A kind of control method of the photovoltaic DC-to-AC converter based on particle swarm algorithm and Learning Algorithm
阅读说明：本技术 一种基于粒子群算法和神经网络学习算法的光伏逆变器的控制方法 (A kind of control method of the photovoltaic DC-to-AC converter based on particle swarm algorithm and Learning Algorithm ) 是由 陈亮 王玉龙 黄帅 金尚忠 徐时清 张淑琴 杨凯 谷振寰 杨家军 祝晓明 徐瑞 于 2019-09-10 设计创作，主要内容包括：本发明公开了一种基于粒子群算法和神经网络学习算法的光伏逆变器的控制方法,属于光伏逆变器领域,包括在逆变器基础上加上一种基于粒子群算法和神经网络学习算法相结合的逆变器控制算法,具体包括通过粒子群算法对光伏系统进行离线寻优,得到最优的光伏逆变器参数和利用神经网络对样本进行训练,得到光伏系统的逆模型,对变化的负载进行补偿,减小负载的变化对系统的影响。本发明通过粒子群算法可以选出当前系统状态下的最好的PI参数值,而通过BP神经网络建立相应的逆模型,可以对控制器进行补偿,从而减小变化的负载对系统的影响。(The invention discloses a kind of control methods of photovoltaic DC-to-AC converter based on particle swarm algorithm and Learning Algorithm, belong to photovoltaic DC-to-AC converter field, including adding a kind of inverter control algorithm combined based on particle swarm algorithm and Learning Algorithm on the basis of inverter, it specifically includes and off-line optimization is carried out to photovoltaic system by particle swarm algorithm, it obtains optimal photovoltaic DC-to-AC converter parameter and sample is trained using neural network, obtain the inversion model of photovoltaic system, the load of variation is compensated, influence of the variation to system of load is reduced.The present invention can select the best PI parameter value under current system conditions by particle swarm algorithm, and establish corresponding inversion model by BP neural network, can compensate to controller, to reduce influence of the load to system of variation.)
1. a kind of control method of the photovoltaic DC-to-AC converter based on particle swarm algorithm and reverse transmittance nerve network learning algorithm, special Sign is, including adding a kind of inversion combined based on particle swarm algorithm and Learning Algorithm on the basis of inverter Device control algolithm, specifically comprises the following steps:
S1, Optimizing Search is instructed by interparticle cooperation in the group in particle swarm algorithm and swarm intelligence, changed each time Dai Zhong, particle update the position and speed of oneself by 2 poles, obtain the PI controller parameter of optimal location and speed:
S101, PI controller parameter representated by current particle is inputted in inversion system model, obtains load voltage；
S102, load voltage is brought into objective function and obtains adaptive value；
S103, processing will be optimized in defeated time particle swarm algorithm of adaptive value；
S104, corresponding PI controller parameter is directly exported if the number of iterations reaches maximum, obtains optimal PI controller ginseng Number；Conversely, the number of iterations is then added 1, S101-S103 is repeated, optimal PI controller parameter is obtained；
S2, sample is trained using neural network algorithm, obtains neural network contrary modeling, the load of variation is mended It repays, reduces influence of the load error to system.
2. the controlling party of the photovoltaic DC-to-AC converter according to claim 1 based on particle swarm algorithm and Learning Algorithm Method, which is characterized in that PI controller parameter described in S1 includes proportionality coefficient, integral coefficient and capacitor current feedback coefficient.
3. the controlling party of the photovoltaic DC-to-AC converter according to claim 1 based on particle swarm algorithm and Learning Algorithm Method, which is characterized in that objective function described in S102 are as follows:
The target function value is smaller, and optimizing region is more excellent.
Technical field
The invention belongs to photovoltaic DC-to-AC converter technical fields, and in particular to one kind is based on particle swarm algorithm and neural network learning The control method of the photovoltaic DC-to-AC converter of algorithm.
Background technique
Under different application environments, ac inverter also has higher requirement to the output waveform of inverter.For Output voltage and preferable power factor to high quality, photovoltaic generating system inverter generally use in outer voltage and electric current The double-loop control strategy of ring determines that the key of double-closed-loop control effect is the determination of its PI parameter.Traditional double-closed-loop control Method, needs first to establish accurate system model, then by gathering the PI parameter that examination method etc. is more suitable for.But since photovoltaic is sent out Load variation often occurs for electric system, and sytem matrix, Closed-loop Eigenvalues and dynamic response can change, cause really Fixed parameter is difficult to be suitble to the system after variation.Especially for the nonlinear characteristic of photovoltaic generating system, parameter, which is gathered, tries difficulty, very Difficulty is to having determined parameter into adjustment.
Summary of the invention
The purpose of the present invention is to provide a kind of photovoltaic DC-to-AC converter based on particle swarm algorithm and Learning Algorithm Control method, to solve the problems mentioned in the above background technology.
To achieve the above object, the invention provides the following technical scheme:
A kind of control method of the photovoltaic DC-to-AC converter based on particle swarm algorithm and reverse transmittance nerve network learning algorithm, packet It includes and is calculated on the basis of inverter plus a kind of inverter control combined based on particle swarm algorithm and Learning Algorithm Method specifically comprises the following steps:
S1, Optimizing Search is instructed by interparticle cooperation in the group in particle swarm algorithm and swarm intelligence, each In secondary iteration, particle updates the position and speed of oneself by 2 poles, obtains the PI controller ginseng of optimal location and speed Number:
S101, PI controller parameter representated by current particle is inputted in inversion system model, obtains load voltage；
S102, load voltage is brought into objective function and obtains adaptive value；
S103, processing will be optimized in defeated time particle swarm algorithm of adaptive value；
S104, corresponding PI controller parameter is directly exported if the number of iterations reaches maximum, obtains optimal PI control Device parameter；Conversely, the number of iterations is then added 1, S101-S103 is repeated, optimal PI controller parameter is obtained；
S2, sample is trained using neural network algorithm, obtains neural network contrary modeling, the load of variation is carried out Compensation reduces influence of the load error to system.
As a preferred embodiment, PI controller parameter described in S1 include proportionality coefficient, integral coefficient and Capacitor current feedback coefficient.
As a preferred embodiment, objective function described in S102 are as follows:
The target function value is smaller, and optimizing region is more excellent.
Compared with prior art, the beneficial effects of the present invention are:
The present invention carries out off-line optimization to system by particle swarm algorithm, to obtain this moment optimal inverter ginseng of system Number improves the difficult problem of inverter parameters adjusting；Sample is trained using neural network, the load of variation is carried out Compensation improves slow, harmonic wave of output voltage content height of system dynamic response etc. to make output that can preferably system for tracking change Problem.
Detailed description of the invention
Fig. 1 is the mathematical model figure of single-phase inverter in the present invention；
Fig. 2 is the particle swarm algorithm offline optimization illustraton of model in the present invention；
Fig. 3 is feedback control model figure before the load current in the present invention；
Fig. 4 is the curve graph of the training error in the present invention.
Specific embodiment
Below with reference to embodiment, the present invention will be further described.
The following examples are intended to illustrate the invention, but cannot be used to limit the scope of the invention.Item in embodiment Part can be adjusted according to actual conditions are further, under concept thereof of the invention all to method simple modifications of the invention Belong to the scope of protection of present invention.
The present invention provides a kind of control method of photovoltaic DC-to-AC converter based on particle swarm algorithm and Learning Algorithm, Including adding a kind of inverter control combined based on particle swarm algorithm and Learning Algorithm on the basis of inverter Algorithm specifically comprises the following steps:
S1, Optimizing Search is instructed by interparticle cooperation in the group in particle swarm algorithm and swarm intelligence, each In secondary iteration, particle updates the position and speed of oneself by 2 poles, obtains the PI controller ginseng of optimal location and speed Number.
Referring to Fig. 1, Fig. 1 is the mathematical model of unidirectional inverter, equivalent transformation is carried out to Fig. 1, and changed accordingly Letter.After having abbreviation calculating, corresponding transmission function is obtained:
It is known that photovoltaic system can change with the variation of load by this transmission function.And in practical situations, Load often variation, so the corresponding control algolithm of inverter continues to change.
The inverter parameters of system optimal this moment are found by particle swarm algorithm；Particle swarm algorithm is a kind of multi- extreme value function The effective ways of global optimization instruct Optimizing Search by the swarm intelligence that cooperation and competition interparticle in group generates；Often In an iteration, particle updates the position and speed of oneself by 2 extreme points, and one is that particle itself arrives institute's energy this moment The optimal solution sought claims Pbest, the other is the optimal solution that entire group to current time is found, abbreviation Gbest；Kth time The renewal equation of i-th particle rapidity vi and position Sik+1 when iteration are as follows:
In formula: k indicates the number of iterations；W is inertia weight；C1 and c2 is Studying factors, and c1 is " itself cognition " part, c2 Be " social recognition " part: r1 and r2 obeys the uniform random number on [0,1].
Referring to Fig. 2, population offline optimization controller parameter model specifically comprises the following steps: in photovoltaic system
S101, controller parameter representated by current particle is inputted in inversion system model, obtains load voltage；
S102, load voltage is entered in objective function, obtains adaptive value；
Objective function are as follows:
The target function value is smaller, and optimizing region is more excellent.
S103, processing will be optimized in defeated time particle swarm algorithm of adaptive value；
If S104, target function value reach minimum, illustrate that particle is optimal in this position；If not reaching minimum Value, then illustrate to need to continue optimizing, then need to substitute into S101-S103 to walk again one time, to obtain optimal PI controller ginseng Numerical value.
In above scheme, it should be noted that objective function is determined with some indexs of photovoltaic system, for photovoltaic For system, load voltage error is the index of most critical.
But contain multiple harmonic in load voltage, to guarantee that error and total harmonic distortion factor are as small as possible, by the exhausted of error To value integral and total harmonic distortion factor as optimization object function, the formula are as follows:
In formula: m, n are constant coefficient, related with load feedback voltage trace command voltage, and guarantee the feedback voltage The harmonic content of waveform is few；THD reflects the percentage of the sum of each harmonic proportion, embodies ac output voltage harmonic wave Content；U1 and U2 is respectively load output voltage fundamental voltage amplitude and each harmonic amplitude.
According to feedback control model before the load current in the simulation parameter value and Fig. 3 of table 1 (seeing below), by K1,KP,H (feedback factor) optimization section is set as [1,5000], [0.01,3], [0,2].It is available by searching for document at the same time m,n,c1,c2It is respectively 0.1,0.9,1.45,0.9 with weight coefficient w.K is obtained by the way that particle swarm optimization algorithm is last1,KP, H difference For 979,0.24,0.013.
Parameter
Symbol
Numerical value
Output voltage
Vin
700V
Output power
P0
6KW
Fundamental frequency
f0
50HZ
Output voltage
U0
220HZ
Switching frequency
fsw
10KHZ
Filter capacitor
C
80μF
Filter inductance
L
480μF
Triangle wave amplitude
Vin
1
S2, sample is trained using neural network algorithm, obtains neural network contrary modeling, the load of variation is carried out Compensation reduces influence of the load error to system.
Referring to Fig. 4, Fig. 4 is feedforward control figure obtained from influencing caused by system in order to compensate for load.D is in figure The modulated signal of inverter bridge；IL is inductive current；Due to using discrete variable in practical control, so setting i-th of inverter The control period is i, and according to the analysis to inverter output voltage influence factor, the period is to reference electricity before the period and it Pressure, output voltage, inductive current, the sampling of inverter pwm control signal, using these collection values as input value, current period Pwm control signal d (k) is as output, building one dynamic BP neural network comprising 9 inputs and 1 output.Choose hidden layer mind It is 15 through first number.A large amount of training sample is needed to be trained neural network contrary modeling later, to the representative of table 1 Inverter carries out emulation experiment, especially to choose different linear and non-linear load values and be tested.I is tested to this Choose the loads of 4 kinds of linearity and non-linearities, be to carry out sample collection in the sampling period to 2 microseconds, one is obtained 40000 samples This.Wherein 39000 samples carry out the training of inversion model, are left 1000 samples and carry out accuracy detection to inversion model.
What the invention patent proposed on the basis of inverter model, which go out, a kind of is combined based on population with neural network Inverter control algorithm, it is not only effectively to find out optimal inverter parameters, but also the dynamic of photovoltaic DC-to-AC converter is improved well State response characteristic reduces harmonic content, more adapts to real complex environment, possesses very big application prospect.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding And modification, the scope of the present invention is defined by the appended.