CN104484833A - Photovoltaic power generation output power tracking algorithm based on genetics algorithm improved RBF-BP neural network - Google Patents

Abstract

The invention discloses a photovoltaic power generation output power tracking algorithm based on a genetics algorithm improved RBF-BP neural network. By building an RBF-BP neural network, an error absolute value between predicted output and expected output of photovoltaic power generation output power is taken as the fitness, and then a genetics algorithm is adopted for selecting, intersecting and mutating data acquired by photovoltaic power generation equipment in order to find out an individual corresponding to the optimal the fitness. The photovoltaic power generation output power tracking algorithm disclosed by the invention combines the advantages that an RBF neural network is high in rate of convergence, good in heap sort performance and the BP neural network is high in self-learning and self-adaptive capabilities, and has the characteristics of better generalization performance, higher rate of convergence, higher prediction precision and the like.

Description

Based on the photovoltaic generation output power tracing algorithm of the RBF-BP neural network that genetic algorithm is improved

Technical field

The present invention relates to a kind of tracing algorithm of photovoltaic generation output power, particularly relate to a kind of photovoltaic generation output power tracing algorithm of the RBF-BP neural network based on genetic algorithm improvement.

Background technology

The problem of environmental pollution caused along with traditional energy consumption becomes increasingly conspicuous, and the utilization of regenerative resource causes to be paid attention to widely.Photovoltaic generation, as a kind of regenerative resource form of emerging emergence, has DEVELOPMENT PROSPECT and commercial value widely.Thus increasing concern is obtained.Large-scale parallel network power generation is the main trend of current photovoltaic generating system, and large-scale photovoltaic parallel in system is applied at present.

Photovoltaic generation is the same with wind-power electricity generation, all belongs to undulatory property and intermittent power supply.Photovoltaic generation is comparatively large by such environmental effects, and particularly intensity of solar radiation, environment temperature etc., because its output power has uncertainty.It makes the integral load forecasting accuracy of bulk power grid reduce after being incorporated to bulk power grid, also must cause the voltage of whole system, the fluctuation of frequency, add conventional electric power generation, and the difficulty of control and operational plan, is unfavorable for the scheduling of whole network system.

So the prediction photovoltaic generation output power of efficiently and accurately just seems particularly necessary for utilizing photovoltaic generation safely and efficiently.The Forecasting Methodology of current photovoltaic generation output power thankss for your hospitality dynamic observation, conductance increment method and neural network etc.It is disturbance observation and or conductance increment method all also exists the fixing problem of a step-length.If step-length is too small, photovoltaic array just can be caused to be trapped in low-power output area for a long time, and step-length is excessive, and system oscillation will be caused to aggravate.Its tracing process of artificial neural network does not need the physical parameter of photovoltaic battery array by contrast, and neural network is by the ability of any Nonlinear Characteristic Curve of learning training programmable single-chip system.So use neural network model in the process of the mod eling and identification of nonlinear system, can not by the restriction of nonlinear model, better adaptability.Neural network mainly comprises RBF and BP neural network prediction method, and they have respective relative merits.But the fast convergence rate generalization ability of RBF neural prediction is poor; But the strong learning algorithm of the self-learning capability of BP neural network can not ensure the misguidance that the result learnt reaches the global minima of square error, training result is easily subject to incorrect training sample set.And both threshold value and weights in the process building neural network need repetition training to revise, and initial value has randomness.

Summary of the invention

In order to overcome RBF generalization ability difference and BP neural metwork training result is easily subject to the defect of the misguidance of incorrect training sample set, the present invention proposes a kind of can the photovoltaic generation output power tracing algorithm of the RBF-BP neural network based on genetic algorithm improvement of prediction photovoltaic generation output power of efficiently and accurately.

The photovoltaic generation output power tracing algorithm of the RBF-BP neural network based on genetic algorithm improvement that the present invention proposes is by setting up a RBF-BP neural network, using the Error Absolute Value between photovoltaic generation output power prediction output with desired output as fitness, then the data using genetic algorithm to collect photovoltaic power generation equipment are selected, the individuality that optimal-adaptive degree is corresponding is found in crossover and mutation operation.RBF-BP neural network prediction genetic algorithm obtains optimum individual to network initial weight and threshold value assignment, predicts after network training to light method generating output power.

The photovoltaic generation output power predicting method of the described RBF-BP neural network based on genetic algorithm improvement, comprises the steps:

(1) according to photovoltaic generation output characteristics, add up under choosing daily weather conditions and gather the factor that each period among one day affects the generating of photovoltaic generation cell panel, intensity of illumination when laying particular emphasis on photovoltaic battery panel working temperature and photovoltaic power generation equipment work in the present invention and photovoltaic generation output power are as the input and output of RBF-BP neural metwork training.Alternative gets the test data of sample data as RBF-BP neural network of each period under equal conditions.

(2) RBF-BP neural network is set up for training the sample data after normalization.The RBF-BP combination neural net that the present invention proposes to set up combines one that the advantages such as the good and BP Neural Network Self-learning of RBF neural fast convergence rate, heap sort performance, adaptive ability are strong set up to form two hidden-layer RBF-BP combination neural net by RBF subnet and BP subnet two parts, it has, and Generalization Capability is better, speed of convergence sooner, precision of prediction more high.The RBF-BP combination neural net that the present invention proposes is divided into: input layer, hidden layer and output layer, and the nodes design of each layer is as follows:

Input layer: for the prediction of photovoltaic maximum power, when photovoltaic maximum power point being affected when the weather condition not considering to suddenly change and local uneven illumination are even, choosing of neural network input quantity mainly considers two parts, the intensity of illumination namely when photovoltaic battery panel working temperature and photovoltaic power generation equipment work.

Hidden layer: the RBF-BP neural network that the present invention relates to, compared to traditional BP neural network, adds RBF neural as sublayer at hidden layer.Sample in input step (1) is first trained through RBF neural subnet, then the input of training result as BP subnet is trained it.The transport function of the hidden layer node of RBF-BP subnet is set to Gaussian function, shown in (1):

$u_i\left(X\right)=\exp [-{\left|\right|X-c_i\left|\right|}^2/{{2\mathrm{\σ}}_i}^2](i=\mathrm{1,2},...,N_2)---\left(1\right)$

Wherein, u _i(X) be the output of i-th hidden node, X is step (1) input amendment, c _ithe center vector of Gaussian function, σ _ifor the base width parameter of node, and for being greater than the number of zero.

Using the input of the output of RBF subnet as the BP subnet in combination neural net.The transport function of its hidden layer node is designed to Sigmoid type function, such as formula (2):

f(x)＝1/(1+e ^-x) (2)

Thus the output of the hidden node of BP sub-network is formula (3):

$O_j=f\left[\underset{i=1}{\overset{N_2}{\mathrm{\Σ}}}{W}_{\mathrm{ij}}{u}_i\left(X\right)\right]---\left(3\right)$

Wherein, W _iji-th, sublayer node to imply a jth node of sublayer weights to BP are implied, N for connecting RBF ₂for RBF implies sublayer nodes.

Output layer: the node number of output layer can according to circumstances be determined.But in order to simplify the design of neural network, the present invention have selected an output node, i.e. the voltage at maximum power point place.The output valve formula calculating output layer node is formula (4):

$y_k=f\left[\underset{j=1}{\overset{N_3}{\mathrm{\Σ}}}{W}_{\mathrm{jk}}{O}_j\right]---\left(4\right)$

Wherein, W _jkthe weights of a jth node in sublayer to an output layer kth node are implied, N for connecting BP ₃for BP implies sublayer nodes.

F _ithe Error Absolute Value calculating output layer with desired output is exported, such as formula (5) according to reality:

$E=\underset{k=1}{\overset{N_4}{\mathrm{\Σ}}}|d_k-y_k|---\left(5\right)$

Wherein, d _kfor desired output.N ₄for output layer nodes, E is Error Absolute Value.

(3) data that step (1) collects are brought in the RBF-BP neural network set up in step (2), obtain the error E of actual output and desired output.As the fitness value for genetic algorithm.

(4) network structure of RBF-BP neural network is determined according to the fitting function input/output argument number of the RBF-BP neural network built in step (2), determine the number of threshold value in RBF-BP neural network and weights, and then determine the length of individuality of genetic algorithm.

(5) the data genetic algorithm that step (1) collects is optimized, concrete implementation step:

A. initialization of population

Each individuality in population comprises all weights and threshold of whole RBF-BP neural network, and individual UVR exposure method is real coding, and each individuality is a real number string, calculates individual fitness value individual by genetic algorithm fitness function.

B. fitness function

Initial weight and the threshold value of RBF-BP neural network is obtained, with the absolute error value E obtained in step (3) as ideal adaptation angle value F according to the individuality of step (1).

C. operation is selected

Genetic algorithm selection operation has the multiple method such as roulette method, championship, and the present invention selects roulette method, the select probability p of each individual i _ifor formula (6):

$\left\{\begin{array}{c}{f}_i=k/F_i\\ p_i=\frac{f_i}{\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{f}_i}\end{array}\right.---\left(6\right)$

F in formula _ifor the fitness value of individual i, because fitness value is the smaller the better, so ask reciprocal to fitness value before individual choice; K is coefficient; N is population at individual number.

D. interlace operation

Because individuality adopts real coding, so interlace operation method also adopts corresponding real number bracketing method, a kth chromosome a _kwith l chromosome a _linterlace operation method in j position is as formula (7):

$\left.\begin{array}{c}{a}_{\mathrm{kj}}=a_{\mathrm{kj}}(1-b)+a_{\mathrm{lj}}b\\ a_{\mathrm{lj}}=a_{\mathrm{lj}}(1-b)+a_{\mathrm{kj}}b\end{array}\right\}---\left(7\right)$

Wherein b is the random number between [0,1].

E. mutation operation

Choose i-th individual jth gene a _ijmake a variation, the individual formula of mutation operation is as (8):

$a_{\mathrm{ij}}=\left\{\begin{array}{cc}{a}_{\mathrm{ij}}=a_{\mathrm{ij}}+(a_{\mathrm{ij}}-a_{\max })*f\left(g\right)& r>0.5\\ a_{\mathrm{ij}}=a_{\mathrm{ij}}+(a_{\min }-a_{\mathrm{ij}})*f\left(g\right)& r\≤0.5\end{array}\right.---\left(8\right)$

In formula (8), a _maxfor gene a _ijthe upper bound; a _minfor gene a _ijlower bound; F (g)=r ₂(1-g/G _max) ²; r ₂it is a random number; G is current iteration number of times; G _maxit is maximum evolution number of times; R is the random number between [0,1].

Further, also comprise:

Find optimal-adaptive angle value corresponding individual by the genetic algorithm of step (4) by selection, crossover and mutation operation the training data of step (1).The optimum individual that RBF-BP neural network genetic algorithm obtains is to networking initial weight and threshold value assignment.RBF-BP neural network after test data test step (1) gathered improves for genetic algorithm, when neural metwork training error is less than target error, network convergence; When network training number of times equals maximum iteration time, training error is still greater than target error, network is not restrained.Now again by the backward learning ability of RBF-BP neural network, the oppositely weights and threshold of amendment neural network, adjustment formula is as (9), (10):

W _jk＝W _jk+λy _k(1-y _k)(d _k-y _k)O _j(9)

θ _k＝θ _k+λy _k(1-y _k)(d _k-y _k) (10)

λ is learning rate.RBF-BP neural network after training can be used for photovoltaic generation peak power output and follows the trail of.

Advantage of the present invention is the tracking for photovoltaic generation output power:

(1) establish one and form two hidden-layer RBF-BP combination neural net by RBF subnet and BP subnet two parts, combine the advantages such as the good and BP Neural Network Self-learning of RBF neural fast convergence rate, heap sort performance, adaptive ability are strong, have that Generalization Capability is better, speed of convergence is faster, precision of prediction more high.

(2) it is individual that utilization genetic algorithm is selected sample data, crossover and mutation operation obtains optimal-adaptive degree, for optimizing RBF-BP neural network further.

(3) not only can the prediction photovoltaic generation output power of efficiently and accurately by the combination of two hidden-layer RBF-BP combination neural net and genetic algorithm, utilize photovoltaic generation safely and efficiently, better can tackle the undulatory property in photovoltaic generation and intermittent power supply simultaneously.

Accompanying drawing explanation

Fig. 1: RBF-BP network structure;

Fig. 2: the algorithm flow chart after genetic algorithm improvement.

Embodiment

As shown in Figure 1, RBF-BP neural network have two hidden layers, input data are trained as the input of BP neural network subnet via after the training of RBF neural subnet, this network has the ability of error backward learning, when training result does not reach accuracy requirement, can oppositely revise the weights and threshold of neural network until training result reaches accuracy requirement.

As shown in Figure 2, the RBF-BP neural network algorithm improved based on genetic algorithm goes when confirming network structure, confirms to export weight threshold length.Using the Error Absolute Value that exports between desired output as fitness, then use genetic algorithm to select data, crossover and mutation operation finds the individuality that optimal-adaptive degree is corresponding, and then confirms the weights and threshold of RBF-BP neural network.After test training, whether the error of network meets the accuracy requirement of prediction photovoltaic generation power.

The photovoltaic generation output power predicting method of the described RBF-BP neural network based on genetic algorithm improvement, is characterized in that comprising the steps:

(1) according to photovoltaic generation output characteristics, add up under choosing daily weather conditions and gather the factor that each period among one day affects the generating of photovoltaic generation cell panel, intensity of illumination when laying particular emphasis on photovoltaic battery panel working temperature and photovoltaic power generation equipment work in the present invention and photovoltaic generation output power are as the input and output of RBF-BP neural metwork training.Alternative gets the test data of sample data as RBF-BP neural network of each period under equal conditions.

(2) RBF-BP neural network is set up for training the sample data after normalization.The RBF-BP combination neural net that the present invention proposes to set up combines one that the advantages such as the good and BP Neural Network Self-learning of RBF neural fast convergence rate, heap sort performance, adaptive ability are strong set up to form two hidden-layer RBF-BP combination neural net by RBF subnet and BP subnet two parts, it has, and Generalization Capability is better, speed of convergence sooner, precision of prediction more high.The RBF-BP combination neural net that the present invention proposes is divided into: input layer, hidden layer and output layer, and the nodes design of each layer is as follows:

Input layer: for the prediction of photovoltaic maximum power, when photovoltaic maximum power point being affected when the weather condition not considering to suddenly change and local uneven illumination are even, choosing of neural network input quantity mainly considers two parts, the intensity of illumination namely when photovoltaic battery panel working temperature and photovoltaic power generation equipment work.

Hidden layer: the RBF-BP neural network that the present invention relates to, compared to traditional BP neural network, adds RBF neural as sublayer at hidden layer.Sample in input step (1) is first trained through RBF neural subnet, then the input of training result as BP subnet is trained it.The transport function of the hidden layer node of RBF-BP subnet is set to Gaussian function, shown in (1):

$u_i\left(X\right)=\exp [-{\left|\right|X-c_i\left|\right|}^2/{{2\mathrm{\σ}}_i}^2](i=\mathrm{1,2},...,N_2)---\left(1\right)$

Wherein, u _i(X) be the output of i-th hidden node, X is step (1) input amendment, c _ithe center vector of Gaussian function, σ _ifor the base width parameter of node, and for being greater than the number of zero.

Using the input of the output of RBF subnet as the BP subnet in combination neural net.The transport function of its hidden layer node is designed to Sigmoid type function, such as formula (2):

f(x)＝1/(1+e ^-x) (2)

Thus the output of the hidden node of BP sub-network is formula (3):

$O_j=f\left[\underset{i=1}{\overset{N_2}{\mathrm{\Σ}}}{W}_{\mathrm{ij}}{u}_i\left(X\right)\right]---\left(3\right)$

Wherein, W _iji-th, sublayer node to imply a jth node of sublayer weights to BP are implied, N for connecting RBF ₂for RBF implies sublayer nodes.

Output layer: the node number of output layer can according to circumstances be determined.But in order to simplify the design of neural network, the present invention have selected an output node, i.e. the voltage at maximum power point place.The output valve formula calculating output layer node is formula (4):

$y_k=f\left[\underset{j=1}{\overset{N_3}{\mathrm{\Σ}}}{W}_{\mathrm{jk}}{O}_j\right]---\left(4\right)$

Wherein, W _jkthe weights of a jth node in sublayer to an output layer kth node are implied, N for connecting BP ₃for BP implies sublayer nodes.

F _ithe Error Absolute Value calculating output layer with desired output is exported, such as formula (5) according to reality:

$E=\underset{k=1}{\overset{N_4}{\mathrm{\Σ}}}|d_k-y_k|---\left(5\right)$

Wherein, d _kfor desired output.N ₄for output layer nodes, E is Error Absolute Value.

(3) data that step (1) collects are brought in the RBF-BP neural network set up in step (2), obtain the error E of actual output and desired output.As the fitness value for genetic algorithm.

(4) network structure of RBF-BP neural network is determined according to the fitting function input/output argument number of the RBF-BP neural network built in step (2), determine the number of threshold value in RBF-BP neural network and weights, and then determine the length of individuality of genetic algorithm.

(5) the data genetic algorithm that step (1) collects is optimized, concrete implementation step:

A. initialization of population

Each individuality in population comprises all weights and threshold of whole RBF-BP neural network, and individual UVR exposure method is real coding, and each individuality is a real number string, calculates individual fitness value individual by genetic algorithm fitness function.

B. fitness function

Initial weight and the threshold value of RBF-BP neural network is obtained, with the absolute error value E obtained in step (3) as ideal adaptation angle value F according to the individuality of step (1).

C. operation is selected

Genetic algorithm selection operation has the multiple method such as roulette method, championship, and the present invention selects roulette method, the select probability p of each individual i _ifor formula (6):

$\left\{\begin{array}{c}{f}_i=k/F_i\\ p_i=\frac{f_i}{\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{f}_i}\end{array}\right.---\left(6\right)$

F in formula _ifor the fitness value of individual i, because fitness value is the smaller the better, so ask reciprocal to fitness value before individual choice; K is coefficient; N is population at individual number.

D. interlace operation

Because individuality adopts real coding, so interlace operation method also adopts corresponding real number bracketing method, a kth chromosome a _kwith l chromosome a _linterlace operation method in j position is as formula (7):

$\left.\begin{array}{c}{a}_{\mathrm{kj}}=a_{\mathrm{kj}}(1-b)+a_{\mathrm{lj}}b\\ a_{\mathrm{lj}}=a_{\mathrm{lj}}(1-b)+a_{\mathrm{kj}}b\end{array}\right\}---\left(7\right)$

Wherein b is the random number between [0,1].

E. mutation operation

Choose i-th individual jth gene a _ijmake a variation, the individual formula of mutation operation is as (8):

$a_{\mathrm{ij}}=\left\{\begin{array}{cc}{a}_{\mathrm{ij}}=a_{\mathrm{ij}}+(a_{\mathrm{ij}}-a_{\max })*f\left(g\right)& r>0.5\\ a_{\mathrm{ij}}=a_{\mathrm{ij}}+(a_{\min }-a_{\mathrm{ij}})*f\left(g\right)& r\≤0.5\end{array}\right.---\left(8\right)$

In formula (8), a _maxfor gene a _ijthe upper bound; a _minfor gene a _ijlower bound; F (g)=r ₂(1-g/G _max) ²; r ₂it is a random number; G is current iteration number of times; G _maxit is maximum evolution number of times; R is the random number between [0,1].

(6) training data of step (1) finds optimal-adaptive angle value corresponding individual by the genetic algorithm of step (4) by selection, crossover and mutation operation.The optimum individual that RBF-BP neural network genetic algorithm obtains is to networking initial weight and threshold value assignment.

(7) the RBF-BP neural network after test data test step (1) gathered improves for genetic algorithm, when neural metwork training error is less than target error, network convergence; When network training number of times equals maximum iteration time, training error is still greater than target error, network is not restrained.Now again by the backward learning ability of RBF-BP neural network, the oppositely weights and threshold of amendment neural network, adjustment formula is as (9), (10):

W _jk＝W _jk+λy _k(1-y _k)(d _k-y _k)O _j(9)

θ _k＝θ _k+λy _k(1-y _k)(d _k-y _k) (10)

In formula, λ is learning rate.RBF-BP neural network after training can be used for photovoltaic generation peak power output and follows the trail of.

The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, for a person skilled in the art, the present invention can have various modifications and variations.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (4)

1., based on a photovoltaic generation output power tracing algorithm for the RBF-BP neural network of genetic algorithm improvement, it is characterized in that, comprise step:

1) according to photovoltaic generation output characteristics, add up under choosing daily weather conditions and gather the factor that each period among one day affects the generating of photovoltaic generation cell panel, using the input of intensity of illumination when photovoltaic battery panel working temperature, photovoltaic power generation equipment work as RBF-BP neural metwork training, using the output of photovoltaic generation output power as RBF-BP neural metwork training, alternative gets the test data of sample data as RBF-BP neural network of each period under equal conditions;

2) RBF-BP neural network is set up for training the sample data after normalization, described RBF-BP neural network forms two hidden-layer RBF-BP combination neural net by RBF subnet and BP subnet two parts, be divided into: input layer, hidden layer and output layer, the nodes design of each layer is as follows:

Input layer: for the prediction of photovoltaic maximum power, when photovoltaic maximum power point being affected when the weather condition not considering to suddenly change and local uneven illumination are even, choosing of neural network input quantity mainly considers two parts, the intensity of illumination namely when photovoltaic battery panel working temperature and photovoltaic power generation equipment work;

Hidden layer: the RBF-BP neural network that the present invention relates to adds RBF neural as sublayer at hidden layer, input step 1) in sample first train through RBF neural subnet, then the input of training result as BP subnet to be trained it;

Output layer: the node number of output layer can according to circumstances be determined, but in order to simplify the design of neural network, the present invention have selected an output node, i.e. the voltage at maximum power point place;

3) by step 1) data that collect bring step 2 into) in the RBF-BP neural network set up, obtain the error E of actual output and desired output, as the fitness value for genetic algorithm;

4) according to step 2) in the fitting function input/output argument number of RBF-BP neural network that builds determine the network structure of RBF-BP neural network, determine the number of threshold value in RBF-BP neural network and weights, and then determine the length of individuality of genetic algorithm;

5) by step 1) the data genetic algorithm that collects is optimized, concrete implementation step:

A. initialization of population

Each individuality in population comprises all weights and threshold of whole RBF-BP neural network, and individual UVR exposure method is real coding, and each individuality is a real number string, calculates individual fitness value individual by genetic algorithm fitness function;

B. fitness function

According to step 1) individuality obtain initial weight and the threshold value of RBF-BP neural network, by step 3) in the absolute error value E that obtains as ideal adaptation angle value F;

C. operation is selected

Genetic algorithm selection operation has the multiple method such as roulette method, championship, and the present invention selects roulette method, the select probability p of each individual i _ifor formula (6):

F in formula _ifor the fitness value of individual i, because fitness value is the smaller the better, so ask reciprocal to fitness value before individual choice; K is coefficient; N is population at individual number;

D. interlace operation

Because individuality adopts real coding, so interlace operation method also adopts corresponding real number bracketing method, a kth chromosome a _kwith l chromosome a _linterlace operation method in j position is as formula (7):

Wherein b is the random number between [0,1];

E. mutation operation

Choose i-th individual jth gene a _ijmake a variation, the individual formula of mutation operation is as (8):

In formula (8), a _maxfor gene a _ijthe upper bound; a _minfor gene a _ijlower bound; F (g)=r ₂(1-g/G _max) ²; r ₂it is a random number; G is current iteration number of times; G _maxit is maximum evolution number of times; R is the random number between [0,1].

2. the photovoltaic generation output power tracing algorithm of the RBF-BP neural network based on genetic algorithm improvement according to claim 1, it is characterized in that, described step 2) in the transport function of hidden layer node of RBF-BP subnet be set to Gaussian function, shown in (1):

u _i(X)＝exp[-||X-c _i|| ²/2σ _i ²](i＝1，2，…，N ₂) (1)

Wherein, u _i(X) be the output of i-th hidden node, X is step (1) input amendment, c _ithe center vector of Gaussian function, σ _ifor the base width parameter of node, and for being greater than the number of zero;

Using the input of the output of RBF subnet as the BP subnet in combination neural net, the transport function of its hidden layer node is designed to Sigmoid type function, such as formula (2):

f(x)＝1/(1+e ^-x) (2)

Thus the output of the hidden node of BP sub-network is formula (3):

Wherein, W _iji-th, sublayer node to imply a jth node of sublayer weights to BP are implied, N for connecting RBF ₂for RBF implies sublayer nodes.

3. the photovoltaic generation output power tracing algorithm of RBF-BP neural network improved based on genetic algorithm according to claim 1, is characterized in that, described step 2) in calculate output layer node output valve formula be formula (4):

Wherein, W _jkthe weights of a jth node in sublayer to an output layer kth node are implied, N for connecting BP ₃for BP implies sublayer nodes;

F _ithe Error Absolute Value calculating output layer with desired output is exported, such as formula (5) according to reality:

Wherein, d _kfor desired output, N ₄for output layer nodes, E is Error Absolute Value.

4. the photovoltaic generation output power tracing algorithm of the RBF-BP neural network based on genetic algorithm improvement according to claim 1, is characterized in that, also comprise step:

A) by step 1) training data by step 4) genetic algorithm by selecting, crossover and mutation operation finds optimal-adaptive angle value corresponding individual, the optimum individual that RBF-BP neural network genetic algorithm obtains is to networking initial weight and threshold value assignment;

B) by step 1) test of the test data that gathers improve for genetic algorithm after RBF-BP neural network, when neural metwork training error is less than target error, network convergence; When network training number of times equals maximum iteration time, training error is still greater than target error, network is not restrained; Now again by the backward learning ability of RBF-BP neural network, the oppositely weights and threshold of amendment neural network, adjustment formula is as (9), (10):

W _jk＝W _jk+λy _k(1-y _k)(d _k-y _k)O _j(9)

θ _k＝θ _k+λy _k(1-y _k)(d _k-y _k) (10)

In formula, λ is learning rate, and the RBF-BP neural network after training can be used for photovoltaic generation peak power output and follows the trail of.