CN105528517B - Based on neural network, wavelet decomposition predicting power of photovoltaic plant method, system - Google Patents
Based on neural network, wavelet decomposition predicting power of photovoltaic plant method, system Download PDFInfo
- Publication number
- CN105528517B CN105528517B CN201510867065.0A CN201510867065A CN105528517B CN 105528517 B CN105528517 B CN 105528517B CN 201510867065 A CN201510867065 A CN 201510867065A CN 105528517 B CN105528517 B CN 105528517B
- Authority
- CN
- China
- Prior art keywords
- signal layer
- wavelet decomposition
- information
- neural network
- humidity
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Abstract
The invention discloses one kind to be based on neural network, wavelet decomposition predicting power of photovoltaic plant method, system, obtains historical record intensity of illumination C, output power P, temperature T, humidity H, wind speed WS, carries out wavelet decomposition respectively；According to the information after the wavelet decomposition, approximation signal layer neural network model and detail signal layer neural network model are established；Future prediction time section temperature T, humidity H, wind speed WS are obtained, respectively through wavelet decomposition；The forecast result of approximation signal layer is obtained using the information of approximation signal layer as the input vector of approximation signal layer neural network model according to the information after the wavelet decomposition；Using the information of detail signal layer as the input vector of detail signal layer neural network model, the forecast result of detail signal layer is obtained；The forecast result of forecast result and detail signal layer to approximation signal layer is reconstructed, and obtains prediction power value.Therefore, the present invention can guarantee the convergence of prediction process, reduce complexity, improve the accuracy of prediction.
Description
Technical field
The present invention relates to energy fields, particularly relate to a kind of based on neural network, wavelet decomposition predicting power of photovoltaic plant
Method, system.
Background technique
Solar energy power generating is that solar radiant energy is converted directly into electric energy using the photovoltaic effect of solar battery
A kind of forms of electricity generation.Currently, the problems such as fossil energy is short, environmental pollution is serious and Global climate change becomes increasingly conspicuous, the world
Energy development show clean, low-carbon, the new trend of high efficiency.China's solar energy resources are close with the U.S., than Europe, day
This is more advantageous, and the potentiality to be exploited having is extremely wide, is in recent years even more to enter the high-speed developing period, becomes new growth
Point.The development and utilization of solar energy meet our country energy development strategy, realize electric power sustainable development, electric power structural adjustment and environment
The needs of protection.China's future will have a large amount of distributed clean energy resource power generations and other forms power generation access power grid, it is desirable that power grid
Ability is received with clean energy resource.
With the development of solar power generation industry, solar grid-connected scale is increasing, but solar power generation has at random
Property it is big, intermittent, demodulate the features such as peak, consumption of the access of large solar power generation to power grid, peak-frequency regulation demand, stabilization
The problems such as property, can generate significant impact, in order to reduce these influences, while guaranteeing the stable operation of power system security, make electric power
Traffic department can change according to photovoltaic power generation quantity to be scheduled in time, is reduced spare capacity and operating cost, is needed to photovoltaic
The generated energy in power station carries out Accurate Prediction.
And traditional photovoltaic plant output power predicting method based on conventional neural networks, only simply influence photovoltaic
The factor of station output is taken into account, and as the input signal of model, photovoltaic plant output power can be effectively predicted in it, but
The periodicity and non-stationary characteristic of photovoltaic plant output power can be such that traditional neural network algorithm is easily trapped into not restrain, in addition sharp
The high fdrequency component that can not be modeled with existing numerical weather forecast does not weed out, and the model of foundation is not accurate enough, and it is restrained
Property is bad, and computation complexity is higher.
Summary of the invention
In view of this, it is an object of the invention to propose one kind in order to solve the deficiency of existing predicting power of photovoltaic plant
Photovoltaic plant output power predicting method improves photovoltaic power precision of prediction.
It is provided by the invention based on neural network, wavelet decomposition predicting power of photovoltaic plant method, packet based on above-mentioned purpose
Include step:
Obtain the historical record of photovoltaic plant position in a period；Wherein, the historical record includes illumination
Intensity C, output power P, temperature T, humidity H, wind speed WS；
The intensity of illumination C, output power P, temperature T, humidity H, wind speed WS are subjected to wavelet decomposition respectively；According to described
Information after wavelet decomposition establishes approximation signal layer neural network model and detail signal layer neural network model；
Obtain the weather forecast value of photovoltaic plant position future prediction time section；Wherein, the weather forecast value
Including temperature T, humidity H, wind speed WS；
By the temperature T of the future prediction time section, humidity H, wind speed WS predicted value respectively through wavelet decomposition；According to institute
Information after stating wavelet decomposition is obtained using the information of approximation signal layer as the input vector of approximation signal layer neural network model
To the forecast result of approximation signal layer；Using the information of detail signal layer as the input of detail signal layer neural network model to
Amount, obtains the forecast result of detail signal layer；
The forecast result of forecast result and detail signal layer to approximation signal layer is reconstructed, and obtains prediction power value.
In some embodiments, the intensity of illumination C that will acquire, output power P, temperature T, humidity H, wind speed WS difference
It carries out wavelet decomposition and it is each to respectively obtain intensity of illumination C, output power P, temperature T, humidity H, wind speed WS using 5 layers of wavelet decomposition
The information of comfortable detail signal layer D1, D2, D3, D4, D5 and the information of approximation signal layer A5；Meanwhile it obtaining output power P and existing
Every layer of coefficient of wavelet decomposition after 5 layers of wavelet decomposition.
In some embodiments, the information according to after the wavelet decomposition establishes approximation signal layer neural network mould
Type includes:
The A5 approximation signal layer information [C after wavelet decomposition respectively by intensity of illumination C, temperature T, humidity H, wind speed WSA5
TA5 HA5 WSA5] it is used as input vector, by A5 approximation signal layer information [0 00 Ps of the output power P after wavelet decompositionA5]
As output vector, A5 approximation signal layer neural network model is established；
In addition, the detail signal layer neural network model of establishing includes:
The D5 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD5 HD5 WSD5]
As input vector, by D5 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD5] it is used as output vector,
Establish D5 detail signal layer neural network model；
The D4 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD4 HD4 WSD4]
As input vector, by D4 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD4] it is used as output vector,
Establish D4 detail signal layer neural network model.
In some embodiments, the predicted value by the temperature T of the future prediction time section, humidity H, wind speed WS point
Not through wavelet decomposition, using 5 layers of wavelet decomposition, respectively obtain each comfortable detail signal layer D1, D2 of temperature T, humidity H, wind speed WS,
The information of D3, D4, D5 and the information of approximation signal layer A5；
In addition, it is described using the information of approximation signal layer as the input vector of approximation signal layer neural network model, it obtains
The forecast result of approximation signal layer includes: that the intensity of illumination C of prediction, temperature T, humidity H, wind speed WS are passed through wavelet decomposition respectively
A5 approximation signal layer information [C afterwards0A5 T0A5 H0A5 WS0A5] it is used as input vector, according to A5 approximation signal layer neural network mould
Type obtains forecast result [0 00 P of the A5 approximation signal layer of photovoltaic plant output power0A5]；
In addition, obtaining details using the information of detail signal layer as the input vector of detail signal layer neural network model
The forecast result of signals layer includes: by the temperature T, humidity H, wind speed WS of prediction the D5 detail signal after wavelet decomposition respectively
Layer information [T0D5 H0D5 WS0D5] be used as input vector that it is defeated to obtain photovoltaic plant according to D5 detail signal layer neural network model
Forecast result [0 0 P of the D5 detail signal layer of power out0D5]；
The D4 detail signal layer information [T after wavelet decomposition respectively by the temperature T, humidity H, wind speed WS of prediction0D4
H0D4 WS0D4] it is used as input vector, according to D4 detail signal layer neural network model, the D4 for obtaining photovoltaic plant output power is thin
Save forecast result [0 0 P of signals layer0D4]。
In some embodiments, described that weight is carried out to the forecast result of approximation signal layer and the forecast result of detail signal layer
Structure, comprising: the coefficient of wavelet decomposition by obtained output power P at D4, D5, A5 layers, respectively multiplied by P0D4、P0D5、P0A5, finally
The results added of multiplication is obtained into prediction power value P0。
In addition the present invention also provides one kind to be based on neural network, wavelet decomposition predicting power of photovoltaic plant system, comprising:
Historical data acquiring unit, for obtaining the historical record of photovoltaic plant position in a period；Wherein, institute
The historical record stated includes intensity of illumination C, output power P, temperature T, humidity H, wind speed WS；
Prediction model establishes unit, for distinguishing the intensity of illumination C, output power P, temperature T, humidity H, wind speed WS
Carry out wavelet decomposition；According to the information after the wavelet decomposition, approximation signal layer neural network model and detail signal layer are established
Neural network model；
Prediction data acquiring unit, for obtaining the weather forecast value of photovoltaic plant position future prediction time section；
Wherein, the weather forecast value includes temperature T, humidity H, wind speed WS；
Power estimation unit, for distinguishing the predicted value of the temperature T of the future prediction time section, humidity H, wind speed WS
Through wavelet decomposition；According to the information after the wavelet decomposition, using the information of approximation signal layer as approximation signal layer neural network
The input vector of model obtains the forecast result of approximation signal layer；Using the information of detail signal layer as detail signal layer nerve
The input vector of network model obtains the forecast result of detail signal layer；
Prediction power reconfiguration unit, the forecast result for forecast result and detail signal layer to approximation signal layer carry out
Reconstruct, obtains prediction power value.
In some embodiments, the prediction model establish the intensity of illumination C that unit will acquire, output power P, temperature T,
Humidity H, wind speed WS carry out wavelet decomposition respectively, using 5 layers of wavelet decomposition, respectively obtain intensity of illumination C, output power P, temperature
T, the information of humidity H, wind speed WS each comfortable detail signal layer D1, D2, D3, D4, D5 and the information of approximation signal layer A5；Together
When, obtain output power P every layer of coefficient of wavelet decomposition after 5 layers of wavelet decomposition.
In some embodiments, the prediction model establishes unit according to the information after the wavelet decomposition, and foundation approaches
Signals layer neural network model includes:
The A5 approximation signal layer information [C after wavelet decomposition respectively by intensity of illumination C, temperature T, humidity H, wind speed WSA5
TA5 HA5 WSA5] it is used as input vector, by A5 approximation signal layer information [0 00 Ps of the output power P after wavelet decompositionA5]
As output vector, A5 approximation signal layer neural network model is established；
In addition, the detail signal layer neural network model of establishing includes:
The D5 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD5 HD5 WSD5]
As input vector, by D5 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD5] it is used as output vector,
Establish D5 detail signal layer neural network model；
The D4 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD4 HD4 WSD4]
As input vector, by D4 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD4] it is used as output vector,
Establish D4 detail signal layer neural network model.
In some embodiments, the prediction data acquiring unit by the temperature T of the future prediction time section, humidity H,
The predicted value of wind speed WS respectively obtains temperature T, humidity H, each leisure of wind speed WS using 5 layers of wavelet decomposition through wavelet decomposition respectively
The information of detail signal layer D1, D2, D3, D4, D5 and the information of approximation signal layer A5；
In addition, it is described using the information of approximation signal layer as the input vector of approximation signal layer neural network model, it obtains
The forecast result of approximation signal layer includes: that the intensity of illumination C of prediction, temperature T, humidity H, wind speed WS are passed through wavelet decomposition respectively
A5 approximation signal layer information [C afterwards0A5 T0A5 H0A5 WS0A5] it is used as input vector, according to A5 approximation signal layer neural network mould
Type obtains forecast result [0 00 P of the A5 approximation signal layer of photovoltaic plant output power0A5]；
In addition, obtaining details using the information of detail signal layer as the input vector of detail signal layer neural network model
The forecast result of signals layer includes: by the temperature T, humidity H, wind speed WS of prediction the D5 detail signal after wavelet decomposition respectively
Layer information [T0D5 H0D5 WS0D5] be used as input vector that it is defeated to obtain photovoltaic plant according to D5 detail signal layer neural network model
Forecast result [0 0 P of the D5 detail signal layer of power out0D5]；
The D4 detail signal layer information [T after wavelet decomposition respectively by the temperature T, humidity H, wind speed WS of prediction0D4
H0D4 WS0D4] it is used as input vector, according to D4 detail signal layer neural network model, the D4 for obtaining photovoltaic plant output power is thin
Save forecast result [0 0 P of signals layer0D4]。
In some embodiments, forecast result and detail signal layer of the prediction power reconfiguration unit to approximation signal layer
Forecast result be reconstructed, comprising: the coefficient of wavelet decomposition by obtained output power P at D4, D5, A5 layers, respectively multiplied by
P0D4、P0D5、P0A5, the results added of multiplication is finally obtained into prediction power value P0。
From the above it can be seen that provided by the invention be based on neural network, wavelet decomposition predicting power of photovoltaic plant
Method and system, realizes a set of simple and easy to do, ensure that the convergence of prediction process, reduces complexity, improves prediction
Accuracy.
Detailed description of the invention
Fig. 1 is that the process based on neural network, wavelet decomposition predicting power of photovoltaic plant method is shown in the embodiment of the present invention
It is intended to；
Fig. 2 is that the present invention can refer in embodiment based on neural network, wavelet decomposition predicting power of photovoltaic plant method
Flow diagram；
Fig. 3 is that the structure based on neural network, wavelet decomposition predicting power of photovoltaic plant system is shown in the embodiment of the present invention
It is intended to.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Attached drawing, the present invention is described in more detail.
As an embodiment of the present invention, as shown in fig.1, it is described based on neural network, wavelet decomposition photovoltaic electric
Power forecasting method of standing includes:
Step 101, the historical record of photovoltaic plant position in a period is obtained.Wherein, the historical record
Including intensity of illumination C, output power P, temperature T, humidity H, wind speed WS.
Step 102, the intensity of illumination C, output power P, temperature T, humidity H, wind speed WS are subjected to wavelet decomposition respectively；
According to the information after the wavelet decomposition, approximation signal layer neural network model and detail signal layer neural network model are established.
In embodiment, intensity of illumination C, output power P, temperature T, humidity H, the wind speed WS that will acquire carry out small echo respectively
It decomposes, using 5 layers of wavelet decomposition, respectively obtains each comfortable details of intensity of illumination C, output power P, temperature T, humidity H, wind speed WS
The information of signals layer D1, D2, D3, D4, D5 and the information of approximation signal layer A5.Meanwhile output power P is obtained in 5 layers of small echo
Every layer of coefficient of wavelet decomposition after decomposition.
Preferably, establishing approximation signal layer neural network model may include: by intensity of illumination C, temperature T, humidity H, wind
The fast WS A5 approximation signal layer information [C after wavelet decomposition respectivelyA5 TA5 HA5 WSA5] it is used as input vector, by output work
A5 approximation signal layer information [0 00 Ps of the rate P after wavelet decompositionA5] it is used as output vector, establish A5 approximation signal layer mind
Through network model.
As another embodiment, D5 detail signal layer mind can be established respectively by establishing detail signal layer neural network model
Through network model and D4 detail signal layer neural network model.
Preferably, establish D5 detail signal layer neural network model can be by passing through temperature T, humidity H, wind speed WS respectively
D5 detail signal layer information [T after crossing wavelet decompositionD5 HD5 WSD5] it is used as input vector, output power P is passed through into small wavelength-division
D5 detail signal layer information [0 0 P after solutionD5] it is used as output vector, establish D5 detail signal layer neural network model.
In addition, as another preferred embodiment, by temperature T, humidity H, wind speed WS respectively after wavelet decomposition
D4 detail signal layer information [TD4 HD4 WSD4] it is used as input vector, D4 details of the output power P after wavelet decomposition is believed
Number floor information [0 0 PD4] it is used as output vector, establish D4 detail signal layer neural network model.
Step 103, the weather forecast value of photovoltaic plant position future prediction time section is obtained.Wherein, the day
Gas predicted value includes temperature T, humidity H, wind speed WS.
Step 104, by the temperature T of the future prediction time section, humidity H, wind speed WS predicted value respectively through small wavelength-division
Solution；According to the information after the wavelet decomposition, using the information of approximation signal layer as the defeated of approximation signal layer neural network model
Incoming vector obtains the forecast result of approximation signal layer；Using the information of detail signal layer as detail signal layer neural network model
Input vector, obtain the forecast result of detail signal layer.
Preferably, by the temperature T of the future prediction time section, humidity H, wind speed WS predicted value respectively through small wavelength-division
Solution respectively obtains temperature T, humidity H, wind speed WS each comfortable detail signal layer D1, D2, D3, D4, D5 using 5 layers of wavelet decomposition
The information of information and approximation signal layer A5.
It should be noted that the forecast result for obtaining approximation signal layer can be by the intensity of illumination C of prediction, temperature T, humidity
H, the wind speed WS A5 approximation signal layer information [C after wavelet decomposition respectively0A5 T0A5 H0A5 WS0A5] it is used as input vector, root
According to A5 approximation signal layer neural network model, the forecast result [0 00 of the A5 approximation signal layer of photovoltaic plant output power is obtained
P0A5]。
And it is worth noting that, it can be using the information of D5, D4 detail signal layer of prediction as D5 detail signal layer
The input vector of the input vector of neural network model, D4 detail signal layer neural network model, obtains D5 detail signal layer
The forecast result of forecast result and D4 detail signal layer.
Preferably, the D5 detail signal layer information after wavelet decomposition respectively by the temperature T of prediction, humidity H, wind speed WS
[T0D5 H0D5 WS0D5] input vector is used as to obtain photovoltaic plant output power according to D5 detail signal layer neural network model
D5 detail signal layer forecast result [0 0 P0D5]。
Preferably, the D4 detail signal layer information after wavelet decomposition respectively by the temperature T of prediction, humidity H, wind speed WS
[T0D4 H0D4 WS0D4] input vector is used as to obtain photovoltaic plant output power according to D4 detail signal layer neural network model
D4 detail signal layer forecast result [0 0 P0D4]。
Step 105, the forecast result of the forecast result to approximation signal layer and detail signal layer is reconstructed, and is predicted
Performance number.
Preferably, the coefficient of wavelet decomposition by obtained output power P at D4, D5, A5 layers, respectively multiplied by the pre- of equivalent layer
Report is as a result, finally obtain prediction power value P for the results added of multiplication0。
As another embodiment that can refer to of the present invention, as shown in fig.2, described based on neural network, wavelet decomposition
Predicting power of photovoltaic plant method can be following process:
Step 201, the historical record of photovoltaic plant position in certain time period is obtained, wherein historical record includes light
According to intensity C, output power P, temperature T, humidity H, wind speed WS.
Step 202, respectively to the intensity of illumination C of acquisition, output power P, temperature T, humidity H, wind speed WS historical data into
Row wavelet decomposition.
Preferably, respectively obtaining intensity of illumination C, output power P, temperature T, humidity H, wind speed WS using 5 layers of wavelet decomposition
Each leisure D1, D2, D3, D4, D5, A5 layer signal.Meanwhile obtaining output power P every layer of wavelet decomposition after 5 layers of wavelet decomposition
Coefficient.
Preferably, one group of original series containing integrated information are resolved into multiple groups not by wavelet filter by wavelet decomposition
With the time series of feature, one group of signal reflects in former time series in variation tendency, i.e. approximation signal；The sequence of remaining group is anti-
Reflect the influence of random perturbation bring, i.e. detail signal.The wavelet decomposition algorithm of time series are as follows:
In formula, t --- time series sequence number；F (t) --- original series；J --- Decomposition order, j=1,2 ..., J,
J=log2N, N are sequence length；H, the filter of G --- wavelet decomposition；Aj--- signal f (t) is in the approximation signal portion of jth layer
Divide the coefficient of wavelet decomposition of (i.e. low frequency part)；Dj--- signal f (t) is in the detail signal part (i.e. high frequency section) of jth layer
Coefficient of wavelet decomposition.
Step 203, by intensity of illumination C, temperature T, humidity H, wind speed WS the A5 approximation signal after wavelet decomposition respectively
Layer information [CA5 TA5 HA5 WSA5] it is used as input vector, by A5 approximation signal layer information of the output power P after wavelet decomposition
[0 0 0 PA5] it is used as output vector, establish A5 approximation signal layer neural network model.
The D5 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD5 HD5 WSD5]
As input vector, by D5 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD5] it is used as output vector,
Establish D5 detail signal layer neural network model.
The D4 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD4 HD4 WSD4]
As input vector, by D4 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD4] it is used as output vector,
Establish D4 detail signal layer neural network model.
Step 204, the weather forecast value of the photovoltaic plant position future prediction time section is obtained, the weather is pre-
Report value includes: temperature T, humidity H, wind speed WS.
Preferably, obtaining the predicted value of the temperature T in 72 hours futures of photovoltaic plant position, humidity H, wind speed WS.
Step 205, the temperature T of the future prediction time section that will acquire, humidity H, wind speed WS predicted value respectively through small echo
It decomposes.
Preferably, using 5 layers of wavelet decomposition, respectively obtain each leisure D1, D2 of temperature T, humidity H, wind speed WS, D3, D4,
D5, A5 layer signal.
Step 206, the A5 of the intensity of illumination C, temperature T, humidity H, wind speed WS of prediction respectively after wavelet decomposition is forced
Nearly signals layer information [C0A5 T0A5 H0A5 WS0A5] it is used as input vector, the A5 approximation signal layer nerve established according to step 203
Network model obtains forecast result [0 00 P of the A5 approximation signal layer of photovoltaic plant output power0A5]。
The D5 detail signal layer information [T after wavelet decomposition respectively by the temperature T, humidity H, wind speed WS of prediction0D5
H0D5 WS0D5] it is used as input vector, according to the D5 detail signal layer neural network model that step 203 is established, obtain photovoltaic plant
Forecast result [0 0 P of the D5 detail signal layer of output power0D5]。
The D4 detail signal layer information [T after wavelet decomposition respectively by the temperature T, humidity H, wind speed WS of prediction0D4
H0D4 WS0D4] it is used as input vector, according to the D4 detail signal layer neural network model that step 203 is established, obtain photovoltaic plant
Forecast result [0 0 P of the D4 detail signal layer of output power0D4]。
Step 207, coefficient of wavelet decomposition of the output power P obtained according to step 202 at D4, D5, A5 layers, to step
206 obtained forecast result P0A5、P0D5And P0D4It is reconstructed, obtains prediction power value P0。
Preferably, coefficient of wavelet decomposition of the output power P that step 202 is obtained at D4, D5, A5 layers, respectively multiplied by
P0D4、P0D5、P0A5, the results added of multiplication is finally obtained into prediction power value P0。
In another aspect of this invention, it provides a kind of based on neural network, wavelet decomposition predicting power of photovoltaic plant system
System based on neural network, wavelet decomposition predicting power of photovoltaic plant system successively includes history number as shown in fig.3, described
Unit 302, prediction data acquiring unit 303, power estimation unit 304 and prediction are established according to acquiring unit 301, prediction model
Power reconfiguration unit 305.Wherein, historical data acquiring unit 301 is for obtaining photovoltaic plant position in a period
Historical record.Wherein, the historical record includes intensity of illumination C, output power P, temperature T, humidity H, wind speed WS.Predict mould
Type establishes unit 302 for the intensity of illumination C, output power P, temperature T, humidity H, wind speed WS to be carried out small wavelength-division respectively
Solution.According to the information after the wavelet decomposition, approximation signal layer neural network model and detail signal layer neural network mould are established
Type.Later, prediction data acquiring unit 303 is used to obtain the weather forecast of photovoltaic plant position future prediction time section
Value.Wherein, the weather forecast value includes temperature T, humidity H, wind speed WS.Finally, power estimation unit 304 will be for will be described
The temperature T of future prediction time section, humidity H, wind speed WS predicted value respectively through wavelet decomposition.After the wavelet decomposition
Information obtains approximation signal layer using the information of approximation signal layer as the input vector of approximation signal layer neural network model
Forecast result；Using the information of detail signal layer as the input vector of detail signal layer neural network model, detail signal is obtained
The forecast result of layer.Prediction power reconfiguration unit 305 is used for the forecast to the forecast result and detail signal layer of approximation signal layer
As a result it is reconstructed, obtains prediction power value.
Preferably, prediction model establishes intensity of illumination C, output power P, temperature T, the humidity H, wind that unit 302 will acquire
Fast WS carries out wavelet decomposition respectively, using 5 layers of wavelet decomposition, respectively obtain intensity of illumination C, output power P, temperature T, humidity H,
The information of wind speed WS each comfortable detail signal layer D1, D2, D3, D4, D5 and the information of approximation signal layer A5.
Preferably, it is by intensity of illumination C, temperature that prediction model, which establishes unit 302 and establishes approximation signal layer neural network model,
Spend T, humidity H, wind speed WS the A5 approximation signal layer information [C after wavelet decomposition respectivelyA5 TA5 HA5 WSA5] as input to
Amount, by A5 approximation signal layer information [0 00 Ps of the output power P after wavelet decompositionA5] it is used as output vector, it establishes A5 and forces
Nearly signals layer neural network model.
As another embodiment, the detail signal layer neural network model that prediction model establishes the foundation of unit 302 can be with
D5 detail signal layer neural network model and D4 detail signal layer neural network model are established respectively.Preferably, D5 details is established
Signals layer neural network model can be by by temperature T, humidity H, wind speed WS the D5 detail signal after wavelet decomposition respectively
Layer information [TD5 HD5 WSD5] it is used as input vector, by D5 detail signal layer information [0 of the output power P after wavelet decomposition
0 PD5] it is used as output vector, establish D5 detail signal layer neural network model.In addition, temperature T, humidity H, wind speed WS are distinguished
D4 detail signal layer information [T after wavelet decompositionD4 HD4 WSD4] it is used as input vector, output power P is passed through into small echo
D4 detail signal layer information [0 0 P after decompositionD4] it is used as output vector, establish D4 detail signal layer neural network model.
It is worth noting that power estimation unit 304 is by the temperature T, humidity H, wind speed WS of the future prediction time section
Predicted value respectively through wavelet decomposition, using 5 layers of wavelet decomposition, respectively obtain each comfortable details letter of temperature T, humidity H, wind speed WS
Number information of floor D1, D2, D3, D4, D5 and the information of approximation signal floor A5.Meanwhile output power P is obtained in 5 layers of small wavelength-division
Every layer of coefficient of wavelet decomposition after solution.
Preferably, the A5 of the intensity of illumination C, temperature T, humidity H, wind speed WS of prediction respectively after wavelet decomposition is approached
Signals layer information [C0A5 T0A5 H0A5 WS0A5] input vector is used as to obtain light according to A5 approximation signal layer neural network model
Forecast result [0 00 P of the A5 approximation signal layer of overhead utility output power0A5]。
And power estimation unit 304 is refreshing using the information of D5, D4 detail signal layer of prediction as D5 detail signal layer
The input vector of input vector, D4 detail signal layer neural network model through network model, obtains the pre- of D5 detail signal layer
Report the forecast result of result and D4 detail signal layer.
Preferably, the D5 detail signal layer information after wavelet decomposition respectively by the temperature T of prediction, humidity H, wind speed WS
[T0D5 H0D5 WS0D5] input vector is used as to obtain photovoltaic plant output power according to D5 detail signal layer neural network model
D5 detail signal layer forecast result [0 0 P0D5]。
Preferably, the D4 detail signal layer information after wavelet decomposition respectively by the temperature T of prediction, humidity H, wind speed WS
[T0D4 H0D4 WS0D4] input vector is used as to obtain photovoltaic plant output power according to D4 detail signal layer neural network model
D4 detail signal layer forecast result [0 0 P0D4]。
As another embodiment of photovoltaic plant output power forecasting system of the present invention, power estimation unit 304
Coefficient of wavelet decomposition by obtained output power P at D4, D5, A5 layers, respectively multiplied by the forecast result of equivalent layer, finally by phase
The results added multiplied obtains prediction power value P0。
It should be noted that being based on neural network, wavelet decomposition predicting power of photovoltaic plant system of the present invention
Specific implementation content, it is described above based in neural network, wavelet decomposition predicting power of photovoltaic plant method in detail
It describes in detail clear, therefore no longer illustrates in this duplicate contents.
It proposes in conclusion the present invention is innovative based on neural network, wavelet decomposition predicting power of photovoltaic plant side
Method and system can be effectively prevented the algorithm because of caused by the periodicity of photovoltaic plant output power and non-stationary characteristic and fall into not
The problem of convergence；The periodicity and non-stationary information for effectively extracting photovoltaic plant output power, pass through theoretical irradiation and environment
The low-frequency information of parameter establishes more accurate model；At the same time, effectively rejecting can not be built using existing numerical weather forecast
The high fdrequency component of mould；In addition, creatively ingenious combine neural net model establishing with wavelet decomposition；Also, it ensure that algorithm
Convergence, reduce computation complexity, improve the precision of algorithm.Finally, described based on neural network, wavelet decomposition photovoltaic
Power station power forecasting method and systems approach are clear, are easily achieved.
It should be understood by those ordinary skilled in the art that: the above is only a specific embodiment of the present invention, and
It is not used in the limitation present invention, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done,
It should be included within protection scope of the present invention.
Claims (2)
1. one kind is based on neural network, wavelet decomposition predicting power of photovoltaic plant method, which is characterized in that comprising steps of
Obtain the historical record of photovoltaic plant position in a period；Wherein, the historical record includes intensity of illumination
C, output power P, temperature T, humidity H, wind speed WS；
The intensity of illumination C, output power P, temperature T, humidity H, wind speed WS are subjected to wavelet decomposition respectively；According to the small echo
Information after decomposition establishes approximation signal layer neural network model and detail signal layer neural network model；
Obtain the weather forecast value of photovoltaic plant position future prediction time section；Wherein, the weather forecast value includes
Temperature T, humidity H, wind speed WS；
By the temperature T of the future prediction time section, humidity H, wind speed WS predicted value respectively through wavelet decomposition；According to described small
Information after Wave Decomposition is forced using the information of approximation signal layer as the input vector of approximation signal layer neural network model
The forecast result of nearly signals layer；Using the information of detail signal layer as the input vector of detail signal layer neural network model, obtain
To the forecast result of detail signal layer；
The forecast result of forecast result and detail signal layer to approximation signal layer is reconstructed, and obtains prediction power value；
Intensity of illumination C, output power P, temperature T, humidity H, the wind speed WS that will acquire carry out wavelet decomposition respectively, small using 5 layers
Wave Decomposition, respectively obtain each comfortable detail signal layer D1, D2 of intensity of illumination C, output power P, temperature T, humidity H, wind speed WS,
The information of D3, D4, D5 and the information of approximation signal layer A5；Meanwhile output power P is obtained every layer after 5 layers of wavelet decomposition
Coefficient of wavelet decomposition；
The information according to after the wavelet decomposition, establishing approximation signal layer neural network model includes:
The A5 approximation signal layer information [C after wavelet decomposition respectively by intensity of illumination C, temperature T, humidity H, wind speed WSA5 TA5
HA5 WSA5] it is used as input vector, by A5 approximation signal layer information [0 00 Ps of the output power P after wavelet decompositionA5] make
For output vector, A5 approximation signal layer neural network model is established；
In addition, the detail signal layer neural network model of establishing includes:
The D5 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD5 HD5 WSD5] conduct
Input vector, by D5 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD5] it is used as output vector, it establishes
D5 detail signal layer neural network model；
The D4 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD4 HD4 WSD4] conduct
Input vector, by D4 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD4] it is used as output vector, it establishes
D4 detail signal layer neural network model；
It is described by the temperature T of the future prediction time section, humidity H, wind speed WS predicted value respectively through wavelet decomposition, using 5
Layer wavelet decomposition, respectively obtain temperature T, humidity H, wind speed WS each comfortable detail signal layer D1, D2, D3, D4, D5 information and
The information of approximation signal layer A5；
In addition, it is described using the information of approximation signal layer as the input vector of approximation signal layer neural network model, it is approached
The forecast result of signals layer include: by the intensity of illumination C of prediction, temperature T, humidity H, wind speed WS respectively after wavelet decomposition
A5 approximation signal layer information [C0A5 T0A5 H0A5 WS0A5] it is used as input vector, according to A5 approximation signal layer neural network model,
Obtain forecast result [0 00 P of the A5 approximation signal layer of photovoltaic plant output power0A5]；
In addition, obtaining detail signal using the information of detail signal layer as the input vector of detail signal layer neural network model
The forecast result of layer includes: that the D5 detail signal layer after wavelet decomposition is believed respectively by the temperature T, humidity H, wind speed WS of prediction
Cease [T0D5 H0D5 WS0D5] input vector is used as to obtain photovoltaic plant output work according to D5 detail signal layer neural network model
Forecast result [0 0 P of the D5 detail signal layer of rate0D5]；
The D4 detail signal layer information [T after wavelet decomposition respectively by the temperature T, humidity H, wind speed WS of prediction0D4 H0D4
WS0D4] it is used as input vector, according to D4 detail signal layer neural network model, obtain the D4 details letter of photovoltaic plant output power
Forecast result [0 0 P of number floor0D4]；
It is described that the forecast result of approximation signal layer and the forecast result of detail signal layer are reconstructed, comprising: defeated by what is obtained
Coefficient of wavelet decomposition of the power P at D4, D5, A5 layers out, respectively multiplied by P0D4、P0D5、P0A5, finally the results added of multiplication is obtained
Obtain prediction power value P0。
2. one kind is based on neural network, wavelet decomposition predicting power of photovoltaic plant system characterized by comprising
Historical data acquiring unit, for obtaining the historical record of photovoltaic plant position in a period；Wherein, described
Historical record includes intensity of illumination C, output power P, temperature T, humidity H, wind speed WS；
Prediction model establishes unit, for carrying out the intensity of illumination C, output power P, temperature T, humidity H, wind speed WS respectively
Wavelet decomposition；According to the information after the wavelet decomposition, approximation signal layer neural network model and detail signal layer nerve are established
Network model；
Prediction data acquiring unit, for obtaining the weather forecast value of photovoltaic plant position future prediction time section；Wherein,
The weather forecast value includes temperature T, humidity H, wind speed WS；
Power estimation unit, for by the temperature T of the future prediction time section, humidity H, wind speed WS predicted value respectively through small
Wave Decomposition；According to the information after the wavelet decomposition, using the information of approximation signal layer as approximation signal layer neural network model
Input vector, obtain the forecast result of approximation signal layer；Using the information of detail signal layer as detail signal layer neural network
The input vector of model obtains the forecast result of detail signal layer；
Prediction power reconfiguration unit, the forecast result for forecast result and detail signal layer to approximation signal layer carry out weight
Structure obtains prediction power value；
The prediction model establish the intensity of illumination C that unit will acquire, output power P, temperature T, humidity H, wind speed WS respectively into
Row wavelet decomposition respectively obtains intensity of illumination C, output power P, temperature T, humidity H, wind speed WS respectively using 5 layers of wavelet decomposition
In the information of detail signal layer D1, D2, D3, D4, D5 and the information of approximation signal layer A5；Meanwhile output power P is obtained 5
Every layer of coefficient of wavelet decomposition after layer wavelet decomposition；
The prediction model establishes unit according to the information after the wavelet decomposition, establishes approximation signal layer neural network model packet
It includes:
The A5 approximation signal layer information [C after wavelet decomposition respectively by intensity of illumination C, temperature T, humidity H, wind speed WSA5 TA5
HA5 WSA5] it is used as input vector, by A5 approximation signal layer information [0 00 Ps of the output power P after wavelet decompositionA5] make
For output vector, A5 approximation signal layer neural network model is established；
In addition, the detail signal layer neural network model of establishing includes:
The D5 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD5 HD5 WSD5] conduct
Input vector, by D5 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD5] it is used as output vector, it establishes
D5 detail signal layer neural network model；
The D4 detail signal layer information [T after wavelet decomposition respectively by temperature T, humidity H, wind speed WSD4 HD4 WSD4] conduct
Input vector, by D4 detail signal layer information [0 0 Ps of the output power P after wavelet decompositionD4] it is used as output vector, it establishes
D4 detail signal layer neural network model；
The prediction data acquiring unit distinguishes the predicted value of the temperature T of the future prediction time section, humidity H, wind speed WS
Through wavelet decomposition, using 5 layers of wavelet decomposition, respectively obtain each comfortable detail signal layer D1, D2 of temperature T, humidity H, wind speed WS,
The information of D3, D4, D5 and the information of approximation signal layer A5；
In addition, it is described using the information of approximation signal layer as the input vector of approximation signal layer neural network model, it is approached
The forecast result of signals layer include: by the intensity of illumination C of prediction, temperature T, humidity H, wind speed WS respectively after wavelet decomposition
A5 approximation signal layer information [C0A5 T0A5 H0A5 WS0A5] it is used as input vector, according to A5 approximation signal layer neural network model,
Obtain forecast result [0 00 P of the A5 approximation signal layer of photovoltaic plant output power0A5]；
In addition, obtaining detail signal using the information of detail signal layer as the input vector of detail signal layer neural network model
The forecast result of layer includes: that the D5 detail signal layer after wavelet decomposition is believed respectively by the temperature T, humidity H, wind speed WS of prediction
Cease [T0D5 H0D5 WS0D5] input vector is used as to obtain photovoltaic plant output work according to D5 detail signal layer neural network model
Forecast result [0 0 P of the D5 detail signal layer of rate0D5]；
The D4 detail signal layer information [T after wavelet decomposition respectively by the temperature T, humidity H, wind speed WS of prediction0D4 H0D4
WS0D4] it is used as input vector, according to D4 detail signal layer neural network model, obtain the D4 details letter of photovoltaic plant output power
Forecast result [0 0 P of number floor0D4]；
The forecast result of approximation signal layer and the forecast result of detail signal layer is reconstructed in the prediction power reconfiguration unit,
It include: the coefficient of wavelet decomposition of the output power P that will obtain at D4, D5, A5 layers, respectively multiplied by P0D4、P0D5、P0A5, finally by phase
The results added multiplied obtains prediction power value P0。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510867065.0A CN105528517B (en) | 2015-12-01 | 2015-12-01 | Based on neural network, wavelet decomposition predicting power of photovoltaic plant method, system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510867065.0A CN105528517B (en) | 2015-12-01 | 2015-12-01 | Based on neural network, wavelet decomposition predicting power of photovoltaic plant method, system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105528517A CN105528517A (en) | 2016-04-27 |
| CN105528517B true CN105528517B (en) | 2019-02-26 |
Family
ID=55770738
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510867065.0A Active CN105528517B (en) | 2015-12-01 | 2015-12-01 | Based on neural network, wavelet decomposition predicting power of photovoltaic plant method, system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105528517B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107194496A (en) * | 2017-04-25 | 2017-09-22 | 甘肃省电力公司风电技术中心 | The Forecasting Methodology and prediction meanss of photovoltaic power |
| CN109840633B (en) * | 2019-01-29 | 2021-03-23 | 合肥工业大学 | Photovoltaic output power prediction method, system and storage medium |
| CN111932007A (en) * | 2020-08-04 | 2020-11-13 | 广州发展新能源股份有限公司 | Power prediction method and device for photovoltaic power station and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6466924B1 (en) * | 1997-12-02 | 2002-10-15 | Denso Corporation | Verification method of neural network and verification apparatus thereof |
| CN104992248A (en) * | 2015-07-07 | 2015-10-21 | 中山大学 | Microgrid photovoltaic power station generating capacity combined forecasting method |
-
2015
- 2015-12-01 CN CN201510867065.0A patent/CN105528517B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6466924B1 (en) * | 1997-12-02 | 2002-10-15 | Denso Corporation | Verification method of neural network and verification apparatus thereof |
| CN104992248A (en) * | 2015-07-07 | 2015-10-21 | 中山大学 | Microgrid photovoltaic power station generating capacity combined forecasting method |
Non-Patent Citations (2)
| Title |
|---|
| 基于小波分析与神经网络的光伏电站功率预测方法;朱红路等;《太阳能学报》;20151130;第36卷(第11期);第2725-2730页 |
| 基于小波变换的神经网络的光伏功率预测;张立影等;《可再生能源》;20150220;第33卷(第2期);第171-176页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105528517A (en) | 2016-04-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Singh et al. | Generalized neural network methodology for short term solar power forecasting | |
| Wang et al. | Analyzing the research subjects and hot topics of power system reliability through the Web of Science from 1991 to 2015 | |
| Sun et al. | Research on ultra-short-term wind power prediction considering source relevance | |
| CN102419394B (en) | Wind/solar power prediction method with variable prediction resolution | |
| CN110889603A (en) | Power system economic dispatching method considering wind power correlation based on PCA-Copula theory | |
| CN107609697A (en) | A kind of Wind power forecasting method | |
| CN105373856A (en) | Wind electricity power short-term combined prediction method considering run detection method reconstruction | |
| CN103077300A (en) | Forecasting method for generating capacity of distributed photovoltaic power supply on basis of type-2 fuzzy logic | |
| Yang et al. | Short-term PV generation system direct power prediction model on wavelet neural network and weather type clustering | |
| CN104268659A (en) | Photovoltaic power station generated power super-short-term prediction method | |
| CN109255728A (en) | The photovoltaic generation power neural network prediction method of chaos phase space optimal reconfiguration | |
| CN103577891A (en) | Multi-island micro-grid optimization cooperation running method containing distributed power source | |
| CN105528517B (en) | Based on neural network, wavelet decomposition predicting power of photovoltaic plant method, system | |
| CN109840633A (en) | Photovoltaic output power predicting method, system and storage medium | |
| CN106130004A (en) | A kind of also site new forms of energy considering stability characteristic (quality) receive the appraisal procedure of ability | |
| Du et al. | Deep power forecasting model for building attached photovoltaic system | |
| Vlasov et al. | Predictive control algorithm for A variable load hybrid power system on the basis of power output forecast | |
| CN114301089A (en) | Energy storage capacity configuration optimization method for wind-solar combined power generation system | |
| CN106485603A (en) | A kind of short-term wind speed forecasting method being applied to wind-power electricity generation | |
| CN105354761B (en) | Safety and efficiency evaluation method and system for accessing wind power into power grid | |
| CN105207255B (en) | A kind of power system peak regulation computational methods suitable for wind power output | |
| CN106339773A (en) | Sensitivity-based active power distribution network distributed power source constant-capacity planning method | |
| Zhen et al. | Multi-level wavelet decomposition based day-ahead solar irradiance forecasting | |
| Li et al. | Dynamic reconfiguration of distribution network considering time-varying characteristics of DG | |
| CN113344283A (en) | Energy internet new energy consumption capacity assessment method based on edge intelligence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20190605 Address after: 100085 Beijing city Haidian District Qinghe small Camp Road No. 15 Co-patentee after: Fibrlink Networks Co., Ltd. Patentee after: Beijing China Power Information Technology Co., Ltd. Address before: 100070 the 28 tier of fortune Fortune Plaza, No.1, hang Feng Road, Fengtai District, Beijing. Co-patentee before: Fibrlink Networks Co., Ltd. Patentee before: Beijing Guodiantong Network Technology Co., Ltd. |