CN106372718A - Photovoltaic generation power output reduction rate estimation method under haze soot formation condition - Google Patents

Abstract

The invention puts forward a photovoltaic generation power output reduction rate estimation method under a haze soot formation condition. On the basis of collecting a great quantity of practical sample data of the PM (Particulate Matter) 2.5 concentration-PM10 concentration-relative humidity-wind speed-photovoltaic generation power output reduction rate, a universal gravitation neural network is used for establishing the prediction model of the photovoltaic generation power output reduction rate, and the model is used for estimating the photovoltaic generation power output reduction rate under the haze soot formation. By use of the method, the problems that the soot formation concentration of a photovoltaic cell panel is difficult in measurement and influence on photovoltaic generation is difficult in quantitative analysis are solved so as to be favorable for improving the prediction accuracy of photovoltaic generation.

Description

Photovoltaic generation power output slip method of estimation under the conditions of haze dust stratification

Technical field

The present invention relates to the photovoltaic generation power output slip method of estimation under the conditions of a kind of haze dust stratification, belong to photovoltaic Generated power forecasting technical field.

Background technology

Increasing to the demand of the energy with society, the problem of " energy security crisis " more highlights, traditional form " global warming " problem caused by utilization of energy is also increasingly serious.Solar energy power generating as one kind inexhaustible, Nexhaustible clean energy resource, has obtained the attention of various countries.Affected by Multiple factors such as illumination condition, temperature, photovoltaic generation work( Rate has undulatory property, carries out its power prediction significant for the stable operation ensureing power system.

Since 2012, China occurs in that large-scale haze weather.Meanwhile, national correlation department starts real-time release sky 6 fundamental surveillance data such as makings volume index (aqi) and pm2.5 and pm10 concentration therein, on January 2nd, 2015, Cover 367 cities.Mist and haze refer to trickle water droplet and the dust granule being suspended in air respectively, are the important of atmospheric aerosol Ingredient.The impact to photovoltaic generation for the haze has two aspects: one is the solar radiation that direct weakening is transmitted in an atmosphere, thus Reduce photovoltaic generation power, two is that fine particle under haze weather constantly settles, and forms dust stratification and covers in photovoltaic battery panel table Face, by occlusion effect, temperature effects, corrosion effect, weakens the solar radiation that photovoltaic cell absorbs, indirectly thus dropping further Its generated output low.

1) occlusion effect of photovoltaic dust stratification: because the presence of dust is so that the transmittance of face glass of photovoltaic cell subtracts Weak, thus reaching the light intensity decreasing on photovoltaic cell surface, photoelectric effect weakens, and photovoltaic generation power reduces.

2) temperature effects of photovoltaic dust stratification: dust, as the less material of heat conductivity, is attached to surface of photovoltaic cell panel resistance Keep off its heat outwards transmission so that the temperature more and more higher of photovoltaic battery panel itself, thus affecting its generating efficiency.

3) corrosion effect of photovoltaic dust stratification: from glass cover-plate on the whole, because acid or alkaline dust is to face glass Corrosion, cause face glass surface roughness increase, energy of reflection light increase, refraction light energy reduce so that inciding Intensity of illumination on photovoltaic cell weakens, and photoelectric effect weakens, and generated energy reduces.

Have some scholars both at home and abroad and carried out the correlational study work that dust stratification affects on photovoltaic cell capable of generating power power, concentrate In the dust stratification concentration of analysis photovoltaic battery panel and the relation of photovoltaic generation power output slip.Some scholars according to laterite, The simulation experiment result of limestone and ashes, establishes theoretical model, can estimate photovoltaic power output according to dust stratification concentration Slip；Some scholars pass through to simulate nature dust indoors it is proposed that dust stratification concentration exports slip with photovoltaic power Linear fit model.Because experimental situation and dust stratification component difference are larger, the studies above work lacks comparability each other, It is to be tested that its conclusion is appropriate for that the haze dust stratification based on fine particle has.In addition, in Practical Project, measuring photovoltaic electric The dust stratification concentration of pond panel need to enter Line Continuity measurement more difficult, thus have impact on by means of high-precision weighing equipment State the practicality of model.

Because the dust stratification concentration of photovoltaic battery panel is difficult to carry out real-time measurement, more efficiently characterizing method need to be found, For calculating the power output slip of photovoltaic generation.The present invention proposes the photovoltaic generation power under the conditions of a kind of haze dust stratification Output slip method of estimation, preferably to carry out the prediction of photovoltaic generation power under haze weather.

Content of the invention

A kind of photovoltaic generation power output slip method of estimation under the conditions of haze dust stratification, methods described is a large amount of in collection Pm2.5 concentration-pm10 concentration-relative humidity-wind speed-photovoltaic generation power exports the basis of the actual sample data of slip On, set up, using gravitation neural net method, the forecast model that photovoltaic generation power exports slip, and utilize this model Carry out photovoltaic generation power output slip to estimate.Comprise the following steps for the method:

A. gathered data construction sample set；

Under the conditions of a certain fixed-illumination of laboratory, the output of the clean photovoltaic battery panel of recording surface is p₀；So Afterwards, photovoltaic battery panel is placed at the outdoor spaciousness under haze weather so as to be α with the inclination angle of level ground, carries out natural product Ash, and with δ t as time interval, periodically under the conditions of laboratory identical fixed-illumination, measure its output, record simultaneously The pm2.5 concentration of corresponding moment air, pm10 concentration, relative humidity rh, wind speed v, for i ＆ lt measurement, record its output work Rate is p_i, the pm2.5 concentration of air is c_i1, pm10 concentration be c_i2, relative humidity is rh_i, wind speed be v_i；By long-term data Record, forms original sample collection { (c_i1, c_i2, rh_i, v_i, p_i), i=1,2 ..., m, m are sample total number；

B. original sample collection is processed, composing training sample set:

Based on original sample collection { (c_i1, c_i2, rh_i, v_i, p_i), calculate i ＆ lt and measure corresponding humidity modifying factor f (rh_i) and wind speed modifying factor f (v_i), and obtain pm2.5 cumulative concentration c_i1 ^*With pm10 cumulative concentration c_i2 ^*, and i ＆ lt measurement Corresponding photovoltaic generation power exports slip η_i:

$f\left({\mathrm{rh}}_i\right)=\frac{1}{1-{\mathrm{rh}}_i/100};$

$f\left(v_i\right)=3.91\frac{1}{v_i};$

$c_{i1}^{*}=\underset{j=1}{\overset{i}{\σ}}{c}_{j1}\·\mathrm{\δ}t\·f\left({\mathrm{rh}}_j\right)f\left(v_j\right);$

$c_{i2}^{*}=\underset{j=1}{\overset{i}{\σ}}{c}_{j2}\·\mathrm{\δ}t\·f\left({\mathrm{rh}}_j\right)f\left(v_j\right);$

${\mathrm{\η}}_i=\frac{p_0-p_i}{p_0}\×100\%;$

Thus obtaining training sample set

Described humidity modifying factor f (rh_i) basis be: the extinction coefficient of aerosol particle thing be subject to relative humidity shadow Sound is larger, in the case that relative humidity is higher, water solublity aerosol particle moisture expantion, and extinction coefficient can increase several times；

Described wind speed modifying factor f (v_i) basis be: higher wind speed is conducive to the diffusion of aerosol particle thing, Reduce photovoltaic battery panel dust stratification；

C. set up three-decker bp neural network prediction model, wherein, input layer nodes are 2, hidden layer Neuron node number is 5, and output layer neuron node number is 1, and hidden layer neuron transfer function uses tanh letter Number, output layer neuron transfer function uses s type function；For sampleTake bp neural network prediction First of model inputs and isSecond inputs and isIt is output as η_i；

D. it is based on training sample set, using the weights to the bp neural network prediction model set up for the gravitation searching algorithm It is optimized with threshold value, specifically comprise the following steps that

1. setting particle population size n and each particle initial position For particle dimension, initialize each particle position by the way of generating random number, and arrange Big iterationses；

2. calculate the fitness function value of each particle:

Define the fitness function f of particle_iFor mean square deviation on training sample for the bp neural network prediction model:

$f_i=\frac{1}{mp}\underset{p=1}{\overset{p}{\σ}}\underset{j=1}{\overset{m}{\σ}}{\left({\hat{x}}_{pj}-x_{pj}\right)}^2$

Wherein: m is output node number；P is the number of training sample；For network desired output；x_pjFor network Real output value；

3. update the f in colony_bestWith f_worst(f_best=min f_j, f_worst=max f_j), it is calculated as follows each particle Quality m_i:

$m_i=\frac{f_i-f_{worst}}{f_{best}-f_{worst}};$

4. the gravitation calculating each particle according to the following formula is made a concerted effort f_iWith acceleration a_i:

$f_i^d\left(t\right)=\underset{j\&notequal;i}{\σ}{\mathrm{rand}}_j{f}_{ij}^d\left(t\right);$

$a_i^d\left(t\right)=\frac{f_i^d\left(t\right)}{m_i\left(t\right)};$

$f_{ij}^d\left(t\right)=g\left(t\right)\frac{m_{pi}\left(t\right)m_{aj}\left(t\right)}{\left|\right|p_i\left(t\right),p_j\left(t\right)|{|}^2}\left(p_j^d\right(t)-p_i^d(t\left)\right),$

In formula: t is iterationses, f_i ^dT () is making a concerted effort f from other particle gravitation of being subject to of i-th particle_iD dimension Component,For the acceleration of the d dimension of i-th particle, rand_jFor the random number between [0,1], g (t) is the gravitation time Constant m_pi(t) and m_ajT () is respectively the passive inertia mass of i-th particle and the active inertia mass of j-th particle, WithIt is respectively the position of the d dimension of i-th particle and j-th particle；

5. update speed v of each particle according to the following formula_iWith position p_i:

$v_i^d(t+1)={\mathrm{rand}}_i\×v_i^d\left(t\right)+a_i^d\left(t\right);$

$p_i^d(t+1)=p_i^d\left(t\right)+v_i^d(t+1);$

In formula:D for i-th particle ties up speed；

6. judging whether to reach maximum iteration time, if reaching maximum iteration time, stopping iteration, now f_bestCorresponding Particle position be optimize after the weights of bp neural network prediction model and threshold value；Otherwise, return to step is 2.；

E. carry out, using gravitation neural network prediction model, the estimation that photovoltaic generation power exports slip, will certain Pm2.5 cumulative concentration c is obtained under one environment_i1 ^*With pm10 cumulative concentration c_i2 ^*Defeated as gravitation neural network prediction model Enter, the output of forecast model is the estimated value that photovoltaic generation power under current environment exports slip

The beneficial effects of the present invention is:

(1) using the pm2.5 cumulative concentration on the ground that can obtain in real time, pm10 cumulative concentration, photovoltaic generation power is exported Slip is estimated, and considers the impact of relative humidity and wind speed；

(2) establish, using neural net method, the forecast model that photovoltaic generation power exports slip, and using universal Gravitation algorithm is optimized to the parameter of neural network prediction model such that it is able to effectively improve model accuracy.

Below in conjunction with the accompanying drawings, preferred embodiment is elaborated.It should be emphasized that the description below is merely exemplary , rather than in order to limit the scope of the present invention and its application.

Brief description

Fig. 1 is photovoltaic generation power output slip forecast model structure chart proposed by the invention；

Three layers of bp neural network structure figure that Fig. 2 adopts by the carried forecast model of the present invention；

The flow chart that Fig. 3 is optimized to bp neural network prediction model using gravitation searching algorithm；

Fig. 4 is estimated result on test samples for the forecast model in embodiment.

Specific embodiment

The present invention is gathering a large amount of pm2.5 concentration-pm10 concentration-relative humidity-wind speed-photovoltaic generation power output minimizings On the basis of the actual sample data of rate, set up photovoltaic generation power using gravitation neural net method and export slip Forecast model, its structure is as shown in figure 1, and carry out photovoltaic generation power output slip estimation using this model.

The present invention carried photovoltaic generation power output slip method of estimation comprises the steps:

A. gathered data construction sample set；

Under the conditions of a certain fixed-illumination of laboratory, the output of the clean photovoltaic battery panel of recording surface is p₀；So Afterwards, photovoltaic battery panel is placed at the outdoor spaciousness under haze weather so as to be α with the inclination angle of level ground, carries out natural product Ash, and with δ t as time interval, periodically under the conditions of laboratory identical fixed-illumination, measure its output, record simultaneously The pm2.5 concentration of corresponding moment air, pm10 concentration, relative humidity rh, wind speed v, for i ＆ lt measurement, record its output work Rate is p_i, the pm2.5 concentration of air is c_i1, pm10 concentration be c_i2, relative humidity is rh_i, wind speed be v_i；By long-term data Record, forms original sample collection { (c_i1, c_i2, rh_i, v_i, p_i), i=1,2 ..., m, m are sample total number；

In preferred embodiment, original sample collection is collected in the Baoding area between in December, 2015 in June, 2016.? Laboratory utilizes 1000w high pressure xenon lamp simulated solar illumination environment, and measurement surface is clean, the photovoltaic battery panel of rated power 30w Output, be recorded as p₀=15.6w；Then, photovoltaic battery panel is placed at the outdoor spaciousness under haze weather so as to The inclination angle of level ground is α=30 °, carries out nature dust stratification and evades rainy weather, and with δ t=24h as time interval, regularly Measure its output under the conditions of laboratory identical fixed-illumination, form original sample collection { (c_i1, c_i2, p_i), i=1, 2 ..., m, m=150；

B. original sample collection is processed, composing training sample set:

Based on original sample collection { (c_i1, c_i2, rh_i, v_i, p_i), calculate i ＆ lt and measure corresponding humidity modifying factor f (rh_i) and wind speed modifying factor f (v_i), and obtain pm2.5 cumulative concentration c_i1 ^*With pm10 cumulative concentration c_i2 ^*, and i ＆ lt measurement Corresponding photovoltaic generation power exports slip η_i:

$f\left({\mathrm{rh}}_i\right)=\frac{1}{1-{\mathrm{rh}}_i/100};$

$f\left(v_i\right)=3.91\frac{1}{v_i};$

$c_{i1}^{*}=\underset{j=1}{\overset{i}{\σ}}{c}_{j1}\·\mathrm{\δ}t\·f\left({\mathrm{rh}}_j\right)f\left(v_j\right);$

$c_{i2}^{*}=\underset{j=1}{\overset{i}{\σ}}{c}_{j2}\·\mathrm{\δ}t\·f\left({\mathrm{rh}}_j\right)f\left(v_j\right);$

${\mathrm{\η}}_i=\frac{p_0-p_i}{p_0}\×100\%;$

Thus obtaining training sample set

C. set up three-decker bp neural network prediction model, as shown in Fig. 2 wherein, input layer nodes are 2 Individual, hidden layer neuron nodes are 5, and output layer neuron node number is 1, and hidden layer neuron transfer function is using double Bent tan, output layer neuron transfer function uses s type function；For sampleTake bp nerve net First of network forecast model inputs and isSecond inputs and isIt is output as η_i；

D. it is based on training sample set, using the weights to the bp neural network prediction model set up for the gravitation searching algorithm It is optimized with threshold value, idiographic flow is as shown in figure 3, comprise the steps:

1. setting particle population size n and each particle initial position For particle dimension, initialize each particle position by the way of generating random number, and arrange Big iterationses；In embodiment, take n=40, the bp neural network input layer set up has l=2 neuron, and hidden layer has h =5 neurons, output layer has m=1 neuron, if w_lh={ w_ji| j=1 ..., l；I=1 ..., h } neural for input layer Connection weight set between unit and hidden layer neuron, w_hm={ w_kj| k=1 ..., m；J=1 ..., h } neural for hidden layer Connection weight set between unit and output layer neuron, b_hAnd b_mRepresent hidden layer neuron and output layer neuron respectively Threshold value, then each particle position is p_i=[w_lh, b_h, w_hm, b_m], therefore d=21, setting maximum iteration time is 1000 times；

2. calculate the fitness function value of each particle:

Define the fitness function f of particle_iFor mean square deviation on training sample for the bp neural network prediction model:

$f_i=\frac{1}{mp}\underset{p=1}{\overset{p}{\σ}}\underset{j=1}{\overset{m}{\σ}}{\left({\hat{x}}_{pj}-x_{pj}\right)}^2$

Wherein: m is output node number；P is the number of training sample；For network desired output；x_pjFor network Real output value；

3. update the f in colony_bestWith f_worst(f_best=min f_j, f_worst=max f_j), it is calculated as follows each particle Quality m_i:

$m_i=\frac{f_i-f_{worst}}{f_{best}-f_{worst}};$

4. the gravitation calculating each particle according to the following formula is made a concerted effort f_iWith acceleration a_i:

$f_i^d\left(t\right)=\underset{j\&notequal;i}{\σ}{\mathrm{rand}}_j{f}_{ij}^d\left(t\right);$

$a_i^d\left(t\right)=\frac{f_i^d\left(t\right)}{m_i\left(t\right)};$

$f_{ij}^d\left(t\right)=g\left(t\right)\frac{m_{pi}\left(t\right)m_{aj}\left(t\right)}{\left|\right|p_i\left(t\right),p_j\left(t\right)|{|}^2}\left(p_j^d\right(t)-p_i^d(t\left)\right),$

In formula: t is iterationses, f_i ^dT () is making a concerted effort f from other particle gravitation of being subject to of i-th particle_iD dimension Component,For the acceleration of the d dimension of i-th particle, rand_jFor the random number between [0,1], g (t) is the gravitation time Constant m_pi(t) and m_ajT () is respectively the passive inertia mass of i-th particle and the active inertia mass of j-th particle, WithIt is respectively the position of the d dimension of i-th particle and j-th particle；

5. update speed v of each particle according to the following formula_iWith position p_i:

$v_i^d(t+1)={\mathrm{rand}}_i\×v_i^d\left(t\right)+a_i^d\left(t\right);$

$p_i^d(t+1)=p_i^d\left(t\right)+v_i^d(t+1);$

In formula:D for i-th particle ties up speed；

6. judging whether to reach maximum iteration time, if reaching maximum iteration time, stopping iteration, now f_bestCorresponding Particle position be optimize after the weights of bp neural network prediction model and threshold value；Otherwise, return to step is 2.；

In embodiment, the weights of bp neural network prediction model after optimization are as shown in the table with threshold value.

Table 1

Table 2

bh(h=1,2,3,4,5)

-4.8752

-2.8568

1.9983

0.7211

3.1285

Table 3

whm={ wkj| k=1,2,3,4,5；J=1 }

-2.1296

0.9454

3.1124

-0.9263

-0.5151

E. carry out, using gravitation neural network prediction model, the estimation that photovoltaic generation power exports slip, will certain Pm2.5 cumulative concentration c is obtained under one environment_i1 ^*With pm10 cumulative concentration c_i2 ^*Defeated as gravitation neural network prediction model Enter, the output of forecast model is the estimated value that photovoltaic generation power under current environment exports slip

In embodiment, k moment pm2.5 concentration c under a certain environment_i1=68ug/m³H, c_i2=252ug/m³H, can count Calculate and end to k moment corresponding pm2.5 cumulative concentration c_k1 ^*=10327ug/m³H, pm10 cumulative concentration c_k2 ^*= 33028ug/m³H, as the input of gravitation neural network prediction model, the output of forecast model is works as front ring Under border, photovoltaic generation power exports the estimated value of slip

In order to verify the precision of set up photovoltaic battery panel power output slip forecast model, in addition choose 21 groups of samples This, as verification sample set, is verified to estimation effect, result as shown in Figure 4 it is known that, verification sample average absolute by mistake Difference is 0.158%, illustrates that model has higher estimated accuracy.

The above, the only present invention preferably specific embodiment, but protection scope of the present invention is not limited thereto, Any those familiar with the art the invention discloses technical scope in, the change or replacement that can guess, all Should be included within the scope of the present invention.

Claims (1)

1. the photovoltaic generation power output slip method of estimation under the conditions of a kind of haze dust stratification, methods described is a large amount of in collection Pm2.5 concentration-pm10 concentration-relative humidity-wind speed-photovoltaic generation power exports the basis of the actual sample data of slip On, set up, using gravitation neural net method, the forecast model that photovoltaic generation power exports slip, and utilize this model Carry out photovoltaic generation power output slip to estimate, it is characterized by the method comprises the following steps:

A. gathered data construction sample set；

Under the conditions of a certain fixed-illumination of laboratory, the output of the clean photovoltaic battery panel of recording surface is p₀；Then, will So as to be α with the inclination angle of level ground at the outdoor spaciousness that photovoltaic battery panel is placed under haze weather, carry out nature dust stratification, and With δ t as time interval, periodically under the conditions of laboratory identical fixed-illumination, measure its output, when recording corresponding simultaneously Carve the pm2.5 concentration of air, pm10 concentration, relative humidity rh, wind speed v, for i ＆ lt measurement, recording its output is p_i, The pm2.5 concentration of air is c_i1, pm10 concentration be c_i2, relative humidity is rh_i, wind speed be v_i；By long-term data record, shape Become original sample collection { (c_i1, c_i2, rh_i, v_i, p_i), i=1,2 ..., m, m are sample total number；

B. original sample collection is processed, composing training sample set:

Based on original sample collection { (c_i1, c_i2, rh_i, v_i, p_i), calculate i ＆ lt and measure corresponding humidity modifying factor f (rh_i) with Wind speed modifying factor f (v_i), and obtain pm2.5 cumulative concentration c_i1 ^*With pm10 cumulative concentration c_i2 ^*, and i ＆ lt measurement is corresponding Photovoltaic generation power exports slip η_i:

$f\left({\mathrm{rh}}_i\right)=\frac{1}{1-{\mathrm{rh}}_i/100};$

$f\left(v_i\right)=3.91\frac{1}{v_i};$

$c_{i1}^{*}=\underset{j=1}{\overset{i}{\mathrm{\σ}}}{c}_{j1}\·\mathrm{\δ}t\·f\left({\mathrm{rh}}_j\right)f\left(v_j\right);$

$c_{i2}^{*}=\underset{j=1}{\overset{i}{\mathrm{\σ}}}{c}_{j2}\·\mathrm{\δ}t\·f\left({\mathrm{rh}}_j\right)f\left(v_j\right);$

${\mathrm{\η}}_i=\frac{p_0-p_i}{p_0}\×\mathrm{100\%;}$

Thus obtaining training sample set

C. set up three-decker bp neural network prediction model, wherein, input layer nodes are 2, hidden layer nerve First nodes are 5, and output layer neuron node number is 1, and hidden layer neuron transfer function uses hyperbolic tangent function, defeated Go out layer neuron transfer function and use s type function；For sampleTake bp neural network prediction model First inputs and isSecond inputs and isIt is output as η_i；

D. it is based on training sample set, using weights and the threshold of the bp neural network prediction model to foundation for the gravitation searching algorithm Value is optimized, and specifically comprises the following steps that

1. setting particle population size n and each particle initial positionI=1,2, 3 ..., n, d are particle dimension, initialize each particle position, and arrange maximum iteration time by the way of generating random number；

2. calculate the fitness function value of each particle:

Define the fitness function f of particle_iFor mean square deviation on training sample for the bp neural network prediction model:

$f_i=\frac{1}{mp}\underset{p=1}{\overset{p}{\mathrm{\σ}}}\underset{j=1}{\overset{m}{\mathrm{\σ}}}{({\hat{x}}_{pj}-x_{pj})}^2$

Wherein: m is output node number；P is the number of training sample；For network desired output；x_pjDefeated for network reality Go out value；

3. update the f in colony_bestWith f_worst(f_best=minf_j, f_worst=maxf_j), it is calculated as follows the quality of each particle m_i:

$m_i=\frac{f_i-f_{worst}}{f_{best}-f_{worst}};$

4. the gravitation calculating each particle according to the following formula is made a concerted effort f_iWith acceleration a_i:

$f_i^d\left(t\right)=\underset{j\&notequal;i}{\mathrm{\σ}}{\mathrm{rand}}_j{f}_{ij}^d\left(t\right);$

$a_i^d\left(t\right)=\frac{f_i^d\left(t\right)}{m_i\left(t\right)};$

$f_{ij}^d\left(t\right)=g\left(t\right)\frac{m_{pi}\left(t\right)m_{aj}\left(t\right)}{\left|\right|p_i\left(t\right),p_j\left(t\right)|{|}^2}\left(p_j^d\right(t)-p_i^d(t\left)\right),$

In formula: t is iterationses, f_i ^dT () is making a concerted effort f from other particle gravitation of being subject to of i-th particle_iD dimension component,For the acceleration of the d dimension of i-th particle, rand_jFor the random number between [0,1], g (t) is gravitation time constant m_pi (t) and m_ajT () is respectively the passive inertia mass of i-th particle and the active inertia mass of j-th particle,WithIt is respectively the position of the d dimension of i-th particle and j-th particle；

5. update speed v of each particle according to the following formula_iWith position p_i:

$v_i^d(t+1)={\mathrm{rand}}_i\×v_i^d\left(t\right)+a_i^d\left(t\right);$

$p_i^d(t+1)=p_i^d\left(t\right)+v_i^d(t+1);$

In formula:D for i-th particle ties up speed；

6. judging whether to reach maximum iteration time, if reaching maximum iteration time, stopping iteration, now f_bestCorresponding grain Sub- position is the weights of bp neural network prediction model and threshold value after optimizing；Otherwise, return to step is 2.；

E. carry out, using gravitation neural network prediction model, the estimation that photovoltaic generation power exports slip, will a certain ring Pm2.5 cumulative concentration c is obtained under border_i1 ^*With pm10 cumulative concentration c_i2 ^*As the input of gravitation neural network prediction model, The output of forecast model is the estimated value that photovoltaic generation power under current environment exports slip