中国南方电网有限责任公司 | Photovoltaic energy storage grid-connected control method based on neural network |
| CN115953042A (en) * | 2023-03-10 | 2023-04-11 | 山东创宇环保科技有限公司 | Distributed energy optimization control management method based on expert system |
| CN116111948A (en) * | 2022-12-07 | 2023-05-12 | 中广核贵州安顺关岭新能源有限公司 | Photovoltaic branch equipment fault early warning system |
-
2014
- 2014-03-27 CN CN201410118453.4A patent/CN104200274A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| 葛鹏江: "《大规模集中接入的光伏电站功率预测》", 《中国优秀硕士论文电子期刊网 工程科技II辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104965558A (en) * | 2015-05-27 | 2015-10-07 | 华北电力大学（保定） | Photovoltaic power generation system maximum power tracking method and apparatus considering the factor of haze |
| CN109978280A (en) * | 2019-04-19 | 2019-07-05 | 上海交通大学 | A kind of generalization photovoltaic cell operating temperature prediction technique and device |
| CN110675278A (en) * | 2019-09-18 | 2020-01-10 | 上海电机学院 | Photovoltaic power short-term prediction method based on RBF neural network |
| CN110796292A (en) * | 2019-10-14 | 2020-02-14 | 国网辽宁省电力有限公司盘锦供电公司 | Photovoltaic power short-term prediction method considering haze influence |
| CN110703847A (en) * | 2019-10-29 | 2020-01-17 | 特变电工西安电气科技有限公司 | Photovoltaic global maximum power point tracking method of improved particle swarm-disturbance observation method |
| CN113031110A (en) * | 2019-12-25 | 2021-06-25 | 国创新能源汽车智慧能源装备创新中心(江苏)有限公司 | Cloud layer monitoring method and device and photovoltaic power prediction method and device |
| CN113031110B (en) * | 2019-12-25 | 2024-01-23 | 国创移动能源创新中心(江苏)有限公司 | Cloud layer monitoring method and device, photovoltaic power prediction method and device |
| CN113326658A (en) * | 2021-06-03 | 2021-08-31 | 中国南方电网有限责任公司 | Photovoltaic energy storage grid-connected control method based on neural network |
| CN113326658B (en) * | 2021-06-03 | 2024-03-12 | 中国南方电网有限责任公司 | Photovoltaic energy storage grid-connected control method based on neural network |
| CN116111948A (en) * | 2022-12-07 | 2023-05-12 | 中广核贵州安顺关岭新能源有限公司 | Photovoltaic branch equipment fault early warning system |
| CN116111948B (en) * | 2022-12-07 | 2024-04-23 | 中广核贵州安顺关岭新能源有限公司 | Photovoltaic branch equipment fault early warning system |
| CN115953042A (en) * | 2023-03-10 | 2023-04-11 | 山东创宇环保科技有限公司 | Distributed energy optimization control management method based on expert system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104200274A (en) | Power prediction method for photovoltaic devices | |
| Suganthi et al. | Applications of fuzzy logic in renewable energy systems–a review | |
| Al-Majidi et al. | A particle swarm optimisation-trained feedforward neural network for predicting the maximum power point of a photovoltaic array | |
| Ayop et al. | Components sizing of photovoltaic stand-alone system based on loss of power supply probability | |
| Khatib et al. | A review on sizing methodologies of photovoltaic array and storage battery in a standalone photovoltaic system | |
| García-Triviño et al. | Optimized operation combining costs, efficiency and lifetime of a hybrid renewable energy system with energy storage by battery and hydrogen in grid-connected applications | |
| Chandra et al. | A detailed literature review on wind forecasting | |
| Capizzi et al. | Recurrent neural network-based control strategy for battery energy storage in generation systems with intermittent renewable energy sources | |
| CN104050517A (en) | Photovoltaic power generation forecasting method based on GRNN | |
| Hong et al. | Interactive multi-objective active power scheduling considering uncertain renewable energies using adaptive chaos clonal evolutionary programming | |
| Xu et al. | Energy allocation strategy based on fuzzy control considering optimal decision boundaries of standalone hybrid energy systems | |
| Mousavi et al. | A real-time energy management strategy for pumped hydro storage systems in farmhouses | |
| Dimopoulou et al. | A Markov decision process for managing a hybrid energy storage system | |
| Al Smadi et al. | Artificial intelligent control of energy management PV system | |
| CN115912466A (en) | Active power distribution network island division method and system based on information gap decision theory | |
| Abdelhak et al. | Optimum sizing of hybrid PV/wind/battery using Fuzzy-Adaptive Genetic Algorithm in real and average battery service life | |
| Hussaian Basha et al. | A novel on design and implementation of hybrid MPPT controllers for solar PV systems under various partial shading conditions | |
| Anu Shalini et al. | Power generation forecasting using deep learning CNN-based BILSTM technique for renewable energy systems | |
| Iqbal et al. | Analysis and comparison of various control strategy of hybrid power generation a review | |
| Boulal et al. | Optimal management of energy flows in a multi-source grid | |
| CN114676921A (en) | Method for calculating wind power receptibility of system by considering source load storage coordination optimization | |
| Mahmud et al. | ANFISPID-based voltage regulation strategy for grid-tied renewable DG system with ESS | |
| Thanigaivel et al. | A hybrid DMO-RERNN based UPFC controller for transient stability analysis in grid connected wind-diesel-PV hybrid system | |
| Divya et al. | ANN Based Solar Power Forecasting in a Smart Microgrid System for Power Flow Management | |
| Eshraghi et al. | Power grid reliability improvement through forecasting with complex-valued neural networks under system contingency |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Gao Zhiqiang Inventor after: Fan Hui Inventor after: Hu Wenping Inventor after: Sun Zhongji Inventor after: Meng Liang Inventor after: Jing Hao Inventor after: Liang Bin Inventor after: Yang Xiao Inventor after: Chu Huayu Inventor before: Gao Zhiqiang Inventor before: Chu Huayu Inventor before: Sun Zhongji Inventor before: Meng Liang Inventor before: Jing Hao Inventor before: Liang Bin Inventor before: Yang Xiao |
|
| COR | Change of bibliographic data | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20141210 |