阅读说明：本技术 一种基于机器学习的光伏系统直流故障电弧检测方法 (Photovoltaic system direct current fault arc detection method based on machine learning ) 是由 孙耀杰 樊宏涛 马磊 于 2019-09-03 设计创作，主要内容包括：本发明属于光伏电气故障检测领域,具体为一种基于机器学习训练的光伏系统直流故障电弧检测方法。该方法包括以下步骤：逆变器采集光伏系统直流侧的实时电流信号；分析得到其频域特征和时域特征；通过三种模型判断故障电弧是否发生；如果判定光伏系统产生了直流故障电弧,则启动报警机制。本发明采用基于机器学习的检测算法可以提高检测准确性,小电流和大电流情况下均实用；也能避免阈值设定不能适应所有情况带来的误检操作,有效降低误检率。(The invention belongs to the field of photovoltaic electrical fault detection, and particularly relates to a photovoltaic system direct-current fault arc detection method based on machine learning training. The method comprises the following steps: the method comprises the steps that an inverter collects real-time current signals of a direct current side of a photovoltaic system; analyzing to obtain frequency domain characteristics and time domain characteristics of the target; judging whether a fault arc occurs or not through the three models; and if the photovoltaic system is judged to generate the direct-current fault arc, starting an alarm mechanism. The invention adopts a detection algorithm based on machine learning, which can improve the detection accuracy and is practical under the conditions of small current and large current; the false detection operation caused by the fact that the threshold setting cannot adapt to all conditions can be avoided, and the false detection rate is effectively reduced.)

1. A photovoltaic system direct current fault arc detection method based on machine learning is characterized in that the detection method is realized through a photovoltaic system; the photovoltaic system comprises a photovoltaic array, a combiner box, an inverter and an alternating current power grid; the output end of the photovoltaic array is connected with the input end of the combiner box, the output end of the combiner box is connected with the input end of the inverter, the output end of the inverter is connected with an alternating current power grid, a current collecting device is arranged between the combiner box and a circuit connected with the inverter, and a switch is arranged between the photovoltaic array and the combiner box; the detection method comprises the following steps:

(1): the inverter acquires a real-time current signal obtained by a photovoltaic array on the direct current side of the photovoltaic system through a current acquisition device; the sampling time is T1 seconds, and the interval time between two adjacent samplings is T2 seconds;

(2): analyzing the real-time current signal acquired in the step (1) to obtain the frequency domain characteristic and the time domain characteristic of the real-time current signal;

When the time domain characteristics show that direct current fault arc occurs, the current average value a1 is suddenly reduced, and the degree of reduction is determined by the specific power level of the photovoltaic system; the current average value a1 is calculated asWherein a1 is the average value of the current, A_iThe value of the sampled current is N, and the number of the samples is N;

When the time domain characteristics show that a direct current fault arc occurs, the current variance a2 suddenly increases, and the degree of the increase is determined by specific photovoltaic system configuration parameters; the current variance a2 is calculated aswherein a2 is the current variance, A_iThe value of the sampled current is a1, the average value of the current is a, and N is the number of samples;

A specific frequency band f when the frequency domain characteristics show that a direct current fault arc occurs_a-f_bThen, the variance a3 of the wavelet coefficient of the wavelet decomposition suddenly increases, and the degree of the increase is determined by the specific power level of the photovoltaic system; the variance a3 is calculated asWherein a3 is the variance, d_iIs the coefficient after wavelet decomposition of the current signal,Is N numberThe average value of the number of samples, N is the number of samples; f. of_aFor the lower limit of the selected frequency band interval, f_bis the upper limit of the selected frequency band interval;

A specific frequency band f when the frequency domain characteristics show that a direct current fault arc occurs_a-f_bThen, the energy a4 of the wavelet coefficient of the wavelet decomposition is suddenly increased, and the degree of the increase is determined by the specific power level of the photovoltaic system; the calculation formula of the energy a4 is a4 ═ d²Wherein d is the coefficient of wavelet decomposition under a specific frequency band; f. of_aFor the lower limit of the selected frequency band interval, f_bIs the upper limit of the selected frequency band interval;

A specific frequency band f when the frequency domain characteristics show that a direct current fault arc occurs_a-f_bThen, the absolute value a5 of the maximum value minus average value of the wavelet coefficient d of the wavelet decomposition is suddenly increased, and the degree of the increase is determined by the specific power level of the photovoltaic system; the absolute value a5 is calculated asIn the formula (d)_maxFor the maxima of the wavelet decomposition coefficients at a particular frequency band,Is the average value of the coefficients of wavelet decomposition under a specific frequency band, f_aFor the lower limit of the selected frequency band interval, f_bIs the upper limit of the selected frequency band interval;

(3): identifying whether the direct-current fault arc occurs in 3 models of a random forest, a support vector machine and a decision tree by using the time-frequency characteristics and the time-domain characteristics of the fault arc characteristics in the step (2); reading a Flag1 value, and calculating a Flag value;

(3.1) if flag is 0; judging that the detected direct-current fault arc does not occur at this time;

(3.2) if flag is 1; judging that the detected direct-current fault arc is possible to occur at this time, and shortening the sampling time;

(3.3) if flag is 2; judging the occurrence of the direct current fault arc detected at this time;

(4): if the photovoltaic system is judged to generate the direct-current fault arc in the step (3), Flag +1, and judgment is carried out:

(4.1) if Flag is 2 at this time; judging the occurrence of direct-current fault arc; the breaker acts to break the circuit and send out alarm information;

(4.2) if Flag is 1 at this time; delaying for 200ms to carry out next detection;

If the direct current fault arc is not detected, the Flag value is still judged:

(4.3) if Flag is 1 at this time; setting Flag1 to 1, and storing the value of Flag1 for waiting for the next detection;

If Flag is 0 at this time; and clearing 0 for Flag1, and waiting for the next detection.

2. The photovoltaic system direct-current fault arc detection method based on machine learning as claimed in claim 1, wherein the current collection device provided between the combiner box and the inverter of the photovoltaic system is a series coil induction type real-time current collection device, which is used for realizing the collection of the real-time current signal on the direct-current side of the photovoltaic system in the step (1).

Technical Field

The invention belongs to the field of photovoltaic electrical fault detection, and particularly relates to a photovoltaic system direct-current fault arc detection method based on machine learning.

background

arcing is a gas discharge phenomenon, which refers to the momentary spark produced by a current passing through some insulating medium (e.g., air). Arc discharge is a self-sustaining discharge that is distinguished from other types of discharge by a very low sustaining voltage. At present, it is difficult to define the arc discharge strictly, and the arc discharge is a discharge with a reduced cathode potential and a large current density, and generally has a negative volt-ampere characteristic, simply from the electrical characteristic of the discharge.

In a photovoltaic system, once a fault electric arc is generated, if a timely and effective protective measure is not taken, a continuous direct current electric arc can generate a high temperature of more than 3000 ℃, and then a fire disaster is caused. In recent years, fire accidents caused by fault arcs occur in European and American countries, and equipment damage is caused to different degrees. The united states electrical code (NEC) of 2011 stipulates that a photovoltaic system should be equipped with a detection device and a circuit breaker for detecting a fault arc. The Underwriters Laboratories (UL) also introduced corresponding development testing methods and mechanisms.

At present, most researchers put forward a detection method for passive detection aiming at the characteristics of the arc, and the defect is that under some large current conditions, the arc characteristics are not obvious, and false detection is easily caused. The false detection can cause shutdown of the whole photovoltaic system once occurring, and unnecessary loss is brought.

disclosure of Invention

The invention aims to provide a photovoltaic system direct-current fault arc detection method based on machine learning, which is obvious in arc characteristic performance and not easy to cause false detection.

the invention provides a photovoltaic system direct current fault arc detection method based on machine learning, which is realized by a photovoltaic system; the photovoltaic system comprises a photovoltaic array, a combiner box, an inverter and an alternating current power grid; the output end of the photovoltaic array is connected with the input end of the combiner box, the output end of the combiner box is connected with the input end of the inverter, the output end of the inverter is connected with an alternating current power grid, a current collecting device is arranged between the combiner box and a circuit connected with the inverter, and a switch is arranged between the photovoltaic array and the combiner box; the detection method comprises the following steps:

(1): the inverter acquires a real-time current signal obtained by a photovoltaic array on the direct current side of the photovoltaic system through a current acquisition device; the sampling time is T1 seconds, and the interval time between two adjacent samplings is T2 seconds;

(2): analyzing the real-time current signal acquired in the step (1) to obtain the frequency domain characteristic and the time domain characteristic of the real-time current signal;

When the time domain characteristics show that direct current fault arc occurs, the current average value a1 is suddenly reduced, and the degree of reduction is determined by the specific power level of the photovoltaic system; the current average value a1 is calculated asWherein a1 is the average value of the current, A_iThe value of the sampled current is N, and the number of the samples is N;

When the time domain characteristics show that a direct current fault arc occurs, the current variance a2 suddenly increases, and the degree of the increase is determined by specific photovoltaic system configuration parameters; the current variance a2 is calculated asWherein a2 is the current variance, A_ithe value of the sampled current is a1, the average value of the current is a, and N is the number of samples;

A specific frequency band f when the frequency domain characteristics show that a direct current fault arc occurs_a-f_bThen, the variance a3 of the wavelet coefficient of the wavelet decomposition suddenly increases, and the degree of the increase is determined by the specific power level of the photovoltaic system; the variance a3 is calculated aswherein a3 is the variance, d_iIs the coefficient after wavelet decomposition of the current signal,the average value of N sampling numbers is obtained, and N is the sampling number; f. of_afor the lower limit of the selected frequency band interval, f_bIs the upper limit of the selected frequency band interval;

a specific frequency band f when the frequency domain characteristics show that a direct current fault arc occurs_a-f_bthen, the energy a4 of the wavelet coefficient of the wavelet decomposition is suddenly increased, and the degree of the increase is determined by the specific power level of the photovoltaic system; the calculation formula of the energy a4 is a4 ═ d²in the formula, d is speciallyand (4) coefficients of wavelet decomposition under a fixed frequency band. f. of_aFor the lower limit of the selected frequency band interval, f_bIs the upper limit of the selected frequency band interval;

a specific frequency band f when the frequency domain characteristics show that a direct current fault arc occurs_a-f_bThen, the absolute value a5 of the maximum value minus average value of the wavelet coefficient d of the wavelet decomposition is suddenly increased, and the degree of the increase is determined by the specific power level of the photovoltaic system; the absolute value a5 is calculated asIn the formula (d)_maxFor the maxima of the wavelet decomposition coefficients at a particular frequency band,The coefficient average value of wavelet decomposition under a specific frequency band is obtained; f. of_aFor the lower limit of the selected frequency band interval, f_bis the upper limit of the selected frequency band interval;

(3): and (3) respectively identifying whether the direct-current fault arc occurs in 3 models of a random forest, a support vector machine and a decision tree by using the time-frequency characteristics and the time-domain characteristics of the fault arc characteristics in the step (2). The Flag1 value was read and the Flag value was calculated.

(3.1) if flag is 0; and judging that the direct current fault arc is not generated in the current detection.

(3.2) if flag is 1; the method judges that the detected direct-current fault arc is possible to occur, and shortens the sampling time.

(3.3) if flag is 2; and judging that the direct current fault arc is generated at the current time.

(4): if the photovoltaic system is judged to generate the direct-current fault arc in the step (3), Flag +1, and judgment is carried out:

(4.1) if Flag is 2 at this time; and judging the occurrence of the direct-current fault arc. The breaker acts to break the circuit and send out alarm information.

(4.2) if Flag is 1 at this time; delaying 200ms for next detection.

If the direct current fault arc is not detected, the Flag value is still judged:

(4.3) if Flag is 1 at this time; setting Flag1 to 1, and storing the value of Flag1 for waiting for the next detection;

If Flag is 0 at this time; and clearing 0 for Flag1, and waiting for the next detection.

In the invention, the current collecting device arranged between the combiner box and the inverter of the photovoltaic system is a coil-in induction type real-time current collecting device which is used for collecting real-time current signals of the direct current side of the photovoltaic system in the step (1).

Compared with the prior art, the invention has the beneficial effects that:

(1) The detection accuracy can be improved by adopting a detection mode based on machine learning, and the method is practical under the conditions of low current and large current;

(2) The method can adapt to various conditions of the photovoltaic system during actual operation, and the false detection rate is reduced.

Drawings

FIG. 1 is a schematic diagram of real-time current collection locations for detecting a DC fault arc in a photovoltaic system.

fig. 2 is a flow chart of steps of a photovoltaic system dc fault arc detection method based on machine learning detection.

Reference numbers in the figures: the photovoltaic array is used as 1, the combiner box is used as 2, the inverter is used as 3, the alternating current power grid is used as 4, and the real-time current acquisition device is used as 5.

Detailed Description

the present invention will be described in detail with reference to the following embodiments, which are provided for illustration and not for limitation.