CN104915899A - Photovoltaic power station operation state classifying method based on characteristic cluster analysis - Google Patents

Abstract

The invention relates to a photovoltaic power station operation state classifying method based on characteristic cluster analysis and belongs to the technical field of photovoltaic power generation. According to the invention, characteristic quantities, including matrix conversion efficiency, inversion conversion efficiency, PV matrix equivalence power generation time and PV system equivalence power generation time, for indicating operation states of the photovoltaic power station; and adopting a cluster analysis algorithm for cluster analysis of obtained sample data and implementing the classification of operation states of the photovoltaic power station. According to the invention, the K-means algorithm is adopted for cluster analysis of the characteristic quantities of the photovoltaic power station and the cluster results indicate the operation states of the photovoltaic power station. The operation range of the photovoltaic power station can be reduced further and the operation time can be reduced according to the obtained operation states, so that power generation earnings of the photovoltaic power station can be guaranteed further. Scientific guide is provided for operation state evaluation of the photovoltaic power station and the development of operation maintenance of the photovoltaic power station is promoted.

Description

A kind of photovoltaic plant running status division methods of feature based cluster analysis

Technical field

The present invention relates to a kind of photovoltaic plant running status division methods of feature based cluster analysis, belong to technical field of photovoltaic power generation.

Background technology

Photovoltaic generation is as the one of clean energy resource, worldwide all receive and pay attention to widely, Chinese large-sized ground light overhead utility, distributed roof photovoltaic power station gets more and more, but the running status of photovoltaic plant, the suitable O&M time, photovoltaic plant raising generated energy has been become to need the major issue of solution badly.

Different time, environment, equipment state all can have an impact to the operation conditions of photovoltaic plant, and therefore photovoltaic plant presents different running statuses.Application number is the patent of 201410618987.3, a kind of appraisal procedure of photovoltaic plant operation characteristic is proposed, wherein relate to photovoltaic power station power generation amount, photovoltaic plant energy ezpenditure is analyzed, photovoltaic power station system efficiency analysis, important indicator such as photovoltaic plant equivalence generating dutation and photovoltaic plant healthy economy coefficient working time etc., and the method quantitative evaluation result utilizing anti-entropy to weigh, can be photovoltaic plant operator and technical support is provided, the main innovate point of the method is can quantitative evaluation result, but only can provide superior according to quantized result, well, generally, poor this kind of macroscopical judged result, and the running status of photovoltaic plant cannot be reflected.

Summary of the invention

The object of this invention is to provide a kind of photovoltaic plant running status division methods of feature based cluster analysis, to realize the division to photovoltaic plant running status.

The present invention provides a kind of photovoltaic plant running status division methods of feature based cluster analysis for achieving the above object, and this division methods comprises the following steps:

1) extract characterize photovoltaic plant running status characteristic quantity as sample data, selected characteristic quantity comprise square formation conversion efficiency, inversion conversion efficiency, PV square formation day generating dutation of equal value and PV system day generating dutation of equal value;

2) cluster algorithm is adopted to step 1) sample data that obtains carries out cluster analysis, realizes the division to photovoltaic plant running status.

Described step 2) in adopt feature clustering analytic process as follows:

A. according to the number k of photovoltaic plant operation characteristic determination photovoltaic plant running status, the cluster number using this number as sample data;

B. from step 1) sample data Stochastic choice k primary data object-point as initial cluster center;

C. to calculate in sample data all the other each data objects to the distance of k initial cluster center, and each data object is incorporated into the cluster at that initial cluster center place nearest apart from it;

D. the center of each initial clustering is calculated, using this position as cluster centre again cluster, no longer change until obtain cluster centre, and average error criterion function presents convergence state, obtain the classification results of cluster result as photovoltaic plant running status using this cluster centre.

Distance in described step C calculates and adopts Euclidean distance.

Photovoltaic plant running status is divided into 4 kinds of operational modes by the cluster result that described step D obtains, i.e. healthy operational mode, efficiency of solar array decline operational mode, inversion efficiency decline operational mode, low irradiance operational mode.

Described square formation av eff is the average energy conversion efficiency of PV square formation, and namely PV square formation outputs to the energy of PV system and the ratio of the energy incided in PV square formation,

η _Amean＝E _A/(A×H _T)

E _A＝Σ _day(P _in×τ _r)

η _ameanfor required square formation av eff; E _afor the output energy of PV square formation in the τ period, unit is kW ˙ h; A is the useful area of PV square formation, and unit is m ²; H _tfor τ period PV square formation dip plane radiant quantity, unit is kW ˙ h/m ²; Symbol Σ _dayrepresent summation per diem; P _in× τ _rrepresent the direct current measurement that in logging interval, in PV square formation, battery component exports, P _infor the PV power input in the inverter parameters that monitors.

Described inverter efficiency eta _iNVrepresenting that direct current measurement is converted to the conversion efficiency of ac electric by inverter, is inverter outgoing side power and the ratio of input side power,

η _iNV=P _export/ P _input

P _exportoutput AC power based on inverter calculates, and equals the active power in the inverter parameters monitored, and unit is kW; P _inputdC input power based on inverter calculates, and equals the PV power input in the inverter parameters monitored, and unit is kW.

Generating dutation Y of equal value of described PV square formation day _arepresent that PV square formation every kW square formation every day monitored is supplied to the energy of PV system, be equivalent to PV square formation with output rating P ₀run the hourage needing work,

Y _A＝E _A/P ₀

Wherein E _afor PV square formation output energy, unit is kW ˙ h.P ₀representing PV system peak watt power, is each battery component peak watt power sum, and the general power of PV system when namely each battery component presses rated power operation, unit is kW.

Described PV system day represents that PV system every kW square formation every day monitored is supplied to the energy of transformer station's input side when equivalence generates electricity, and is equivalent to PV system with output rating P ₀need the hourage of work,

Y _F＝E _AC/P ₀

Wherein E _aCfor PV system exports gross energy, unit is kW ˙ h; P ₀representing PV system peak watt power, is each battery component peak watt power sum, and the general power of PV system when namely each battery component presses rated power operation, unit is kW.

The invention has the beneficial effects as follows: the present invention first extract characterize photovoltaic plant running status characteristic quantity as sample data, comprise square formation conversion efficiency, inversion conversion efficiency, PV square formation day generating dutation of equal value and PV system day generating dutation of equal value; Then adopt cluster algorithm to carry out cluster analysis to the sample data obtained, realize the division to photovoltaic plant running status.The present invention adopts K-means algorithm to carry out cluster analysis to the characteristic quantity of photovoltaic plant, cluster result characterizes photovoltaic plant running status, photovoltaic plant O&M scope can be reduced further according to obtaining running status, shortening the O&M time, ensure photovoltaic power station power generation income further.For the assessment of photovoltaic plant running status provides scientific guidance, facilitate the development of photovoltaic plant operation maintenance.

Accompanying drawing explanation

Fig. 1 be the present invention adopt clustering algorithm photovoltaic plant operation state mode divide process flow diagram.

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described further.The present invention

Photovoltaic plant Feature change is on the impact of power station running status, the characteristic quantity characterizing photovoltaic plant running status primarily of square formation conversion efficiency, inversion conversion efficiency, PV square formation day generating dutation of equal value, PV system day generating dutation of equal value form, by extracting the characteristic quantity characterizing photovoltaic plant running status, the photovoltaic plant running status change that Feature change causes is analyzed, and carry out the cluster analysis of photovoltaic plant sample data based on this, the specific implementation process of the method is as follows:

1. obtain and characterize photovoltaic plant running status characteristic quantity

1) square formation av eff η _ameanrepresent the average energy conversion efficiency of PV square formation, namely PV square formation outputs to the energy of PV system and the ratio of the energy incided in PV square formation.This efficiency represents the ability of PV square formation switching energy, and numerical value is high, and represent that the ability of PV square formation switching energy is strong, under other equipment non-failure conditions, the energy being supplied to whole system by PV square formation is many.The average conversion efficiency of battle array mainly reflects that photovoltaic array loses through line, Low emissivity loses, eclipsing loss, cell panel surface is stained, angle is lost, MPPT follows the tracks of loss, radiant quantity measurement mistake, the energy conversion efficiency of inverter is transported to after the loss such as frequency spectrum loss, main reflection PV square formation running status, very strong correlativity is had with the day of PV square formation generating dutation of equal value, can be and judge that photovoltaic array line loses, Low emissivity loses, eclipsing loss, cell panel surface is stained, angle is lost, MPPT follows the tracks of loss, the square formation fallback that the reasons such as radiant quantity measurement mistake cause provides data reference.

In record slot τ, square formation av eff is defined as:

η _Amean＝E _A/(A×H _T)

E _A＝Σ _day(P _in×τ _r)

E _afor the output energy of PV square formation in the τ period, unit is kW ˙ h; Symbol Σ _dayrepresent summation per diem; P _in× τ _rrepresent the direct current measurement that in logging interval, in PV square formation, battery component exports, ignore the line loss of battery component to inverter, P _infor " PV power input " in the inverter parameters that monitors; A is the useful area of PV square formation, and unit is m ²; H _tfor τ period PV square formation dip plane radiant quantity, unit is kW ˙ h/m ².

2) inverter efficiency eta _iNVrepresent that direct current measurement is converted to the conversion efficiency of ac electric by inverter, the ability of reflection inverter switching energy, numerical value is inverter outgoing side power and the ratio of input side power.Numerical value is high, and represent that the ability of inverter switching energy is strong, the energy being supplied to box change input side by PV system is many.Inversion efficiency is the running status evaluating inverter, the important indicator whether assessment invertor operation environment is good, can be the running status judging inverter, and whether assessment invertor operation environment well provides important reference.And the output of PV square formation energy and inverter MPPT tracking mode have very strong corresponding relation, therefore the change of this index relatedly can affect the change of efficiency of solar array and generating dutation of equal value, has very strong correlativity with other parameters.

η _iNV=P _export/ P _input

P _exportoutput AC power based on inverter calculates, and equals " active power " in the inverter parameters monitored, and unit is kW; P _inputdC input power based on inverter calculates, and equal " PV power input " in the inverter parameters monitored, unit is kW.

3) PV square formation day generating dutation Y of equal value _arepresent that PV square formation every kW square formation every day monitored is supplied to the energy of PV system, be equivalent to PV square formation with output rating P ₀run the hourage needing work.PV square formation day, generating dutation of equal value was not only relevant with efficiency of solar array, inversion efficiency, more relevant with weather condition, the change of environment directly can cause the change of the generating dutation of equal value of PV square formation, but affecting relatively little on square formation conversion efficiency and inverter conversion efficiency, is participate in judging the important auxiliary foundation whether square formation running status is good.

Y _A＝E _A/P ₀

E _afor PV square formation output energy, unit is kW ˙ h; P ₀represent PV system peak watt power, be each battery component peak watt power sum, when namely each battery component presses rated power operation, the general power of PV system, unit is kW.

4) represent that PV system every kW square formation every day monitored is supplied to the energy of box-type substation input side during the equivalence generating of PV system day, be equivalent to PV system with output rating P ₀need the hourage of work.PV system equivalence generating dutation is subject to the impact of efficiency of solar array, inversion efficiency and environmental factor equally, PV system equivalence generating dutation is directly decided by PV square formation generating dutation of equal value, both have positive correlation, PV system equivalence generating dutation is the key index characterizing photovoltaic plant overall operation state, is the important auxiliary foundation participating in judging whole operation of electric power system state.

Y _F＝E _AC/P ₀

E _aCfor PV system exports gross energy, unit is kW ˙ h; P ₀represent PV system peak watt power, be each battery component peak watt power sum, when namely each battery component presses rated power operation, the general power of PV system, unit is kW.

2. adopt clustering algorithm to carry out cluster analysis, photovoltaic plant running status to the characteristic quantity obtained

Clustering algorithm adopts K-means algorithm, first specify need to divide bunch number k value; Then select k primary data object-point as initial cluster centre randomly; 3rd, calculate the distance (here generally adopt distance as similarity measurement) of remaining each data object to this k initial cluster center, data object is incorporated in bunch class at that center nearest apart from it; Finally, adjust new class and recalculate the center of the class that makes new advances, if there is not any change in the cluster centre calculated for twice, so just can illustrate that the adjustment of data object is own through terminating, that is the criterion function that cluster adopts is convergence, represent that algorithm terminates, namely final cluster result.Be a generator unit with 1MW below, the photovoltaic plant of installed capacity of power station N MW is that example is described, and the composition of sample of this photovoltaic plant is:

$\stackrel{\·}{u}=\left\{{\stackrel{\·}{u}}_1,{\stackrel{\·}{u}}_2,{\stackrel{\·}{u}}_3\mathrm{...}{\stackrel{\·}{u}}_n\right\}=\left|\begin{array}{cccc}{u}_{11}& u_{12}& u_{13}& u_{14}\\ u_{21}& u_{22}& u_{23}& u_{24}\\ \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}.\\ .\\ .\end{array}\\ u_{n1}& u_{n2}& u_{n3}& u_{n4}\end{array}\right|$

Each sample data contains each characteristic quantity

${\stackrel{\·}{u}}_i=\[u_{i1},u_{i2},u_{i3},u_{i4}\]$

Wherein, u _i1, u _i2, u _i3, u _i4represent square formation conversion efficiency respectively, inversion efficiency, PV square formation day generating dutation of equal value, PV system day generating dutation of equal value 4 proper vectors.

(1) according to photovoltaic plant operation characteristic, sample object being gathered is 4 classes, i.e. k ₁, k ₂, k ₃, k ₄; k ₁represent photovoltaic plant normal operational condition; k ₂representative affects power station running status by weather conditions; k ₃represent because assembly dust stratification is serious, assembly damage etc. reduces by efficiency of solar array the power station running status situation of change brought; k ₄represent because inversion efficiency reduces the power station running status situation of change brought;

(2) random selecting k ₁, k ₂, k ₃, k ₄4 class initial cluster centers the distance of each sample data to 4 initial cluster centers is asked for according to Euclidean distance;

$d(u_i,u_a)=\sqrt{\underset{k=1}{\overset{4}{\Σ}}{(u_{ik}-u_{ak})}^2},i=1,\mathrm{2...}n;$ $d(u_i,u_b)=\sqrt{\underset{k=1}{\overset{4}{\Σ}}{(u_{ik}-u_{bk})}^2},i=1,\mathrm{2...}n;$

$d(u_i,u_c)=\sqrt{\underset{k=1}{\overset{4}{\Σ}}{(u_{ik}-u_{ck})}^2},i=1,\mathrm{2...}n;$

$d(u_i,u_d)=\sqrt{\underset{k=1}{\overset{4}{\Σ}}{(u_{ik}-u_{dk})}^2},i=1,\mathrm{2...}n;$

According to minimal distance principle, if namely there is following relation:

$\begin{array}{ccc}d(u_i,u_a)=\min \[d_{ij}^{\left(k\right)}\]& i=1,2...n;& j=a,b,c,d\end{array}$

Then belong to for the k of core ₁classification, by that analogy, by each sample data graduation to above-mentioned 4 classes bunch, is respectively

(3) average error criterion function under the calculating initial classes heart

$E={\underset{i=1}{\overset{k}{\Σ}}\underset{p\∈X_i}{\Σ}||p-{u_i}^{\left(0\right)}||}^2$

Wherein k=4, represents the number of institute's cluster dividing; u _i ⁽⁰⁾represent the center of 4 bunches

(4) new cluster centre is calculated

$u_a^{\left(1\right)}=\frac{1}{N_1}\underset{u_i\∈w_1^{\left(0\right)}}{\Σ}{u}_i$ N ₁expression belongs to element number

$u_b^{\left(1\right)}=\frac{1}{N_2}\underset{u_i\∈w_2^{\left(0\right)}}{\Σ}{u}_i$ N ₂expression belongs to element number

$u_c^{\left(1\right)}=\frac{1}{N_3}\underset{u_i\∈w_3^{\left(0\right)}}{\Σ}{u}_i$ N ₃expression belongs to element number

$u_d^{\left(1\right)}=\frac{1}{N_4}\underset{u_i\∈w_4^{\left(0\right)}}{\Σ}{u}_i$ N ₄expression belongs to element number

Wherein N ₁+ N ₂+ N ₃+ N ₄=n n represents sample data total amount

(5) above-mentioned steps 2,3,4 is repeated, until meet following relation

$u_a^{\left(s\right)}=u_a^{(s+1)},u_b^{\left(s\right)}=u_b^{(s+1)},u_c^{\left(s\right)}=u_c^{(s+1)},u_d^{\left(s\right)}=u_d^{(s+1)},$ And there is E ^(s+1)≤ E ^(s), illustrate that, after s cluster iteration, average error criterion function presents convergence state, and a bunch center no longer changes, stop iteration, so far cluster result completes.

By cluster result, photovoltaic plant running status can be divided into 4 kinds of operational modes, be respectively healthy operational mode, low irradiance operational mode, square formation inefficient operating mode, inversion inefficient operating mode, as shown in table 1.

Table 1

According to above-mentioned cluster result, the O&M solution formulation that can be photovoltaic plant provides reference frame, thus promotes the long-term efficient stable operation of photovoltaic plant.

Claims (8)

1. a photovoltaic plant running status division methods for feature based cluster analysis, it is characterized in that, this division methods comprises the following steps:

1) extract characterize photovoltaic plant running status characteristic quantity as sample data, selected characteristic quantity comprise square formation conversion efficiency, inversion conversion efficiency, PV square formation day generating dutation of equal value and PV system day generating dutation of equal value;

2) cluster algorithm is adopted to step 1) sample data that obtains carries out cluster analysis, realizes the division to photovoltaic plant running status.

2. the photovoltaic plant running status division methods of feature based cluster analysis according to claim 1, is characterized in that, described step 2) in adopt feature clustering analytic process as follows:

A. according to the number k of photovoltaic plant operation characteristic determination photovoltaic plant running status, the cluster number using this number as sample data;

B. from step 1) sample data Stochastic choice k primary data object-point as initial cluster center;

C. to calculate in sample data all the other each data objects to the distance of k initial cluster center, and each data object is incorporated into the cluster at that initial cluster center place nearest apart from it;

D. the center of each initial clustering is calculated, using this position as cluster centre again cluster, no longer change until obtain cluster centre, and average error criterion function presents convergence state, obtain the classification results of cluster result as photovoltaic plant running status using this cluster centre.

3. the photovoltaic plant running status division methods of feature based cluster analysis according to claim 2, is characterized in that, the distance in described step C calculates and adopts Euclidean distance.

4. the photovoltaic plant running status division methods of feature based cluster analysis according to claim 2, it is characterized in that, photovoltaic plant running status is divided into 4 kinds of operational modes by the cluster result that described step D obtains, i.e. healthy operational mode, efficiency of solar array decline operational mode, inversion efficiency decline operational mode, low irradiance operational mode.

5. the photovoltaic plant running status division methods of feature based cluster analysis according to claim 2, it is characterized in that, described square formation av eff is the average energy conversion efficiency of PV square formation, and namely PV square formation outputs to the energy of PV system and the ratio of the energy incided in PV square formation

η _Amean＝E _A/(A×H _T)

E _A＝Σ _day(P _in×τ _r)

η _ameanfor required square formation av eff; E _afor the output energy of PV square formation in the τ period, unit is kW ˙ h; A is the useful area of PV square formation, and unit is m ²; H _tfor τ period PV square formation dip plane radiant quantity, unit is kW ˙ h/m ²; Symbol Σ _dayrepresent summation per diem; P _in× τ _rrepresent the direct current measurement that in logging interval, in PV square formation, battery component exports, P _infor the PV power input in the inverter parameters that monitors.

6. the photovoltaic plant running status division methods of feature based cluster analysis according to claim 2, is characterized in that, described inverter efficiency eta _iNVrepresenting that direct current measurement is converted to the conversion efficiency of ac electric by inverter, is inverter outgoing side power and the ratio of input side power,

η _iNV=P _export/ P _input

P _exportoutput AC power based on inverter calculates, and equals the active power in the inverter parameters monitored, and unit is kW; P _inputdC input power based on inverter calculates, and equals the PV power input in the inverter parameters monitored, and unit is kW.

7. the photovoltaic plant running status division methods of feature based cluster analysis according to claim 2, is characterized in that, generating dutation Y of equal value of described PV square formation day _arepresent that PV square formation every kW square formation every day monitored is supplied to the energy of PV system, be equivalent to PV square formation with output rating P ₀run the hourage needing work,

Y _A＝E _A/P ₀

Wherein E _afor PV square formation output energy, unit is kW ˙ h.P ₀representing PV system peak watt power, is each battery component peak watt power sum, and the general power of PV system when namely each battery component presses rated power operation, unit is kW.

8. the photovoltaic plant running status division methods of feature based cluster analysis according to claim 2, it is characterized in that, described PV system day represents that PV system every kW square formation every day monitored is supplied to the energy of transformer station's input side when equivalence generates electricity, and is equivalent to PV system with output rating P ₀need the hourage of work,

Y _F＝E _AC/P ₀

Wherein E _aCfor PV system exports gross energy, unit is kW ˙ h; P ₀representing PV system peak watt power, is each battery component peak watt power sum, and the general power of PV system when namely each battery component presses rated power operation, unit is kW.