CN114217165A - Fault diagnosis method and device for cascade H-bridge photovoltaic inverter - Google Patents

Abstract

The invention provides a fault diagnosis method and device for a cascade H-bridge photovoltaic inverter, and belongs to the technical field of photovoltaic power generation. The method comprises the following steps: setting a topological structure of the cascaded H-bridge photovoltaic inverter, and controlling the grid-connected current of the cascaded H-bridge photovoltaic inverter; analyzing the fault mode of the cascade H-bridge photovoltaic inverter; extracting features by adopting wavelet packet energy entropy; solving the optimal kernel function bandwidth and bias by adopting an improved bacterial foraging optimization algorithm; and according to the optimal kernel function bandwidth and bias, training data by using a support vector machine and performing fault diagnosis. Compared with an SVM fault diagnosis method, the method for optimizing the SVM fault diagnosis by using the improved bacterial optimization algorithm realizes the characteristic extraction of the switching tube fault of the cascaded H-bridge photovoltaic inverter and the diagnosis and classification of the switching tube fault of the cascaded H-bridge photovoltaic inverter, and can effectively improve the accuracy of the open-circuit fault diagnosis of the three-phase cascaded H-bridge photovoltaic inverter.

Description

Fault diagnosis method and device for cascade H-bridge photovoltaic inverter

Technical Field

The invention belongs to the technical field of photovoltaic power generation, and particularly relates to a fault diagnosis method and device for a cascade H-bridge photovoltaic inverter.

Background

The energy is a power developed by the country, the main energy supply form of China is thermal power generation at present, the steam power is obtained by burning fossil fuel to push a generator rotor to rotate for generating electricity, but the renewable speed of the fossil fuel is very low, the output and the consumption are unbalanced, in addition, the consumption of the fossil fuel can cause great harm to the ecological environment, and in order to alleviate the problems, new energy with cleanness and sustainability is generated at the right moment, and the new energy becomes a main direction of technical attack.

The photovoltaic power generation is clean and sustainable, and is an important component in a new energy structure by converting energy in solar illumination into electric energy by using a photovoltaic panel and transmitting the electric energy to a power grid or an energy storage battery. In a photovoltaic power generation system, the output voltage of a photovoltaic panel is direct current, but electric energy transmission in China is mostly transmitted through alternating current, so that equipment for converting direct current into alternating current, namely an inverter, is essential for power generation grid connection or formation of a microgrid. An inverter in a traditional photovoltaic grid-connected system usually adopts a two-level three-phase bridge type topological structure, but with the increase of grid-connected scale, the pressure born by a switch tube is increased, the number of the switch tubes in a multi-level topological structure is greatly increased, and the integral performance of a circuit is favorably improved. The cascaded H bridge is one of multi-level topological structures, has the characteristics of simple circuit and easy expansion, but for a system with more cascaded H bridge units, the number of switching tubes is multiplied, and each switching tube is a potential fault point.

Therefore, effective fault judgment and diagnosis need to be performed on the faults of the switching tubes of the cascaded H-bridge photovoltaic inverter.

Disclosure of Invention

The invention aims to at least solve one of the technical problems in the prior art and provides a fault diagnosis method and device for a cascade H-bridge photovoltaic inverter.

In one aspect of the present invention, a fault diagnosis method for a cascade H-bridge photovoltaic inverter is provided, which includes the following steps:

setting a topological structure of a cascade H-bridge photovoltaic inverter;

controlling the grid-connected current of the cascade H-bridge photovoltaic inverter;

analyzing the fault mode of the cascade H-bridge photovoltaic inverter;

extracting features by adopting wavelet packet energy entropy;

solving the optimal kernel function bandwidth and bias by adopting an improved bacterial foraging optimization algorithm;

and according to the optimal kernel function bandwidth and the bias, training data by using a support vector machine and performing fault diagnosis.

Optionally, the controlling the grid-connected current of the cascade H-bridge photovoltaic inverter includes:

the sinusoidal quantity is converted into direct current quantity for control by using an abc-dq coordinate transformation method, and an output equation of a grid-connected current control link is as follows:

wherein u is_d、u_q，i_d、i_qThe output voltage and the output current of the inverter under the dq coordinate system are respectively; i.e. i^* _qThe reactive current instruction value is set according to the requirement of the power grid on reactive power; i.e. i^* _dFor active current command value, it is necessaryTo match the output power capability of the photovoltaic module; omega is the angular frequency of the alternating current; e.g. of the type_d、e_qThe grid voltage is the grid voltage under the dq coordinate system;

is the formula for PI, K_pIs a proportional parameter, K_iIs an integration parameter, 1/S represents integration; ω is 2 × fpwm, representing the angular frequency, fpwm representing the frequency of the PWM modulated wave.

Optionally, the analyzing the fault mode of the cascade H-bridge photovoltaic inverter includes:

and acquiring IGBT open circuit fault data by establishing a simulation model.

Optionally, the performing feature extraction by using wavelet packet energy entropy includes:

carrying out j-layer wavelet packet decomposition on the fault signal to obtain 2^jSub-signals of different frequency bands are rearranged according to the sequence from the low frequency band to the high frequency band;

wavelet packet reconstruction is performed on each frequency band, and the reconstruction coefficient is expressed as S_j,k(k＝0,1,...,2^j-1)；

Calculating the wavelet packet energy value E of each frequency band_j,k(k＝0,1,...,2^j-1)；

In the formula, t_i-1-t_iIs the start-stop time of the frequency band signal;

normalizing the energy of each frequency band to obtain normalized energy;

and (4) measuring the energy entropy of the kth wavelet packet on the jth layer after the fault signal is decomposed.

Optionally, the wavelet packet energy value of each frequency band adopts the following relation:

and/or the presence of a gas in the gas,

the normalized energy P_j,kThe following relationship is used:

and/or the presence of a gas in the gas,

the energy entropy measure of the kth wavelet packet at the jth layer adopts the following relational expression:

optionally, the improved bacterial foraging optimization algorithm is implemented by adding a Levy flight optimization strategy into the bacterial foraging optimization algorithm, and the update formula of the new position of the flora is as follows:

wherein u and v are both subject to a normal distribution,

representing the position of the current best flora, alpha representing a step factor, and performing point-to-point multiplication with L (lambda);

the position of the current population is indicated,

representing the current population location after the Levy flight update.

Optionally, the bacterial foraging optimization algorithm mainly comprises the following steps:

initializing parameters p, S, N_C、N_S、N_re、N_ed、P_edC (i ═ 1,2, …, S) and θⁱ；

The migration operation cycle l is l + 1;

a copy operation cycle k is k + 1;

trending the operating cycle j ═ j + 1;

if j < N_CReturning to the fourth step to perform the tropism operation;

for a given k, l and each1,2, …, S, and converting the bacterial energy value J to_healthArranged from small to large, eliminating the former S_rS after replication/2 smaller energy value bacteria_rBacteria with larger energy values, each bacteria dividing into identical bacteria;

if k is less than N_reReturning to the third step;

migration: after several generations of replication operations of the flora, each bacterium has a probability P_edIs re-randomly distributed into the optimization space;

if l is less than N_edReturning to the second step, otherwise ending the optimization.

Optionally, the trend operation cycle j ═ j +1 includes:

bacteria i were driven to one step as follows: 1,2, …, S;

calculating an adaptive value function J (i, J, k, l);

let J_last(i, j, k, l) stored as the best current fitness value for bacterium i;

rotating: generating a random vector Δ (i) ∈ R^PEach element of which is Δ_m(i) (m-1, 2, …, p) are distributed in [ -1, 1 [ -1 [ ]]A random number of (c);

moving: order to

Wherein, C (i) is the step size of the bacteria i moving along the direction randomly generated after rotation;

calculating J (i, J +1, k, l), and let

J(i,j+1,k,l)＝J(i,j,k,l)+J_CC(θⁱ(j+1,k,l),P(j+1,k,l))；

Swimming: (ii) m is 0; m < N_s(ii) a Let m equal m +1, if J (i, J +1, k, l) < J_lastLet J_lastJ (i, J +1, k, l) and

returning to the sixth step by thetaⁱ(j+1, k, l) to calculate a new J (i, J +1, k, l);

otherwise, let m equal to N_s(ii) a Returning to the second step, processing the next bacterium i + 1.

Optionally, the performing fault diagnosis by using support vector machine training data according to the optimal kernel function bandwidth and bias includes:

substituting the optimal kernel function bandwidth and bias into the expression of the support vector machine to train and diagnose the fault signal; wherein the content of the first and second substances,

the expression of the final classification of the SVM is as follows:

wherein alpha is_iIs Lagrange multiplier, b is threshold, K (x)_iX) is a kernel function; x is the number of_i∈Rⁿ,y_i∈{1,-1},i＝1,2,3,…,n,。

In another aspect of the present invention, a fault diagnosis apparatus for a cascade H-bridge photovoltaic inverter is provided, including: the device comprises a topological structure setting module, a control module, an analysis module, a feature extraction module, a solving module and a fault diagnosis module; wherein the content of the first and second substances,

the topological structure setting module is used for setting a topological structure of the cascaded H-bridge photovoltaic inverter;

the control module is used for controlling the grid-connected current of the cascade H-bridge photovoltaic inverter;

the analysis module is used for analyzing the fault mode of the cascade H-bridge photovoltaic inverter;

the characteristic extraction module is used for extracting characteristics by adopting wavelet packet energy entropy;

the solving module is used for solving the optimal kernel function bandwidth and bias by adopting an improved bacterial foraging optimization algorithm;

and the fault diagnosis module is used for adopting support vector machine training data to carry out fault diagnosis according to the optimal kernel function bandwidth and the bias.

The invention provides a fault diagnosis method of a cascade H-bridge photovoltaic inverter, which comprises the following steps: setting a topological structure of a cascade H-bridge photovoltaic inverter, and controlling the grid-connected current of the cascade H-bridge photovoltaic inverter; analyzing the fault mode of the cascade H-bridge photovoltaic inverter; extracting features by adopting wavelet packet energy entropy; solving the optimal kernel function bandwidth and bias by adopting an improved bacterial foraging optimization algorithm; and according to the optimal kernel function bandwidth and the bias, training data by using a support vector machine and performing fault diagnosis. Compared with an SVM fault diagnosis method, the SVM fault diagnosis method optimized by using the improved bacterial optimization algorithm provided by the invention has the advantages that the characteristic extraction of the switching tube fault of the cascaded H-bridge photovoltaic inverter is realized, the diagnosis and classification of the switching tube fault of the cascaded H-bridge photovoltaic inverter are realized, and the open-circuit fault diagnosis accuracy of the three-phase cascaded H-bridge photovoltaic inverter can be effectively improved.

Drawings

Fig. 1 is a block flow diagram of a fault diagnosis method for a cascade H-bridge photovoltaic inverter according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a three-phase cascade H-bridge photovoltaic inverter system according to another embodiment of the invention;

FIG. 3 is a schematic structural diagram of an H-bridge unit according to another embodiment of the present invention;

FIG. 4 is a diagram of abc-dq coordinate transformation vectors according to another embodiment of the present invention;

FIG. 5 is a flow chart of a bacterial foraging algorithm according to another embodiment of the present invention;

FIG. 6 is a schematic diagram of an optimal hyperplane view of another embodiment of the present invention;

fig. 7 is a schematic diagram of a fault diagnosis apparatus for a cascaded H-bridge photovoltaic inverter according to another embodiment of the present invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the described embodiments of the invention without any inventive step, are within the scope of protection of the invention.

As shown in fig. 1, in one aspect of the present invention, a fault diagnosis method S100 for a cascade H-bridge photovoltaic inverter is provided, which includes the following specific steps S110 to S160:

and S110, setting a topological structure of the cascaded H-bridge photovoltaic inverter.

As shown in fig. 2, in the three-phase cascade H-bridge photovoltaic inverter system structure, each phase of the inverter side is formed by connecting H-bridge units end to end, and the direct current sides of the H-bridge units are connected with photovoltaic modules independent of each other, so that each phase of the inverter side outputs a voltage u_ao、u_bo、u_coRespectively equal to the sum of the output voltages of the H bridge units of each phase. e.g. of the type_a、e_b、e_cRepresenting the three-phase grid voltage, a three-phase filter is provided between the inverter side and the grid voltage, and fig. 2 illustrates a single inductor filter, so that the voltage across the inductor directly determines the characteristics of the grid-connected current, such as amplitude and phase.

As shown in FIG. 3, given the topology of each H-bridge unit, the actions of V1 and V2 are complementary, the actions of V3 and V4 are complementary, and the H-bridge output voltages under different switching action combinations are different, so that the H-bridge unit can output u_dc、0、-u_dcThree levels, if each phase is cascaded by n cells, a total of 2n +1 levels may be output.

And S120, controlling the grid-connected current of the cascade H-bridge photovoltaic inverter.

Specifically, the three-phase cascade H-bridge photovoltaic inverter topology is shown in fig. 2, and the following voltage balance equation can be written according to the related knowledge of the circuit:

wherein L represents the filter inductance, i_a、i_b、i_cFor three-phase grid-connected current, u_ao、u_bo、u_coFor the three-phase inverter side output voltage, e_a、e_b、e_cFor three-phase mains voltage, u_onFor a voltage between o and n, for Sinusoidal Pulse Width Modulation (SPWM), u is determined without taking into account the high-frequency voltage components due to the switching action_onEqual to 0, the grid-connected current is directly influenced by the output voltage of the inverter, and obviously, the output voltage of the inverter is a control means in a grid-connected current control loop. Therefore, the input of the grid-connected current control link is the error between the grid-connected current instruction and the feedback, and the error is calculated by a regulator and the like to obtain the output of the grid-connected current control link, namely the output voltage of the inverter side of the control means. For the cascade H-bridge photovoltaic inverter, the output voltage of each H-bridge unit is accumulated to obtain the total output voltage of the phase, so that after the total output voltage is obtained by calculating a grid-connected current loop, how to bear the total output voltage by each unit is a degree of freedom, but considering the maximization of the total output power, the unit with large output power capability correspondingly bears more output voltages by taking power distribution as a bearing principle. Because the grid-connected current is in sinusoidal variation, if the regulator adopts a traditional PI structure, the regulating effect is poor, and therefore, the method of converting the sinusoidal quantity into the direct-current quantity by using abc-dq coordinate transformation can be considered for control.

As shown in fig. 4, the principle of abc-dq coordinate transformation is as follows: e represents a grid voltage vector, and the projection of the grid voltage vector on three axes of a, b and c is E_a、e_b、e_cI is a grid-connected current vector, and the projection of the grid-connected current vector on three axes a, b and c is I_a、i_b、i_cInstantaneous value of (a). When the included angle between the q axis and the a axis and the voltage e of the power grid phase_aIf the phase angles of the two vectors are the same, the d axis and the vector E are superposed to form an active axis, and the q axis is perpendicular to the E to form a reactive axis, so that the projection I of the vector I on the d axis_dCan be regarded as a real current component, i projection on the q axis_qFor the reactive current component, from the geometric relationship, the following coordinate transformation expression can be obtained.

Theta in the above formula can be calculated by a phase-locked loop, and details are not described, the angle can be taken as a known quantity when a circuit model is analyzed, and the circuit equation in the dq coordinate system can be obtained by substituting the above formula into the formula voltage balance equation.

With i_d、i_qIs a controlled object, then i_d、i_qThe increment of (i) can be regarded as an error, i.e. (i #)_d-i_d)、(i*_q-i_q) Wherein i^* _dIs an active current command value, i^* _qFor reactive current command values, i for cascaded H-bridge photovoltaic inverters^* _dNeed to match the output power capability of the photovoltaic module, i^* _qThen i can be set according to the reactive power demand of the grid^* _d、i^* _qCan be translated from power requirements and will not be described in detail herein. The time increment dt can be regarded as action time to be removed from the expression, the PI regulator has large direct current gain and good tracking effect, and can be used as a controller under a dq coordinate system to replace a coefficient L in the expression. According to the above description, the above formula can be rewritten as a grid-connected current control link output equation.

Is the formula for PI, K_pIs a proportional parameter, K_iIs an integral parameterWherein 1/S represents integration; ω is 2 × fpwm, representing the angular frequency, fpwm representing the frequency of the PWM modulated wave.

And S130, analyzing the fault mode of the cascade H-bridge photovoltaic inverter.

Specifically, the switching device faults are divided into an IGBT fault and a diode fault, and the faults can be divided into two cases, namely a short-circuit fault and an open-circuit fault. In practical applications, when a short-circuit fault occurs in the IGBT, a protection circuit inside the module is triggered and operated, so that the fuse is broken to form an open-circuit fault. When the short circuit condition is very serious, the protection circuit does not work, and the phenomenon of inverter explosion is often caused.

Therefore, the present embodiment mainly studies the IGBT fault diagnosis for its open circuit fault. IGBT open-circuit fault data are obtained by establishing a simulation model, wavelet packet entropy characteristic extraction is carried out on the open-circuit fault data, and the open-circuit faults are classified by using an SVM (support vector machine) solved by an improved bacterial optimization algorithm.

And S140, extracting the features by adopting wavelet packet energy entropy.

The wavelet packet can decompose the signals in the low-frequency band and the high-frequency band at the same time, adaptively determine the resolution ratio of the signals on different frequency bands, and the signals in each decomposition frequency band are independent, have no redundancy and are not overlooked. The probability that the energy entropy represents the number of states present in the signal and the corresponding probability can be used to assess the complexity of the signal. A fault voltage signal U (t) to be analyzed is set, the length of the signal is N, and the wavelet packet energy entropy solving process of the section of the signal is represented as follows:

1. carrying out j-layer wavelet packet decomposition on the fault signal to obtain 2^jSub-signals of different frequency bands and rearranged in the order of low frequency band to high frequency band.

2. Wavelet packet reconstruction is performed on each frequency band, and the reconstruction coefficient is expressed as S_j,k(k＝0,1,...,2^j-1)。

3. The wavelet packet energy value E of each frequency band is calculated according to the following formula_j,k(k＝0,1,...,2^j-1)。

In the formula, t_i-1-t_iIs the start-stop time of the band signal.

4. Normalizing the energy of each frequency band to obtain normalized energy P_j,k：

5. The energy entropy measure of the kth wavelet packet on the jth layer after the fault signal decomposition is as follows:

and S150, solving the optimal kernel function bandwidth and bias by adopting an improved bacterial foraging optimization algorithm.

Specifically, the basic principle of bacterial foraging optimization algorithm is as follows, assuming that a minimum value of J (theta) is required, where theta ∈ R^PAnd gradient of

It cannot be quantitatively or qualitatively analyzed. To solve the non-gradient function optimization problem, the bacterial foraging algorithm simulates four main operations in a real bacterial system: tendency, aggregation, replication and migration, each real bacterium is regarded as an optimal solution of the optimization problem, namely a test solution moving on a function surface, and the process of finding food by the movement of escherichia coli is the process of finding the optimal solution.

To simulate the behavior of actual bacteria, a symbolic description is first introduced, as shown in the following table:

let P (i, k, l) be { theta }ⁱ(i, k, l) | i ═ 1,2, …, S } indicates that individual in population is in the jth tropism operation,J (i, J, k, l) represents the adaptation value function value of the bacterium i after the jth tropism operation, the kth replication operation and the l migration operation.

Tropism operation:

coli has two basic movements throughout the foraging process: rotation and play. Rotation is the turning to find a new direction, while swimming refers to forward movement that keeps the direction the same. The trending operation of the BFA algorithm is a simulation of these two basic actions. Generally, bacteria will swim more in areas of abundance or moderate acidity-alkalinity of the environment, and will rotate more, i.e., stay in place, in areas of food deficiency or high acidity-alkalinity of the environment.

The trending operation of the BFA algorithm is as follows:

firstly, moving one step in a random direction; if the adaptive value in the direction is higher than the adaptive value of the position where the previous step is located, moving forward along the random direction; if the maximum number of attempts is reached, the bacteria's tropism operation is stopped, and the next bacteria is skipped to perform the tropism operation.

Each step tropism operation of the bacteria is shown as follows:

in the above formula, one unit vector in a random direction is represented.

And (3) aggregation operation:

during the process of the flora seeking food, the bacterial individuals achieve aggregation behavior through interaction with each other. There are both attractive and repulsive forces between cells. The bacteria are gathered together by the attraction force, and even the phenomenon of holding the bacteria in a ball is generated. Repulsion forces allow each cell to have a position at which it can acquire energy to sustain survival. This aggregative operation is simulated in the BFA algorithm. The mathematical expression for the aggregation behavior between bacteria is:

in the above formula, d_attractantDepth of gravity, w_attractantWidth of gravity, h_repellantHeight of repulsion, w_repellantIn order to be the width of the repulsive force,

the m-th component of bacterium i, θ_mIs the m-th component of other bacteria in the entire population.

The above formula essentially describes the sum of the forces generated by the entire flora at the location of the bacteria i.

In general d_attractant＝h_repellant。

Due to J_cc(θ, P (j, k, l)) represents the influence of signal transmission between the population bacteria, so after introducing the aggregation operation in the tropism cycle, the fitness value for the ith bacterium is calculated as:

minf(x₁,x₂,...,x_n)

as shown in the above formula, the aggregation operation corrects the fitness value by the above formula so that the bacteria achieve the purpose of aggregation.

A replicative operation:

the biological evolution process always follows Darwin evolution rules, namely survival, excellence and decline of suitable persons. After the BFA algorithm executes a foraging process for a period of time, part of the bacteria which are weak in food source searching capacity (high in fitness value, and the minimum value of the function is mainly used as a description object) are naturally eliminated, and the rest of the bacteria which are strong in food searching capacity (low in fitness value) are bred in order to maintain the population size unchanged. Simulating this phenomenon in the BFA algorithm is called replicative operation.

For a given k, l and each i ═ 1,2, …, S, the definitions are as follows:

the above formula is a health function (also called energy function) of the bacteria i, and is used to measure the energy obtained by the bacteria.

Larger means that the bacterium i is healthier and has stronger foraging ability. Energy J_healthArranged in the order from small to large, eliminating the former S_rS after replication/2 smaller energy value bacteria_rA bacterium with a large energy value to generate S again_rThe progeny bacteria which are completely the same as the parent bacteria with larger original energy value, namely the generated progeny bacteria have the same foraging capacity as the parent bacteria, or the progeny bacteria and the parent bacteria are positioned at the same positions.

Migratory manipulation

The local area where bacteria live in the actual environment may change gradually (e.g., food consumption is over) or suddenly (e.g., temperature rises suddenly, etc.). This may result in bacterial populations living in this local area migrating to a new area or collectively being killed by external forces. Simulating this phenomenon in the BFA algorithm is called migratory operation.

Migration operations, which destroy the bacteria's chemotactic behavior, may therefore seek areas where food is more abundant, and are therefore beneficial for the flora to feed in the long term. To simulate this process, the bacteria are replicated with a given probability P after several generations of the flora in the algorithm_edAnd executing the migration operation, and randomly reallocating the migration operation to the optimization interval. Namely: if a certain bacterial individual in the population meets the probability of migration occurrence, the bacterial individual is killed, a new individual is randomly generated at any position of the solution space, and the new individual and the original individual may have different positions, namely different foraging capacities. The new individual randomly generated by the migration behavior can be closer to the global optimal solution, so that the local optimal solution can be better jumped by the trending operation, and the global optimal solution can be searched.

The bacterial foraging optimization algorithm mainly comprises the following steps:

step 1: first stageInitialization parameters p, S, N_C、N_S、N_re、N_ed、P_edC (i ═ 1,2, …, S) and θⁱ；

Step 2: the migration operation cycle l is l + 1;

and step 3: a copy operation cycle k is k + 1;

and 4, step 4: trending the operating cycle j ═ j + 1;

(1) bacteria i were driven to one step as follows: 1,2, …, S;

(2) calculating an adaptive value function J (i, J, k, l);

let J (i, J, k, l) equal to J (i, J, k, l) + J_CC(θⁱ9j, k, l), P (j, k, l)) (i.e., increasing the force of intercellular attraction to mimic aggregation behavior).

(3) Let J_last(i, j, k, l) stored as the best current fitness value for bacterium i;

(4) rotating: generating a random vector Δ (i) ∈ R^PEach element of which is Δ_m(i) (m-1, 2, …, p) are distributed in [ -1, 1 [ -1 [ ]]A random number of (c);

(5) moving: order to

Wherein, C (i) is the step size of the bacteria i moving along the direction randomly generated after rotation;

(6) calculating J (i, J +1, k, l), and let

J(i,j+1,k,l)＝J(i,j,k,l)+J_CC(θⁱ(j+1,k,l),P(j+1,k,l))；

(7) Swimming: (ii) m is 0; m < N_s(ii) a Let m equal m +1, if J (i, J +1, k, l) < J_lastLet J_lastJ (i, J +1, k, l) and

returning to the sixth step (6) by using the value of thetaⁱ(J +1, k, l) calculating a new J (i, J +1, k, l);

otherwise, let m equal to N_s。

(8) Returning to the second step (2), processing the next bacterium i + 1.

And 5: if j < N_CReturning to the fourth step (4) to perform the tropism operation;

step 6: and (6) copying. For a given k, l and each i ═ 1,2, …, S, the bacterial energy values J are measured_healthArranged from small to large, eliminating the former S_rS after replication/2 smaller energy value bacteria_rThe bacteria with larger energy values, each of which divides into identical bacteria.

And 7: if k is less than N_reReturning to the third step (3);

and 8: migration: after several generations of replication operations of the flora, each bacterium has a probability P_edAre re-randomly distributed into the optimization space. If l is less than N_edReturning to the second step, otherwise ending the optimization.

The specific flow of the above process is shown in fig. 5.

Further, based on the bacterial foraging optimization algorithm, in the improved bacterial foraging optimization algorithm adopted in this embodiment, a Levy flight optimization strategy is added to the bacterial foraging optimization algorithm.

Levy flight (Levy flight) is named after the name of the French mathematician Paul Pierre Levy, and is used to describe the characteristics of an object that is distributed from the heavy tail (heavy-tail) in steps when the object is randomly walked. The heavy tail distribution means that a large value can be taken with a large probability, that is, a large jump is performed at a local position with a large probability to jump out a local optimum so as to enlarge a search range. After observing organisms in nature, the behavior tracks of many organisms (such as fruit flies, reindeer and the like) are very similar to those of Laiwei flight. Therefore, adding this strategy to the flora update formula can be expressed as:

in the formula: α represents a step factor, which is multiplied point-to-point with L (λ). Typically the formula is a random equation with respect to random step size. The random walk follows a markov chain, i.e. the next position depends on the current position and the transfer coefficient. L (λ) represents a random path expressed as:

L(λ)～u＝t ^-λ1＜λ＜3

the updating formula of the new position of the flora is as follows:

wherein u and v are normally distributed.

Representing the location of the current best flora;

the position of the current population is indicated,

representing the current population location after the Levy flight update.

And S160, according to the optimal kernel function bandwidth and the optimal bias, training data by using a support vector machine and performing fault diagnosis.

It should be noted that the principle of Support Vector Machine (SVM) is that the system randomly generates a hyperplane that can be moved, and the movement of this plane makes different classification points in the training set on both sides of the plane, which is essential to find an optimal classification plane, as shown in fig. 6.

The expression of the hyperplane in fig. 6 is (ω · x) + b ═ 0, where ω is the hyperplane normal vector; b is a threshold value. For the training set (x)_i,x_j)，x_i∈Rⁿ,y_iE {1, -1}, i ═ 1,2, 3.

Converting the above problem into an equality problem, the lagrange function is defined as follows:

in the formula, alpha_iIs a lagrange multiplier. Meanwhile, the SVM kernel function selected by the text adopts an RBF kernel function K (x)_iX), so that an expression of the final classification of the SVM can be obtained, as follows:

in the model, two parameters to be optimized are respectively kernel function bandwidth sigma and bias b, the classification accuracy is used as a fitness function, and an improved bacterial foraging optimization algorithm is used for solving the optimal kernel function bandwidth and bias.

As shown in fig. 7, in another aspect of the present invention, there is provided a fault diagnosis apparatus 200 for a cascade H-bridge photovoltaic inverter, including: a topology setting module 210, a control module 220, an analysis module 230, a feature extraction module 240, a solving module 250, and a fault diagnosis module 260; the topology structure setting module 210 is configured to set a topology structure of a cascaded H-bridge photovoltaic inverter; the control module 220 is used for controlling the grid-connected current of the cascade H-bridge photovoltaic inverter; an analysis module 230, configured to analyze a failure mode of the cascaded H-bridge photovoltaic inverter; a feature extraction module 240, configured to perform feature extraction by using wavelet packet energy entropy; a solving module 250 for solving the optimal kernel function bandwidth and bias using an improved bacterial foraging optimization algorithm; and the fault diagnosis module 260 is configured to perform fault diagnosis by using support vector machine training data according to the optimal kernel function bandwidth and bias.

It should be noted that the device of the present embodiment adopts the method described above for fault diagnosis of the cascade H-bridge photovoltaic inverter, and details are not repeated herein.

The following description will be made of a fault diagnosis method for a cascade H-bridge photovoltaic inverter in a specific embodiment:

a MATLAB-based simulink platform builds a three-phase cascade H-bridge photovoltaic inverter simulation, one group of H-bridges is selected to set open-circuit faults for each photovoltaic grid-connected inverter, 1000 groups of open-circuit fault data are extracted, 200 groups of noise signals are randomly extracted from 1000 groups of fault data, and the data with the noise signals and the data without the noise signals are combined into fault data.

1. Carrying out j-layer wavelet packet decomposition on the extracted fault signal to obtain 2^jSub-signals of different frequency bands and rearranged in the order of low frequency band to high frequency band.

2. Wavelet packet reconstruction is performed on each frequency band, and the reconstruction coefficient is expressed as S_j,k(k＝0,1,...,2^j-1)；

3. The wavelet packet energy value E of each frequency band is calculated according to the following formula_j，k(k＝0，1，...，2^j-1)：

In the formula, t_i-1-t_iIs the start-stop time of the band signal.

4. Normalizing the energy of each frequency band to obtain normalized energy P_j,k：

5. The energy entropy measure of the kth wavelet packet of the 4 th layer after the fault signal decomposition is as follows:

6. the entropy difference obtained by wavelet packet decomposition to layer 4 is large and can be used as a characteristic value of the fault.

7. When SVM training data are used, kernel function bandwidth and bias of an SVM need to be solved, an improved bacterial foraging algorithm is used for solving, and accuracy obtained by SVM solution is used as a fitness function;

8. in bacterial foraging algorithm, parameters p, S and N are initialized_C、N_S、N_re、N_ed、P_edC (i ═ 1,2, …, S) and θⁱ。

9. Carrying out Levy flight operation on the bacterial population;

10. the migration operation cycle l is l + 1;

11. a copy operation cycle k is k + 1;

12. the operation cycle j is trended to j + 1.

(1) Bacteria i were driven to one step as follows: i is 1,2, …, S.

(2) An adaptive value function J (i, J, k, l) is calculated. Let J (i, J, k, l) equal to J (i, J, k, l) + J_CC(θⁱ(j, k, l), P (j, k, l)) (i.e., increasing the force of intercellular attraction to mimic aggregation behavior).

(3) Let J_last(i, j, k, l) is stored as the best current adaptation value for bacterium i.

(4) Rotating: generating a random vector Δ (i) ∈ R^PEach element of which is Δ_m(i) (m-1, 2, …, p) are distributed in [ -1, 1 [ -1 [ ]]The random number of (2).

(5) Moving: order to

Wherein C (i) is the step size of bacteria i moving along the direction randomly generated after rotation.

(6) Calculating J (i, J +1, k, l), and let

J(i,j+1,k,l)＝J(i,j,k,l)+J_CC(θⁱ(j+1,k,l),P(j+1,k,l))

(7) Swimming:

①m＝0；

②m＜N_s；

let m equal m +1, if J (i, J +1, k, l) < J_lastLet J_lastJ (i, J +1, k, l) and

returning to step (6) by using the value of thetaⁱ(J +1, k, l) calculating a new J (i, J +1, k, l);

otherwise, let m equal to N_s。

(8) Returning to the step (2), and processing the next bacterium i + 1.

13. If j < N_CReturning to the step 11 to perform the trending operation;

14. and (6) copying. For a given k, l and each i ═ 1,2, …, S, the bacterial energy values J are measured_healthArranged in the order from small to large. Eliminate S before elimination_rS after replication/2 smaller energy value bacteria_rThe bacteria with larger energy values, each of which divides into identical bacteria.

15. If k is less than N_reThen returning to the step 10;

16. and (4) migration. After several generations of replication operations of the flora, each bacterium has a probability P_edAre re-randomly distributed into the optimization space. If l is less than N_edReturning to the step 9, otherwise ending the optimization.

17. Substituting the optimized SVM kernel function bandwidth and bias solution, training the fault signal and diagnosing.

The invention provides a fault diagnosis method and a fault diagnosis device for a cascade H-bridge photovoltaic inverter, which have the following beneficial effects compared with the prior art:

firstly, the invention can provide a method for processing open-circuit faults of a three-phase cascade H-bridge photovoltaic inverter and a method for extracting fault characteristics and carrying out fault diagnosis, and provides a new idea for diagnosis of a grid-connected inverter of a photovoltaic power station.

Secondly, compared with the SVM fault diagnosis method, the SVM fault diagnosis method optimized by using the improved bacterial optimization algorithm can effectively improve the accuracy of open-circuit fault diagnosis of the three-phase cascade H-bridge photovoltaic inverter.

It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.

Claims (10)

1. A fault diagnosis method for a cascade H-bridge photovoltaic inverter is characterized by comprising the following steps:

setting a topological structure of a cascade H-bridge photovoltaic inverter;

controlling the grid-connected current of the cascade H-bridge photovoltaic inverter;

analyzing the fault mode of the cascade H-bridge photovoltaic inverter;

extracting features by adopting wavelet packet energy entropy;

solving the optimal kernel function bandwidth and bias by adopting an improved bacterial foraging optimization algorithm;

and according to the optimal kernel function bandwidth and the bias, training data by using a support vector machine and performing fault diagnosis.

2. The method of claim 1, wherein the controlling the cascaded H-bridge photovoltaic inverter grid-connected current comprises:

the sinusoidal quantity is converted into direct current quantity for control by using an abc-dq coordinate transformation method, and an output equation of a grid-connected current control link is as follows:

wherein u is_d、u_q，i_d、i_qThe output voltage and the output current of the inverter under the dq coordinate system are respectively; i.e. i^* _qThe reactive current instruction value is set according to the requirement of the power grid on reactive power; i.e. i^* _dThe active current instruction value needs to be matched with the output power capability of the photovoltaic module; omega is the angular frequency of the alternating current; e.g. of the type_d、e_qThe grid voltage is the grid voltage under the dq coordinate system;

is the formula for PI, K_pIs a proportional parameter, K_iIs an integration parameter, 1/S represents integration; ω is 2 × fpwm, representing the angular frequency, fpwm representing the frequency of the PWM modulated wave.

3. The method of claim 1, wherein analyzing the cascaded H-bridge photovoltaic inverter failure mode comprises:

and acquiring IGBT open circuit fault data by establishing a simulation model.

4. The method according to claim 1, wherein the feature extraction using wavelet packet energy entropy comprises:

carrying out j-layer wavelet packet decomposition on the fault signal to obtain 2^jSub-signals of different frequency bands are rearranged according to the sequence from the low frequency band to the high frequency band;

wavelet packet reconstruction is performed on each frequency band, and the reconstruction coefficient is expressed as S_j,k(k＝0,1,...,2^j-1)；

Calculating the wavelet packet energy value E of each frequency band_j,k(k＝0,1,...,2^j-1)；

In the formula, t_i-1-t_iIs the start-stop time of the frequency band signal;

normalizing the energy of each frequency band to obtain normalized energy;

and (4) measuring the energy entropy of the kth wavelet packet on the jth layer after the fault signal is decomposed.

5. The method of claim 4, wherein the wavelet packet energy value for each frequency band is represented by the following relationship:

and/or the presence of a gas in the gas,

the normalized energy P_j,kBy using a lower partThe following relation:

and/or the presence of a gas in the gas,

the energy entropy measure of the kth wavelet packet at the jth layer adopts the following relational expression:

6. the method of claim 1, wherein the improved bacterial foraging optimization algorithm is a Levy flight optimization strategy added to the bacterial foraging optimization algorithm, and the new location of the flora is updated according to the formula:

wherein u and v are both subject to a normal distribution,

representing the position of the current best flora, alpha representing a step factor, and performing point-to-point multiplication with L (lambda);

the position of the current population is indicated,

representing the current population location after the Levy flight update.

7. The method of claim 6, wherein the bacterial foraging optimization algorithm comprises the following main steps:

initializing parameters p, S, N_C、N_S、N_re、N_ed、P_edC (i ═ 1,2, …, S) andθⁱ；

the migration operation cycle l is l + 1;

a copy operation cycle k is k + 1;

trending the operating cycle j ═ j + 1;

if j < N_CReturning to the fourth step to perform the tropism operation;

for a given k, l and each i ═ 1,2, …, S, the bacterial energy values J are measured_healthArranged from small to large, eliminating the former S_rS after replication/2 smaller energy value bacteria_rBacteria with larger energy values, each bacteria dividing into identical bacteria;

if k is<N_reReturning to the third step;

migration: after several generations of replication operations of the flora, each bacterium has a probability P_edIs re-randomly distributed into the optimization space;

if l<N_edReturning to the second step, otherwise ending the optimization.

8. The method of claim 7, wherein the trending operation cycle j ═ j +1 comprises:

bacteria i were driven to one step as follows: 1,2, …, S;

calculating an adaptive value function J (i, J, k, l);

let J_last(i, j, k, l) stored as the best current fitness value for bacterium i;

rotating: generating a random vector Δ (i) ∈ R^PEach element of which is Δ_m(i) (m-1, 2, …, p) are distributed in [ -1, 1 [ -1 [ ]]A random number of (c);

moving: order to

Wherein, C (i) is the step size of the bacteria i moving along the direction randomly generated after rotation;

calculating J (i, J +1, k, l), and let

J(i,j+1,k,l)＝J(i,j,k,l)+J_CC(θⁱ(j+1,k,l),P(j+1,k,l))；

Swimming: (ii) m is 0; m < N_s(ii) a Let m equal m +1, if J (i, J +1, k, l) < J_lastLet J_lastJ (i, J +1, k, l) and

returning to the sixth step by thetaⁱ(J +1, k, l) calculating a new J (i, J +1, k, l);

otherwise, let m equal to N_s(ii) a Returning to the second step, processing the next bacterium i + 1.

9. The method of claim 1, wherein the performing fault diagnosis using support vector machine training data according to the optimal kernel function bandwidth and bias comprises:

substituting the optimal kernel function bandwidth and bias into the expression of the support vector machine to train and diagnose the fault signal; wherein the content of the first and second substances,

the expression of the final classification of the SVM is as follows:

wherein alpha is_iIs Lagrange multiplier, b is threshold, K (x)_iX) is a kernel function; x is the number of_i∈Rⁿ,y_i∈{1,-1},i＝1,2,3,...,n,。

10. A fault diagnosis device of a cascade H-bridge photovoltaic inverter is characterized by comprising: the device comprises a topological structure setting module, a control module, an analysis module, a feature extraction module, a solving module and a fault diagnosis module; wherein the content of the first and second substances,

the topological structure setting module is used for setting a topological structure of the cascaded H-bridge photovoltaic inverter;

the control module is used for controlling the grid-connected current of the cascade H-bridge photovoltaic inverter;

the analysis module is used for analyzing the fault mode of the cascade H-bridge photovoltaic inverter;

the characteristic extraction module is used for extracting characteristics by adopting wavelet packet energy entropy;

the solving module is used for solving the optimal kernel function bandwidth and bias by adopting an improved bacterial foraging optimization algorithm;

and the fault diagnosis module is used for adopting support vector machine training data to carry out fault diagnosis according to the optimal kernel function bandwidth and the bias.

Images