CN106558878A - A kind of analysis photovoltaic is exerted oneself the method for undulatory property - Google Patents

Abstract

A kind of method of undulatory property the present invention relates to analysis photovoltaic is exerted oneself, including：Collection photovoltaics power station history photovoltaic goes out force data；Input odd-numbered day weather pattern；Obtain the likelihood function of the undulating value probability distribution under the weather pattern；Maximum Likelihood Estimation based on EM estimates three-component mixed Gaussian probability distribution parameters, generates Fitted probability distribution curve；Confirm the effectiveness of fitting of distribution.By way of fitting, the photovoltaic for analyzing different weather type and different cloud layer state is exerted oneself wave characteristic, obtains the uncertain data under different weather.Contribute to instructing electric power system dispatching to run, ability and operation security and stability are received so as to improve power system photovoltaic generation.

Description

A kind of analysis photovoltaic is exerted oneself the method for undulatory property

Technical field

The present invention relates to a kind of analysis method, and in particular to a kind of analysis photovoltaic is exerted oneself the method for undulatory property.

Background technology

Photovoltaic generation is the renewable energy technologies after wind-powered electricity generation with maximum potential and using value, is matched somebody with somebody China is a series of Under set policy support, photovoltaic generation quickly grows.To the year two thousand twenty, China's Photovoltaic generation installed capacity is up to 50GW, counts Draw and increase by more than 15% in year.But the random and undulatory property feature of photovoltaic generation can bring impact to the peak regulation of electrical network and frequency modulation.Cause This, the wave characteristic that research photovoltaic is exerted oneself will be the offer such as power system peak-frequency regulation, operation control method experience and foundation.

The research of undulatory property of exerting oneself to photovoltaic at present is mainly the undulating value that statistical analysiss different time scales photovoltaic is exerted oneself, and The photovoltaic for not analyzing different weather type and different cloud layer state is exerted oneself wave characteristic.

The content of the invention

For the problems referred to above, the present invention provides a kind of method that analysis photovoltaic exerts oneself undulatory property, to different weather type lower body Now the probability distribution of the probabilistic undulating value of cloud layer state is fitted, and contributes to instructing electric power system dispatching to run, from And improve power system photovoltaic generation and receive ability and operation security and stability.

The purpose of the present invention is realized using following technical proposals：

A kind of analysis photovoltaic is exerted oneself the method for undulatory property, including：

(1) collection photovoltaics power station history photovoltaic goes out force data；

(2) it is input into odd-numbered day weather pattern；

(3) obtain the likelihood function of the undulating value probability distribution under the weather pattern；

(4) Maximum Likelihood Estimation based on EM estimates three-component mixed Gaussian probability distribution parameters, generates Fitted probability Distribution curve；

(5) confirm the effectiveness of fitting of distribution.

Preferably, the collection in the step (1) includes：The history of conversion photovoltaic is exerted oneself P (i, t), and uses following formula (1) express headroom theory to exert oneself and exert oneself relatively：

P (i, t)=P_DCI(i,t)·P_N(i,t) (1)

In formula (1), P_DCI(i, t) and P_N(i, t) is respectively i-th day t and is exerted oneself and phase based on the headroom theory of clearance model To exerting oneself.

Further, by the relative P that exerts oneself_N(i, t) is decomposed into the set of power average value and undulating value, with following formula (2) Expression：

P_N(i, t)=P_M(i)+P_F(i,t) (2)

In formula (2), P_F(i, t) is the undulating value that i-th day t is caused by different cloud layer states and state of weather, P_M(i) For the power average value of i-th day, reflection same day photovoltaic was exerted oneself degree.

Preferably, the weather pattern of the step (2) includes fine day, cloudy, cloudy and change weather.

Preferably, the step (3) includes：With (3) three points of following formula Undulating value P under each weather pattern of amount mixed Gaussian Probability Distribution Fitting_FThe likelihood function of (i, t) probability distribution：

In formula (3), α_j、μ_jAnd σ_jThe weight coefficient of respectively j-th Gaussian component, average and standard deviation (j=1,2,3), z_i∈[z₁,z₂,…z_N] represent i-th observed value；

Parameter θ is estimated in order_j=[α_j,μ_j,σ_j], (j=1,2,3), observed value Z=[z₁,z₂,…z_N], obtain likelihood function：

The both sides of formula (4) are taken the logarithm acquisition following formula (5)：

(5) in formula, z_i∈[z₁,z₂,…z_N] represent i-th observed value, numbers of the N for observed value.

Preferably, in the step (4), estimate that three-component mixed Gaussian probability distribution parameters generate Fitted probability distribution bent Line comprises the steps：

4-1 calculates posterior probability；

4-2 updates weight coefficient, average and standard deviation matrix；

4-3 judges convergence.

Further, the step (4-1) includes, calculates weight coefficient α_jPosterior probability be：

Further, the step (4-2) includes：Using the weight coefficient α in formula (6) newer (3)_j, Value μ_jWith standard deviation sigma_j：

Further, the step (4-3) judges that convergence includes：Iterative step (4-1) and (4-2), circulation is more Newly, until meeting | l (θ_j)-l'(θ_j) | ＜ ε, ε<10⁵；Wherein, l (θ_j) for undulating value P_F(i, t) probability distribution is seemingly The logarithm of right function, l'(θ_j) it is value after each iteration updates, ε is perunit value.

Preferably, the step (5) includes：Confirmed using residuals squares SSE checking three-component mixed Gaussian probability distribution The effectiveness of fitting is represented with following formula (10)：

In formula (10),For i-th observed value z_iThree-component mixed Gaussian Probability Distribution Fitting probability density function Value, y_iFor actual probabilities density value, length of the n for probability density value；If SSE is less than ε (ε=0.2～0.3), it is fitted Effectively.

Compared with immediate prior art, the beneficial effect that the present invention reaches is：

The probability distribution for dividing weather pattern that undulating value is exerted oneself using three-component mixed Gaussian Probability Distribution Fitting photovoltaic, Neng Gouzhun The feature that photovoltaic under each weather pattern goes out fluctuation is really embodied, can be exerted oneself in photovoltaic generation for a long time as early stage base application In time series modeling, it is the production analog simulation of the sequential containing extensive new forms of energy, annual new energy digestion capability analysis etc. The generation of the analogue simulation data of needs provides prior art means.

Probability distribution to the probabilistic undulating value of cloud layer state is embodied under different weather type is fitted, and contributes to referring to Conductive Force system management and running, receive ability and operation security and stability so as to improve power system photovoltaic generation.

Description of the drawings

Fig. 1 is that a kind of analysis photovoltaic proposed by the present invention is exerted oneself the method flow diagram of undulatory property；

Fig. 2 is undulating value Probability Distribution Fitting curve and actual probability distribution curve synoptic diagram under 4 kinds of weather patterns；

A () fine day, at (b) cloudy, (c) cloudy day, (d) changes weather.

Specific embodiment

A kind of method of undulatory property as shown in figure 1, analysis photovoltaic is exerted oneself, including：

(1) collection photovoltaics power station history photovoltaic goes out force data；The collection in step (1) includes：Conversion photovoltaic History is exerted oneself P (i, t), and is exerted oneself with following formula (1) expression headroom theory and relative exerted oneself：

P (i, t)=P_DCI(i,t)·P_N(i,t) (1)

In formula (1), P_DCI(i, t) and P_N(i, t) is respectively i-th day t and is exerted oneself and phase based on the headroom theory of clearance model To exerting oneself.

By the relative P that exerts oneself_N(i, t) is decomposed into the set of power average value and undulating value, is expressed with following formula (2)：

P_N(i, t)=P_M(i)+P_F(i,t) (2)

In formula (2), P_F(i, t) is the undulating value that i-th day t is caused by different cloud layer states and state of weather, P_M(i) For the power average value of i-th day, reflection same day photovoltaic was exerted oneself degree.

(2) it is input into odd-numbered day weather pattern；Including fine day, cloudy, cloudy and change weather.

(3) obtain the likelihood function of the undulating value probability distribution under the weather pattern；Step (3) includes：Using three points Amount Undulating value P under each weather pattern of mixed Gaussian Probability Distribution Fitting_FThe likelihood function of (i, t) probability distribution, its expression formula is：

In formula (3), α_j、μ_jAnd σ_jThe weight coefficient of respectively j-th Gaussian component, average and standard deviation (j=1,2,3), z_i∈[z₁,z₂,…z_N] represent i-th observed value；

Parameter θ is estimated in order_j=[α_j,μ_j,σ_j], (j=1,2,3), observed value Z=[z₁,z₂,…z_N], obtain likelihood function：

The both sides of formula (4) are taken the logarithm acquisition following formula：

Wherein, z_i∈[z₁,z₂,…z_N] represent i-th observed value, numbers of the N for observed value.

(4) Maximum Likelihood Estimation based on EM estimates three-component mixed Gaussian probability distribution parameters, generates fitting general Rate distribution curve；As shown in Fig. 2 being the Fitted probability distribution curve and actual probability distribution under different weather type in figure The schematic diagram of curve.

In step (4), estimate that three-component mixed Gaussian probability distribution parameters specifically include following step：

4-1 calculates posterior probability；

4-2 updates weight coefficient, average and standard deviation matrix；

4-3 judges convergence.

Step (4-1) includes, calculates weight coefficient α_jPosterior probability be：

Step (4-2) includes：Using the weight coefficient α in formula (6) newer (3)_j, mean μ_jWith standard deviation sigma_j：

Step (4-3) judges that convergence includes：Iterative step (4-1) and (4-2), are cyclically updated, until meeting |l(θ_j)-l'(θ_j) | ＜ ε, ε<10⁵；Wherein, l (θ_j) for undulating value P_FThe logarithm of the likelihood function of (i, t) probability distribution, l'(θ_j) it is value after each iteration updates, ε is perunit value.

(5) confirm the effectiveness of fitting of distribution.

Step (5) includes：Verify that three-component mixed Gaussian probability distribution confirms the effectiveness of fitting using residuals squares SSE Represented with following formula (10)：

In formula (10),For i-th observed value z_iThree-component mixed Gaussian Probability Distribution Fitting probability density function Value, y_iFor actual probabilities density value, length of the n for probability density value；If SSE is less than ε (ε=0.2～0.3), it is fitted Effectively.

Finally it should be noted that：Above example only to illustrate technical scheme rather than a limitation, although The present invention is described in detail with reference to above-described embodiment, those of ordinary skill in the art should be understood：Still The specific embodiment of the present invention can be modified or equivalent, and appointing without departing from spirit and scope of the invention What modification or equivalent, which all should be covered in the middle of scope of the presently claimed invention.

Claims (10)

1. a kind of analysis photovoltaic is exerted oneself the method for undulatory property, it is characterised in that include：

(1) collection photovoltaics power station history photovoltaic goes out force data；

(2) it is input into odd-numbered day weather pattern；

(3) obtain the likelihood function of the undulating value probability distribution under the weather pattern；

(4) Maximum Likelihood Estimation based on EM estimates three-component mixed Gaussian probability distribution parameters, generates Fitted probability Distribution curve；

(5) confirm the effectiveness of fitting of distribution.

2. the method for claim 1, it is characterised in that the collection in the step (1) includes：Conversion The history of photovoltaic is exerted oneself P (i, t), and is exerted oneself with following formula (1) expression headroom theory and relative exerted oneself：

P (i, t)=P_DCI(i,t)·P_N(i,t) (1)

In formula (1), P_DCI(i, t) and P_N(i, t) is respectively i-th day t and is exerted oneself and phase based on the headroom theory of clearance model To exerting oneself.

3. method as claimed in claim 2, it is characterised in that by the relative P that exerts oneself_N(i, t) is decomposed into power averaging Value and the set of undulating value, are expressed with following formula (2)：

P_N(i, t)=P_M(i)+P_F(i,t) (2)

In formula (2), P_F(i, t) is the undulating value that i-th day t is caused by different cloud layer states and state of weather, P_M(i) For the power average value of i-th day, reflection same day photovoltaic was exerted oneself degree.

4. the method for claim 1, it is characterised in that the weather pattern of the step (2) includes fine day, many Cloud, cloudy day and change weather.

5. the method for claim 1, it is characterised in that the step (3) includes：With (3) three points of following formula Undulating value P under each weather pattern of amount mixed Gaussian Probability Distribution Fitting_FThe likelihood function of (i, t) probability distribution：

$f\left(x\right)={\mathrm{\&alpha;}}_1\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_1}{e}^{-\frac{1}{2{\mathrm{\&sigma;}}_1^2}{(z_i-{\mathrm{\&mu;}}_1)}^2}+{\mathrm{\&alpha;}}_2\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_2}{e}^{-\frac{1}{2{\mathrm{\&sigma;}}_2^2}{(z_i-{\mathrm{\&mu;}}_2)}^2}+{\mathrm{\&alpha;}}_3\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_3}{e}^{-\frac{1}{2{\mathrm{\&sigma;}}_3^2}{(z_i-{\mathrm{\&mu;}}_3)}^2}---\left(3\right)$

In formula (3), α_j、μ_jAnd σ_jThe weight coefficient of respectively j-th Gaussian component, average and standard deviation (j=1,2,3), z_i∈[z₁,z₂,…z_N] represent i-th observed value；

Parameter θ is estimated in order_j=[α_j,μ_j,σ_j], (j=1,2,3), observed value Z=[z₁,z₂,…z_N], obtain likelihood function：

$L\left(Z\right|{\mathrm{\&theta;}}_j)=\underset{i=1}{\overset{N}{\&Pi;}}f\left(x\right)\left(z_i\right)---\left(4\right)$

The both sides of formula (4) are taken the logarithm acquisition following formula (5)：

$l\left({\mathrm{\&theta;}}_j\right)=ln\&lsqb;L\left(Z\right|{\mathrm{\&theta;}}_j)\&rsqb;=\underset{i=1}{\overset{N}{\&Sigma;}}ln\underset{j=1}{\overset{3}{\&Sigma;}}{\mathrm{\&alpha;}}_j\frac{1}{\sqrt{2\mathrm{\&pi;}}{\mathrm{\&sigma;}}_j}{e}^{-\frac{1}{2{\mathrm{\&sigma;}}_j^2}{(z_i-{\mathrm{\&mu;}}_j)}^2}---\left(5\right)$

(5) in formula, z_i∈[z₁,z₂,…z_N] represent i-th observed value, numbers of the N for observed value.

6. the method for claim 1, it is characterised in that in the step (4), estimates that three-component mixing is high This probability distribution parameters generates Fitted probability distribution curve and comprises the steps：

4-1 calculates posterior probability；

4-2 updates weight coefficient, average and standard deviation matrix；

4-3 judges convergence.

7. method as claimed in claim 6, it is characterised in that the step (4-1) includes, calculates weight coefficient α_j Posterior probability be：

${\mathrm{\&beta;}}_j\left(z_i\right)=\frac{{\mathrm{\&alpha;}}_{j}N(z_i;{\mathrm{\&mu;}}_j,{\mathrm{\&sigma;}}_j^2)}{\underset{j=1}{\overset{k}{\&Sigma;}}{\mathrm{\&alpha;}}_{j}N(z_i;{\mathrm{\&mu;}}_j,{\mathrm{\&sigma;}}_j^2)}---\left(6\right)$

8. method as claimed in claim 6, it is characterised in that the step (4-2) includes：Using formula (6) more Weight coefficient α in new-type (3)_j, mean μ_jWith standard deviation sigma_j：

${\mathrm{\&alpha;}}_j^{\&prime;}=\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{\&beta;}}_j\left(z_i\right),j=1,2,3---\left(7\right)$

${\mathrm{\&mu;}}_j^{\&prime;}=\frac{\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{\&beta;}}_j\left(z_i\right)z_i}{\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{\&beta;}}_j\left(z_i\right)},j=1,2,3---\left(8\right)$

${{\mathrm{\&sigma;}}_j^2}^{\&prime;}=\frac{\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{\&beta;}}_j{(z_i-{\mathrm{\&mu;}}_j^{\&prime;})}^T(z_i-{\mathrm{\&mu;}}_j^{\&prime;})}{\underset{i=1}{\overset{N}{\&Sigma;}}{\mathrm{\&beta;}}_j\left(z_i\right)},j=1,2,3---\left(9\right)$

9. method as claimed in claim 6, it is characterised in that the step (4-3) judges that convergence includes：Repeatedly Ride instead of walk rapid (4-1) and (4-2), is cyclically updated, until meeting | l (θ_j)-l'(θ_j) | ＜ ε, ε<10⁵；Wherein, l (θ_j) For undulating value P_FThe logarithm of the likelihood function of (i, t) probability distribution, l'(θ_j) it is value after each iteration updates, ε is perunit Value.

10. the method for claim 1, it is characterised in that the step (5) includes：Using residuals squares SSE Checking three-component mixed Gaussian probability distribution confirms that the effectiveness of fitting is represented with following formula (10)：

$SSE=\underset{i=1}{\overset{n}{\&Sigma;}}{({\hat{y}}_i-y_i)}^2---\left(10\right)$

In formula (10),For i-th observed value z_iThree-component mixed Gaussian Probability Distribution Fitting probability density function values, y_iFor actual probabilities density value, length of the n for probability density value；If SSE is less than ε (ε=0.2～0.3), fitting is effective.