CN116432006A - Photovoltaic grid-connected inverter fault diagnosis method based on CEEMDAN-SE-IHHO-LSTM model - Google Patents

Abstract

The invention relates to a photovoltaic grid-connected inverter fault diagnosis method based on CEEMDAN-SE-IHHO-LSTM model, which comprises the following steps: firstly, obtaining the original data of the voltage of a direct current bus of a photovoltaic inverter containing fault signals, and decomposing the original data into a plurality of eigenmode functions (intrinsic mode function, IMF) by using CEEMDAN; then, calculating a Sample Entropy (SE) value corresponding to each sequence, and reconstructing each decomposed sequence according to the SE size; and finally, improving a standard Harris eagle optimization algorithm, optimizing the super parameters of the LSTM by using the improved algorithm, and realizing the parametric fault diagnosis of the output capacitance of the photovoltaic inverter based on the optimized LSTM model. Through calculation example analysis, the effectiveness of the method for carrying out the parametric fault diagnosis on the DC bus capacitance in the photovoltaic grid-connected inverter is verified, and the accuracy of the method for carrying out the parametric fault diagnosis on the DC bus capacitance in the photovoltaic grid-connected inverter can be effectively improved.

Description

Photovoltaic grid-connected inverter fault diagnosis method based on CEEMDAN-SE-IHHO-LSTM model

Technical Field

The invention belongs to the technical field of fault diagnosis of photovoltaic inverters, and particularly relates to a fault diagnosis method of a photovoltaic grid-connected inverter based on a CEEMDAN-SE-IHHO-LSTM model.

Background

With the increasingly serious global environmental pollution, continuous reduction and exhaustion of fossil energy and other problems, photovoltaic power generation is taken as a renewable energy power generation technology with the most sustainable development prospect, and is widely regarded by relevant scholars at home and abroad, and safe, stable and reliable operation of a photovoltaic grid-connected inverter is important for a photovoltaic power generation system and controlling the impact on a power system after grid connection.

The photovoltaic power generation system is an important device which can utilize the photoelectric effect of an internal photovoltaic array to generate direct current, then the generated variable direct current voltage is inverted into alternating current through a photovoltaic inverter, and finally electric energy is integrated into a power grid or directly provided for a load to use. Once the photovoltaic grid-connected inverter fails, the whole system and other equipment may be damaged or even the personal safety is jeopardized, so that the fault diagnosis of the photovoltaic grid-connected inverter is researched, and the method has extremely important practical significance for solving the actual production safety problem. The faults of the photovoltaic grid-connected inverter are mainly divided into parametric faults and structural faults, the structural faults are usually caused by the fact that circuit elements are damaged to cause system structural change, further circuit operation states are changed, even serious abnormal conditions are caused, structural faults caused by open circuit or short circuit of the circuit elements are obvious in fault characteristics, fault diagnosis is more studied, and the parametric faults are faults caused by performance degradation and parameter degradation of circuit devices under the action of various working stresses. The existing research has less research on parametric fault diagnosis, and the difference of the fault characteristic distinction is not easy to carry out fault diagnosis, so the patent provides a high-accuracy fault diagnosis method which has important significance for reducing the loss of system operation, keeping equipment in normal and stable operation and formulating inverter protection measures.

Disclosure of Invention

Aiming at the problems that the extraction of fault characteristic quantity of a photovoltaic inverter direct current bus capacitor is difficult and the diagnosis accuracy of a diagnosis model is low under different aging degrees, the invention provides a photovoltaic grid-connected inverter fault diagnosis method based on a CEEMDAN-SE-IHHO-LSTM model. The specific process comprises the following steps: firstly, obtaining the original data of the voltage of a direct current bus of a photovoltaic inverter containing fault signals, and decomposing the original data into a plurality of eigenmode functions (intrinsic mode function, IMF) by using CEEMDAN; then, calculating a Sample Entropy (SE) value corresponding to each sequence, and reconstructing each decomposed sequence according to the SE size; and finally, improving a standard Harris eagle optimization algorithm, optimizing the super parameters of the LSTM by using the improved algorithm, and realizing the parametric fault diagnosis of the output capacitance of the photovoltaic inverter based on the optimized LSTM model.

The invention adopts the technical scheme that: a photovoltaic grid-connected inverter fault diagnosis method based on CEEMDAN-SE-IHHO-LSTM model comprises the following steps:

s1: analyzing the parametric faults of the photovoltaic grid-connected inverter;

s2: obtaining voltage signal data of a direct current bus with faults, and decomposing the voltage signal data into a plurality of eigen model functions with different frequencies through CEEMDAN;

s3: calculating SE values of all subsequences, and reconstructing all components after CEEMDAN decomposition into high-frequency, low-frequency and trend sequences;

s4: improving a standard Harris eagle optimization algorithm;

s5: optimizing the super-parameters of LSTM by IHHO, and realizing fault identification of the output capacitor of the photovoltaic inverter;

s6: and (5) evaluating the model diagnosis effect.

In particularThe step S1: and (5) analyzing the parametric faults of the photovoltaic grid-connected inverter. The photovoltaic inverter generally adopts a two-stage structure, the front stage adopts a DC-DC power conversion circuit, the rear stage adopts a three-phase full-bridge inverter circuit, and a double closed-loop control strategy of a direct-current voltage outer ring and a network side current inner ring is adopted to realize direct-current grid connection. The L, T, D, cdc constitutes a front-stage DC-DC power conversion circuit, realizes the tracking control of the maximum power point of the photovoltaic array, converts the voltage output by the photovoltaic array into the direct-current voltage required by the inversion part, and realizes grid-connected control for a three-phase bridge type inversion circuit formed by six insulated gate bipolar transistors with anti-parallel diodes at the rear stage. In a double closed-loop control strategy of the photovoltaic grid-connected inverter, the voltage of the direct-current bus is kept stable by the voltage outer loop control; the inner ring controls the output current, and adjusts the output grid current to be in phase with the grid voltage in the same frequency, so that the unit power factor grid connection is realized. Currently, an aluminum electrolytic capacitor with larger capacity is mostly adopted as a direct current bus capacitor C of a photovoltaic grid-connected inverter _dc Therefore, a large amount of higher harmonic signals generated by the inverter in a high-frequency working mode are absorbed, the effects of stabilizing the voltage of the side of the direct-current bus and storing energy after boosting the inverter are achieved, and the purposes of stably conveying and eliminating harmonic waves to a power grid are achieved. The components of the photovoltaic inverter usually operate in a high-frequency mode, so that the Equivalent Series Resistance (ESR) of the capacitor is increased, the capacitance value is reduced, the aging of the DC bus capacitor is accelerated, the performance of the DC bus capacitor is reduced, and finally, the parametric fault is caused, and even the inverter is damaged or the system is broken. The standard for failure judgment is generally to increase the electrolytic capacitor ESR to 2 to 3 times the initial value or to decrease the capacitance value to 80% of the initial value. However, the direct current bus capacitor ESR value of the photovoltaic inverter system is small and is easily influenced by the working condition of the circuit, and the detection precision is not high. Therefore, the invention reduces the capacitance value to 80% of the initial value as the criterion of the parametric fault of the direct current bus capacitor of the photovoltaic grid-connected inverter.

Specifically, the step S2: and acquiring voltage signal data of the direct current bus with faults, and decomposing the voltage signal data into a plurality of eigen model functions with different frequencies through CEEMDAN. The photovoltaic grid-connected inverter circuit state of Cdc under different aging degrees is less in difference, so that the method has important significance in extracting the characteristics under the fault state. The CEEMDAN can effectively avoid modal aliasing generated when processing parameter fault signals, and can better process and analyze complex data. CEEMDAN is able to decompose raw data into IMFs that contain information of the raw data in different frequency ranges. It introduces an extra signal-to-noise ratio for white noise to control the noise level during each decomposition. The IMF can be completely reconstructed into the original data, almost no noise exists, compared with Empirical Mode Decomposition (EMD), ensemble Empirical Mode Decomposition (EEMD) and Complementary Ensemble Empirical Mode Decomposition (CEEMD), the IMF has higher decomposition rate, and the adaptive noise can be added to eliminate the modal aliasing, and the CEEMDAN is specifically decomposed as follows:

1) Adding Gaussian white noise with the mean value of 0 to the sequence x (t) to be decomposed to construct a sequence xi (t) to be decomposed for K times (i=1, 2,3, …, K).

x _i (t)＝x(t)+εδ _i (t)

Wherein: epsilon is a Gaussian white noise weight coefficient; delta _i (t) is the white noise sequence added the i-th time.

2) EMD decomposing the xi (t) to obtain a first IMF and a residual signal r ₁ (t)。

r ₁ (t)＝x(t)-IMF ₁ (t)

3) And adding specific noise to the residual signal in the j-th stage obtained after decomposition, and continuing EMD decomposition.

r _j (t)＝r _j-1 (t)-IMF _j (t)

And (3) repeatedly executing the step (3), if the residual signal rn (t) obtained by the nth decomposition is a monotone signal, ending the decomposition, otherwise, continuing executing the decomposition.

Specifically, the step S3: and calculating SE values of the subsequences, and reconstructing each component after CEEMDAN decomposition into high-frequency, low-frequency and trend sequences.

The SE can evaluate the complexity of the time series data, and the calculation of the SE does not depend on the length of the data and has better consistency compared with the approximate entropy. The specific calculation process is as follows:

1) The time series X is constructed as an m-dimensional vector:

X(i)＝{x(i),x(i+1),...,x(i+m-1)}

where i=1, 2,..

2) Defining the distance between X (i) and X (j) as d [ X (i), X (j) ] (i not equal to j), wherein the distance is the one with the largest difference value in the corresponding elements:

3) Given a threshold r (r > 0), the number of d [ X (i), X (j) ] < r and the ratio to the total vector N-m:

4) All results from the above formula are averaged:

5) When the dimension m+1 is repeated, the sample entropy of the sequence is theoretically:

6) In practice, however, N cannot be infinity, but is a finite value, and then the estimated value of the sample entropy is:

specifically, the step S4: the standard harris eagle optimization algorithm is improved. The standard Harris eagle optimization algorithm is a novel intelligent optimization algorithm for simulating the predation behavior of the Harris eagle, which is proposed by Heidari et al in 2019, and mainly comprises two stages: the global search stage and the local development stage, wherein the harris hawk inhabitation position is randomly determined in the global search stage, and the following strategy is generally adopted for hunting:

wherein X (t+1) and X (t) are the positions of the t+1st and t-th iterations of the harris eagle, X _rd (t) and X _bset (t) represents the random individual and prey positions, r ₁ 、r ₂ 、r ₃ 、r ₄ Q are all represented by [0,1]Random number, X, in interval _m (t) represents the average position of the population, expressed as follows:

the energy of the prey can change in the escape process, and the energy factor E controls the transition of the global search and the local development stage, and the calculation formula of E is as follows:

wherein T and T are the current iteration number and the maximum iteration number, E ₀ Indicating the initial energy level, at [ -1,1]Values are randomly selected within the interval. When |E|<And 1, carrying out local search by the algorithm, otherwise, carrying out global optimization. In the local development stage, after the Harris eagle determines an attack target, the game is started to enclose the prey, and the algorithm divides the possibility of enclosing into four strategies to simulate HarrisAttack behaviour of the eagle. The choice of the enclosing strategy is mainly based on the relative magnitude of the energy factor E, and the parameter alpha represents the probability of the prey being killed and is 0,1]Values within the interval.

1) When |E| is not less than 0.5 and alpha is not less than 0.5, the harris eagle adopts a soft enclosing attack strategy to kill the hunting, and the hunting escape energy is sufficient at the moment, but the hunting escape energy is captured due to lack of escape opportunities, and the position updating formula is as follows:

X(t+1)＝X _best (t)-X(t)-E PX _best (t)-X(t)|

in the formula, PX _r (t) -X (t) is the distance of the prey from the current harris eagle, p=2 (1-r) ₅ ) Indicating the ability of the prey to escape r ₅ Is [0,1]Random numbers within the interval.

2) When the E is less than 0.5 and the alpha is more than or equal to 0.5, the harris eagle adopts a hard surrounding attack strategy to catch up the hunting, and the hunting is caught due to the lack of energy, and the position updating formula is as follows:

X(t+1)＝X _best (t)-E|PX _best (t)-X(t)|

3) When the I E I is more than or equal to 0.5 and the alpha is less than 0.5, the harris eagle adopts a progressive rapid diving soft enclosing attack strategy to prey the prey, the prey is in a state with sufficient energy and has larger probability of successfully escaping, for the harris eagle transformation strategy to prey the prey, soft enclosing attack is carried out before formal prey, and the position updating formula is as follows:

Y ₁ ＝X _best (t)-E|PX _best (t)-X(t)|

Y ₂ ＝Y ₁ +Q×LF(d)

Wherein f is an objective function, Q is a d-dimensional random variable, and LF is a Levy flight function.

4) When |E| <0.5 and alpha <0.5, the harris eagle adopts a progressive rapid diving hard surrounding attack strategy to prey on the hunting object, and the hunting object has a larger chance of getting out of the body but lacks energy, so the harris eagle firstly carries out hard surrounding attack to shorten the distance between the harris eagle and the hunting object to achieve the purpose of prey, and the position updating formula is as follows:

Y ₁ ＝X _best (t)-E|PX _best (t)-X _m (t)|

Y ₂ ＝Y ₁ +Q×LF(d)

(2) Improved harris eagle optimization algorithm

In order to avoid the problems of local extremum, premature convergence and the like of an algorithm, firstly, the population diversity is increased by adopting chaotic initialization, so that the initial Harris eagle population is uniformly distributed in a solution space, the population is initialized by adopting Logistic chaotic mapping, and the Logistic chaotic mapping formula is as follows:

Z _n+1 ＝Z _n ·μ·(1-Z _n )

wherein mu is a Logistic control parameter and takes a value in a [0,4] interval; z is a random number between [0,1 ]; mapping the generated chaotic sequence into a new solution space to obtain a position updating formula of the harris eagle, wherein the position updating formula comprises the following steps:

X(t+1)＝X _l +(X _u -X _l )·Z _n+1

wherein X is _u 、X _l Respectively the upper and lower boundaries of the solution space.

Second, in standard HHO algorithms, energy E is the primary regulator in the transitional phase, generally exhibiting a linearly decreasing trend, so that the use of a random shrinkage exponential function is more suitable for expressing energy changes during prey escape in order to more accurately simulate interaction of prey. Therefore, an improved energy linear decreasing adjustment method is proposed, a random shrinkage exponential function is combined into the change process of the hunting energy, and the corresponding energy equation after improvement is as follows:

Finally, in order to improve the local optimizing capability of the algorithm, an adaptive weight factor is introduced, and the hunting behavior of the harris eagle in four enclosing strategies is aimed at, so that the hunting updates the position with smaller adaptive weight, and the corresponding formula is as follows:

in the formula, T is the maximum iteration number, and T is the current iteration number.

Specifically, the step S5: and optimizing the super-parameters of the LSTM by IHHO, and realizing fault identification of the output capacitor of the photovoltaic inverter. The recurrent neural network (Recurrent Neural Network, RNN) can better handle time series problems, but cannot solve long-term dependency problems, i.e. as the length of the input sequence increases, the model cannot use earlier data information in the sequence. The LSTM neural network replaces the neurons in the RNN hidden layer with the memory units with long-term memory effect, so that the long-term dependence problem can be effectively solved. The LSTM is very suitable for processing data highly related to a time sequence, has excellent effects on identification and prediction of complex time sequence data, and can be used for solving the problem of insufficient identification precision of parametric fault diagnosis of the photovoltaic grid-connected inverter. In RNN, because the network layer updates information without limit, the information becomes chaotic and disappears easily, resulting in gradient vanishing problem, while LSTM network adds forgetting unit and memory unit in hidden layer, discards secondary information when inputting new information, retains important information in long-term memory, these units are called gates in LSTM, the gates in memory unit include forgetting gate, input gate and output gate 3 parts, forgetting gate discards irrelevant information, input gate decides new information stored in unit state, output gate controls output of hidden layer node, these gate control units make LSTM have the ability to update and control information flow in different area blocks. However, the parameter identification effect of the LSTM is affected by key parameter settings such as hidden layer nodes, training times and the like. Therefore, HHO is adopted to optimize LSTM key parameters, and the high-precision identification of the faults of the photovoltaic inverter is realized, and the specific realization steps are as follows:

1) Setting algorithm initial parameters, initializing a population by adopting chaotic mapping, and setting LSTM super-parameter optimizing range;

2) Constructing an IHHO-LSTM photovoltaic inverter fault parameter identification model, and taking the Harris eagle individual fitness value as a judgment standard of prediction precision;

3) Calculating hunting energy E according to a random shrinkage index function formula;

4) When the absolute value of the prey energy E is larger than 1, the algorithm performs global search to generate a global optimal solution, otherwise, the position of the prey is updated according to the self-adaptive weight, and then local search is performed;

5) Updating Harris eagle individuals and global optimal solutions;

6) Step 3, 4, until the maximum iteration times or iteration precision is reached, outputting an optimal solution and an adaptability value, wherein the optimal solution is an optimal network super-parameter of LSTM optimization;

7) And constructing an identification model according to the IHHO optimized optimal super parameters, and outputting an identification result and an error.

The step S6: and (5) evaluating the model diagnosis effect. Using mean absolute percentage error (M _APE ) Determining coefficient (R) ² ) Evaluating the model identification effect, M _APE Smaller values indicate better prediction effect, R ² The value range is [0,1 ]]，R ² The closer to 1, the higher the identification accuracy is. The specific calculation expression is as follows:

Wherein y is _i Representing the actual value, y' _i The predicted value is represented by a value of the prediction,

for the average value, n represents the number of predicted samples.

The invention has the beneficial effects that: aiming at the problems of difficult extraction of characteristic quantity, low diagnosis accuracy and the like of the parametric faults of the three-phase photovoltaic grid-connected inverter, the invention provides a photovoltaic grid-connected inverter fault diagnosis method based on a CEEMDAN-SE-IHHO-LSTM model, and the effectiveness of the method for carrying out the parametric fault diagnosis on the DC bus capacitor in the photovoltaic grid-connected inverter is verified through the analysis of an example. The CEEMDAN-SE decomposition reconstruction method is superior to the CEEMDAN, EMD and other single signal decomposition methods, a corresponding accurate diagnosis model can be established according to each trend, the number of unit diagnosis models is reduced, fault characteristic quantity is effectively extracted, and model training efficiency and diagnosis precision are improved; the IHHO is adopted to optimize the LSTM parameters, so that compared with the HHO algorithm, the convergence speed is faster, and the convergence precision is higher; experimental results show that the combination method can effectively improve the accuracy of the parameter fault diagnosis of the direct current bus capacitor in the photovoltaic grid-connected inverter.

Drawings

FIG. 1 is a two-stage three-phase photovoltaic grid-connected inverter topology of the present invention;

FIG. 2 is a flow chart of an improved Harris eagle optimization algorithm of the present invention;

FIG. 3 is a block diagram of an LSTM neural unit of the present invention;

FIG. 4 is a flow chart of fault diagnosis based on CEEMDAN-SE-IHHO-LSTM model of the present invention;

FIG. 5 is a waveform diagram showing the normal capacitance value and the decrease of the capacitance value to 75% of the initial value according to the present invention;

FIG. 6 is an exploded view of a voltage waveform CEEMDAN with fault signals according to the present invention;

FIG. 7 shows SE values for the respective subsequences of the invention after CEEMDAN decomposition;

FIG. 8 is a sequence waveform after reconstruction using SE classification in accordance with the present invention;

FIG. 9 is a fitness curve of the present invention when optimizing LSTM parameters using IHHO;

FIG. 10 shows the result of the test set parameter fault recognition according to the comparison method of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. All other embodiments, which can be made by those skilled in the art without making any inventive effort, based on the embodiments of the present invention are within the scope of the present invention, and are specifically described below in connection with the embodiments.

The steps of the present invention will be described in detail with reference to a CEEMDAN-SE-IHHO-LSTM based fault diagnosis model shown in FIG. 4:

S1: and (5) analyzing the parametric faults of the photovoltaic grid-connected inverter. According to the illustration of fig. 1, the photovoltaic inverter generally adopts a two-stage structure, the front stage adopts a DC-DC power conversion circuit, the rear stage adopts a three-phase full-bridge inverter circuit, and a double closed-loop control strategy of a direct-current voltage outer loop and a network side current inner loop is adopted to realize direct-current grid connection. Therein, L, T, D, C _dc The three-phase bridge type inverter circuit is composed of six insulated gate bipolar transistors with anti-parallel diodes, and grid-connected control is realized. In a double closed-loop control strategy of the photovoltaic grid-connected inverter, the voltage of the direct-current bus is kept stable by the voltage outer loop control; the inner ring controls the output current, and adjusts the output grid current to be in phase with the grid voltage in the same frequency, so that the unit power factor grid connection is realized. Currently, an aluminum electrolytic capacitor with larger capacity is mostly adopted as a direct current bus capacitor C of a photovoltaic grid-connected inverter _dc Therefore, a large amount of higher harmonic signals generated by the inverter in a high-frequency working mode are absorbed, the effects of stabilizing the voltage of the side of the direct-current bus and storing energy after boosting the inverter are achieved, and the purposes of stably conveying and eliminating harmonic waves to a power grid are achieved. The components of the photovoltaic inverter usually operate in a high-frequency mode, so that the Equivalent Series Resistance (ESR) of the capacitor is increased, the capacitance value is reduced, the aging of the DC bus capacitor is accelerated, the performance of the DC bus capacitor is reduced, and finally, the parametric fault is caused, and even the inverter is damaged or the system is broken. The standard for failure judgment is generally to increase the electrolytic capacitor ESR to 2 to 3 times the initial value or to decrease the capacitance value to 80% of the initial value. However, the direct current bus capacitor ESR value of the photovoltaic inverter system is small and is easily influenced by the working condition of the circuit, and the detection precision is not high. Therefore, the invention reduces the capacitance value to 80% of the initial value as the criterion of the parametric fault of the direct current bus capacitor of the photovoltaic grid-connected inverter.

S2: and acquiring voltage signal data of the direct current bus with faults, and decomposing the voltage signal data into a plurality of eigen model functions with different frequencies through CEEMDAN. C (C) _dc The photovoltaic grid-connected inverter circuit states under different aging degrees are less in difference, so that the method has important significance in extracting the characteristics under the fault state. The CEEMDAN can effectively avoid modal aliasing generated when processing parameter fault signals, and can better process and analyze complex data. CEEMDAN is able to decompose raw data into IMFs that contain information of the raw data in different frequency ranges. It introduces an extra signal-to-noise ratio for white noise to control the noise level during each decomposition. The IMF can be completely reconstructed into the original data, almost no noise exists, compared with Empirical Mode Decomposition (EMD), ensemble Empirical Mode Decomposition (EEMD) and Complementary Ensemble Empirical Mode Decomposition (CEEMD), the IMF has higher decomposition rate, and the adaptive noise can be added to eliminate the modal aliasing, and the CEEMDAN is specifically decomposed as follows:

1) Adding Gaussian white noise with the mean value of 0K times to a sequence x (t) to be decomposed to construct a sequence x to be decomposed for K times _i (t)(i＝1,2,3,…,K)。

x _i (t)＝x(t)+εδ _i (t)

Wherein: epsilon is a Gaussian white noise weight coefficient; delta _i (t) is the white noise sequence added the i-th time.

2) For x as above _i (t) performing EMD decomposition to obtain a first IMF and a residual signal r ₁ (t)。

r ₁ (t)＝x(t)-IMF ₁ (t)

3) And adding specific noise to the residual signal in the j-th stage obtained after decomposition, and continuing EMD decomposition.

r _j (t)＝r _j-1 (t)-IMF _j (t)

Repeating the step 3), if the residual signal r obtained by the nth decomposition _n And (t) is a monotone signal, ending the decomposition, otherwise continuing to execute the decomposition.

Specifically, the step S3: and calculating SE values of the subsequences, and reconstructing each component after CEEMDAN decomposition into high-frequency, low-frequency and trend sequences.

The SE can evaluate the complexity of the time series data, and the calculation of the SE does not depend on the length of the data and has better consistency compared with the approximate entropy. The specific calculation process is as follows:

1) The time series X is constructed as an m-dimensional vector:

X(i)＝{x(i),x(i+1),...,x(i+m-1)}

where i=1, 2,..

2) Defining the distance between X (i) and X (j) as d [ X (i), X (j) ] (i not equal to j), wherein the distance is the one with the largest difference value in the corresponding elements:

3) Given a threshold r (r > 0), the number of d [ X (i), X (j) ] < r and the ratio to the total vector N-m:

4) All results from the above formula are averaged:

5) When the dimension m+1 is repeated, the sample entropy of the sequence is theoretically:

6) In practice, however, N cannot be infinity, but is a finite value, and then the estimated value of the sample entropy is:

specifically, the step S4: the standard harris eagle optimization algorithm is improved. The standard Harris eagle optimization algorithm is a novel intelligent optimization algorithm for simulating the predation behavior of the Harris eagle, which is proposed by Heidari et al in 2019, and mainly comprises two stages: the global search stage and the local development stage, wherein the harris hawk inhabitation position is randomly determined in the global search stage, and the following strategy is generally adopted for hunting:

wherein X (t+1) and X (t) are the positions of the t+1st and t-th iterations of the harris eagle, X _rd (t) and X _bset (t) represents the random individual and prey positions, r ₁ 、r ₂ 、r ₃ 、r ₄ Q are all represented by [0,1]Random number, X, in interval _m (t) represents the average position of the population, expressed as follows:

the energy of the prey can change in the escape process, and the energy factor E controls the transition of the global search and the local development stage, and the calculation formula of E is as follows:

wherein T and T are the current iteration number and the maximum iteration number, E ₀ Indicating the initial energy level, at [ -1,1]Values are randomly selected within the interval. When |E|<And 1, carrying out local search by the algorithm, otherwise, carrying out global optimization. In the local development stage, after the harris eagle determines an attack target, the game is started to carry out the enclosing attack on the prey, and the algorithm divides the possibility of the enclosing attack into four strategies to simulate the attack behavior of the harris eagle. The choice of the enclosing strategy is mainly based on the relative magnitude of the energy factor E, and the parameter alpha represents the probability of the prey being killed and is 0,1]Values within the interval.

1) When |E| is not less than 0.5 and alpha is not less than 0.5, the harris eagle adopts a soft enclosing attack strategy to kill the hunting, and the hunting escape energy is sufficient at the moment, but the hunting escape energy is captured due to lack of escape opportunities, and the position updating formula is as follows:

X(t+1)＝X _best (t)-X(t)-E PX _best (t)-X(t)|

in the formula, PX _r (t) -X (t) is the distance of the prey from the current harris eagle, p=2 (1-r) ₅ ) Indicating the ability of the prey to escape r ₅ Is [0,1]Random numbers within the interval.

2) When the E is less than 0.5 and the alpha is more than or equal to 0.5, the harris eagle adopts a hard surrounding attack strategy to catch up the hunting, and the hunting is caught due to the lack of energy, and the position updating formula is as follows:

X(t+1)＝X _best (t)-E|PX _best (t)-X(t)|

3) When the I E I is more than or equal to 0.5 and the alpha is less than 0.5, the harris eagle adopts a progressive rapid diving soft enclosing attack strategy to prey the prey, the prey is in a state with sufficient energy and has larger probability of successfully escaping, for the harris eagle transformation strategy to prey the prey, soft enclosing attack is carried out before formal prey, and the position updating formula is as follows:

Y ₁ ＝X _best (t)-E|PX _best (t)-X(t)|

Y ₂ ＝Y ₁ +Q×LF(d)

Wherein f is an objective function, Q is a d-dimensional random variable, and LF is a Levy flight function.

4) When |E| <0.5 and alpha <0.5, the harris eagle adopts a progressive rapid diving hard surrounding attack strategy to prey on the hunting object, and the hunting object has a larger chance of getting out of the body but lacks energy, so the harris eagle firstly carries out hard surrounding attack to shorten the distance between the harris eagle and the hunting object to achieve the purpose of prey, and the position updating formula is as follows:

Y ₁ ＝X _best (t)-E|PX _best (t)-X _m (t)

Y ₂ ＝Y ₁ +Q×LF(d)

(2) Improved harris eagle optimization algorithm

In order to avoid the problems of local extremum, premature convergence and the like of an algorithm, firstly, the population diversity is increased by adopting chaotic initialization, so that the initial Harris eagle population is uniformly distributed in a solution space, the population is initialized by adopting Logistic chaotic mapping, and the Logistic chaotic mapping formula is as follows:

Z _n+1 ＝Z _n ·μ·(1-Z _n )

wherein mu is a Logistic control parameter and takes a value in a [0,4] interval; z is a random number between [0,1 ]; mapping the generated chaotic sequence into a new solution space to obtain a position updating formula of the harris eagle, wherein the position updating formula comprises the following steps:

X(t+1)＝X _l +(X _u -X _l )·Z _n+1

wherein X is _u 、X _l Respectively the upper and lower boundaries of the solution space.

Second, in standard HHO algorithms, energy E is the primary regulator in the transitional phase, generally exhibiting a linearly decreasing trend, so that the use of a random shrinkage exponential function is more suitable for expressing energy changes during prey escape in order to more accurately simulate interaction of prey. Therefore, an improved energy linear decreasing adjustment method is proposed, a random shrinkage exponential function is combined into the change process of the hunting energy, and the corresponding energy equation after improvement is as follows:

Finally, in order to improve the local optimizing capability of the algorithm, an adaptive weight factor is introduced, and the hunting behavior of the harris eagle in four enclosing strategies is aimed at, so that the hunting updates the position with smaller adaptive weight, and the corresponding formula is as follows:

in the formula, T is the maximum iteration number, and T is the current iteration number.

Fig. 2 shows the flow of the improved IHHO algorithm.

Specifically, the step S5: and optimizing the super-parameters of the LSTM by IHHO, and realizing fault identification of the output capacitor of the photovoltaic inverter. The recurrent neural network (RecurrentNeuralNetwork, RNN) can better handle time series problems, but cannot solve long-term dependence problems, i.e. as the length of the input sequence increases, the model cannot use earlier data information in the sequence. The LSTM neural network replaces the neurons in the RNN hidden layer with the memory units with long-term memory effect, so that the problem of long-term dependence can be effectively solved, and the following formulas are established according to the information flow direction as shown in the LSTM unit structure in the figure 3:

f _t ＝σ(W _f x _t +U _f h _t-1 +b _f )

i _t ＝σ(W _i x _i +U _i h _i-1 +b _i )

o _t ＝σ(W _o x _t +U _o h _t-1 +b _o )

h _t ＝o _t ⊙tanh(c _t )

wherein: f (f) _t 、i _t 、o _t For the state quantity of three gates at time t, c _t Cell status at time t; tanh is the activation function; b _f 、b _i 、b _o 、b _c Bias terms for three gates and cell states; sigma is a sigmoid activation function; w (W) _f 、W _i 、W _o 、W _c Three gates and a weight matrix of cell states; h is a _t The hidden output at the moment t; x is x _t Input at time t; the symbol ". Ii represents the multiplication of matrix elements.

The LSTM is very suitable for processing data highly related to a time sequence, has excellent effects on identification and prediction of complex time sequence data, and can be used for solving the problem of insufficient identification precision of parametric fault diagnosis of the photovoltaic grid-connected inverter. In RNN, because the network layer updates information without limit, the information becomes chaotic and disappears easily, resulting in gradient vanishing problem, while LSTM network adds forgetting unit and memory unit in hidden layer, discards secondary information when inputting new information, retains important information in long-term memory, these units are called gates in LSTM, the gates in memory unit include forgetting gate, input gate and output gate 3 parts, forgetting gate discards irrelevant information, input gate decides new information stored in unit state, output gate controls output of hidden layer node, these gate control units make LSTM have the ability to update and control information flow in different area blocks. However, the parameter identification effect of the LSTM is affected by key parameter settings such as the hidden layer number, the hidden layer neuron number, and the batch processing size. Therefore, HHO is adopted to optimize LSTM key parameters, and the high-precision identification of the faults of the photovoltaic inverter is realized, and the specific realization steps are as follows:

1) Setting algorithm initial parameters, initializing a population by adopting chaotic mapping, and setting LSTM super-parameter optimizing range;

2) Constructing an IHHO-LSTM photovoltaic inverter fault parameter identification model, and taking the Harris eagle individual fitness value as a judgment standard of prediction precision;

3) Calculating hunting energy E according to a random shrinkage index function formula;

4) When the absolute value of the prey energy E is larger than 1, the algorithm performs global search to generate a global optimal solution, otherwise, the position of the prey is updated according to the self-adaptive weight, and then local search is performed;

5) Updating Harris eagle individuals and global optimal solutions;

6) Performing the 3 rd and 4) loops until the maximum iteration times or iteration precision are reached, outputting an optimal solution and an adaptability value, wherein the optimal solution is an optimal network super-parameter of LSTM optimization;

7) And constructing an identification model according to the IHHO optimized optimal super parameters, and outputting an identification result and an error.

The step S6: and (5) evaluating the model diagnosis effect. Using mean absolute percentage error (M _APE ) Determining coefficient (R) ² ) Evaluating the model identification effect, M _APE Smaller values indicate better prediction effect, R ² The value range is [0,1 ]]，R ² The closer to 1, the higher the identification accuracy is. The specific calculation expression is as follows:

Wherein y is _i Representing the actual value, y' _i The predicted value is represented by a value of the prediction,

for the average value, n represents the number of predicted samples.

The validity of the present invention is verified as follows:

according to the power topological diagram of the photovoltaic inverter shown in the attached drawing 1, a simulation model of the photovoltaic inverter is built in Matlab/Simulink, the open-circuit voltage is set to 363V, the maximum power tracking voltage of a photovoltaic panel is 300V, the capacitance value of a direct-current bus is 3300 mu F, the switching frequency is 50kHz, the filter inductance is 500 mu H, the filter capacitance is 100F, and the rated power is 100kW. When the photovoltaic inverter works, the capacitance value is set to be changed within the range of 0-50% to simulate the situation that the capacitance performance is degraded and fails under different degrees, wherein the situation is regarded as normal when the capacitance value is changed within 20%; and acquiring 1000 groups of three-phase line voltage signals of the photovoltaic grid-connected inverter through simulation. In U shape _ab For example, when the dc bus capacitor is at a normal value, the voltage waveform is shown in fig. 5 (a), and fig. 5 (b) is a voltage waveform when the dc bus capacitor is in an aging stage, and it can be seen from the graph that the voltage waveform of the dc bus capacitor under the aging degree is similar to the voltage waveform under the normal condition, and the aging fault condition cannot be intuitively seen, so that it is important to accurately separate the characteristic representing the parametric fault of the dc bus capacitor from the voltage waveform.

The CEEMDAN is adopted to decompose the voltage waveform of the direct current bus with the aging fault, 8 IMFs and 1 residual component are obtained after decomposition, then SE values of all subsequences are calculated, and as shown in the figure, the SE values of the first 3 IMFs (IMF 1, IMF2 and IMF 3) are far higher than those of other IMFs, the complexity and the instability are higher, the values of the middle 3 IMFs (IMF 4, IMF5 and IMF 6) and the last 3 IMFs (IMF 7, IMF8 and Re) are lower, a certain change trend is achieved, and the complexity and the fluctuation are lower. Thus, these 9 IMFs are combined into new components, respectively: high frequency portion IMF (IMF 1-IMF 3), low frequency sequence IMF (IMF 4-IMF 6), and trend sequence IMF3 (IMF 6, IMF7, re). The reconstructed components are shown in fig. 8, where imf, imf, imf3 represent new high frequency components, low frequency components, and trend components, respectively.

Based on the above, IHHO is adopted to optimize the super parameters of LSTM, including the batch size of LSTM network, the number of hidden layer neurons, the training times and the learning rate. The IHHO algorithm maximum iteration number is set to be 200, and the population number is set to be 20. The batch size, the hidden layer number, the hidden layer neuron number and the learning rate range are respectively [1,64], [1,10], [1,100] and [0.001,0.01].

To verify the effectiveness of the fault diagnosis method, the following control groups were set for comparative analysis:

method 1: performing fault parameter diagnosis by adopting an EMD-BP combined model, decomposing a direct current bus voltage signal containing faults by using EMD, respectively establishing a BP diagnosis model for each decomposed IMF by adopting a BP neural network to perform fault diagnosis, and then reconstructing all diagnosis results to obtain a final diagnosis result, wherein the number of BP input neurons is set to be 10, the number of output neurons is set to be 1, the number of hidden layer neurons is set to be 1, and the number of hidden layer neurons is set to be 32;

method 2: performing fault parametric diagnosis by adopting a CEEMDAN-LSTM combined model, performing CEEMDAN decomposition on voltage signals containing faults, then respectively establishing an LSTM diagnosis model for each decomposed IMF by adopting an LSTM neural network to perform fault diagnosis, reconstructing all diagnosis results to obtain a final diagnosis result, setting the learning interest rate of the LSTM model to be 0.001, setting the number of hidden layer to be 1, setting the number of hidden layer neurons to be 50, and setting the batch processing size to be 32;

method 3: performing parametric diagnosis on faults by adopting a CEEMDAN-SE-LSTM combination model, decomposing voltage waveforms containing fault signals by adopting CEEMDAN, then calculating SE values of each IMF, classifying and reconstructing each subsequence into three IMF components with different change trends according to the SE values, respectively establishing an LSTM diagnosis model for performing fault diagnosis by adopting an LSTM neural network for each reconstructed IMF, reconstructing all diagnosis results to obtain final diagnosis results, setting the learning interest rate of the LSTM model to be 0.001, the number of hidden layer to be 1, the number of hidden layer neurons to be 50 and the batch processing size to be 32;

Method 4: the method comprises the steps of performing parameter diagnosis on faults by using a CEEMDAN-SE-HHO-LSTM combination model, decomposing voltage waveforms containing fault signals by using the CEEMDAN, then calculating SE values of each IMF, classifying and reconstructing each subsequence into three IMF components with different change trends according to the SE values, respectively establishing an LSTM diagnosis model for each reconstructed IMF by using an HHO-optimized LSTM model to perform fault diagnosis, reconstructing all diagnosis results to obtain final diagnosis results, and performing HHO optimization to obtain optimal parameters of the LSTM, wherein the optimal parameters are set as follows: the batch size is 10, the number of hidden layers is 2, the number of hidden layer neurons is 32 and 64 respectively, and the learning rate is 0.005.

Method 5: the method comprises the steps of performing parameter diagnosis on faults by using a CEEMDAN-SE-IHHO-LSTM combination model, decomposing voltage waveforms containing fault signals by using the CEEMDAN, then calculating SE values of each IMF, classifying and reconstructing each subsequence into three IMF components with different change trends according to the SE values, respectively establishing an LSTM diagnosis model for each reconstructed IMF by using an IHHO-optimized LSTM model to perform fault diagnosis, reconstructing all diagnosis results to obtain final diagnosis results, and performing IHHO optimization to obtain optimal parameters of the LSTM, wherein the optimal parameters are set as follows: the batch size is 20, the number of hidden layers is 2, the number of hidden layer neurons is 50 and 100 respectively, and the learning rate is 0.008.

The LSTM super parameters are optimized by HHO and IHHO, mean square error of training samples is used as a fitness function, adam is used as an optimization algorithm of weight and bias, a comparison curve of the change of the fitness value along with the iteration number is obtained, the diagnosis identification result of the capacitance value under each method is shown in figure 9, and the evaluation index value is shown in table 1.

Table 1 comparison of diagnostic results of various methods

As can be seen from fig. 9, IHHO reaches convergence about 70 times, and HHO converges about 100 times, which is faster in convergence speed and higher in convergence accuracy than before improvement, indicating the effectiveness of the improvement method.

As can be seen from fig. 10 and table 1, method 1 uses EMD for signal decomposition, followed by utilization ofThe BP neural network performs fault diagnosis, the diagnosis effect is worst in all methods, the diagnosis precision is to be improved, through comprehensive comparison, the characteristic extraction method of CEEMDAN-SE is superior to the single signal decomposition method of CEEMDAN, EMD and the like, through SE reconstruction decomposition signals, each subsequence after decomposition is reconstructed according to categories, then LSTM diagnosis model of each reconstruction sequence is established, fault diagnosis can be realized more efficiently, because the reconstructed sequences have obvious characteristic trend corresponding to each trend, corresponding accurate diagnosis model can be established according to each trend, compared with the diagnosis model of each sequence only after signal decomposition, the diagnosis model is established after reconstruction, the number of unit diagnosis models can be reduced, further the model training efficiency and the diagnosis precision are improved, on the basis, the super-parameters of LSTM can be calculated more accurately by adopting IHHO (fast fourier transform) optimization algorithm for LSTM parameters after optimization, compared with the R < R > _MSE The value is reduced by 1.38 percent, M _APE The value is reduced by 0.58 percent, R ² The method improves the fault diagnosis effect of the direct current bus capacitor by 0.95%, and the method 5 has the best fault diagnosis effect on the direct current bus capacitor, and fully verifies the validity of the method for carrying out parameter fault diagnosis on the direct current bus capacitor in the photovoltaic grid-connected inverter.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims (8)

1. A photovoltaic grid-connected inverter fault diagnosis method based on CEEMDAN-SE-IHHO-LSTM model comprises the following steps:

s1: analyzing the parametric faults of the photovoltaic grid-connected inverter;

s2: obtaining voltage signal data of a direct current bus with faults, and decomposing the voltage signal data into a plurality of eigen model functions with different frequencies through CEEMDAN;

S3: calculating SE values of all subsequences, and reconstructing all components after CEEMDAN decomposition into high-frequency, low-frequency and trend sequences;

s4: improving a standard Harris eagle optimization algorithm;

s5: optimizing the super-parameters of LSTM by IHHO, and realizing fault identification of the output capacitor of the photovoltaic inverter;

s6: and (5) evaluating the model diagnosis effect.

2. The method for diagnosing faults of a photovoltaic grid-connected inverter based on the CEEMDAN-SE-IHHO-LSTM model as claimed in claim 1, wherein in step S1, the parameters of the photovoltaic grid-connected inverter are analyzed; the photovoltaic inverter generally adopts a two-stage structure, the front stage adopts a DC-DC power conversion circuit, the rear stage adopts a three-phase full-bridge inverter circuit, and a double closed-loop control strategy of a direct-current voltage outer ring and a network side current inner ring is adopted to realize direct-current grid connection; therein, L, T, D, C _dc The three-phase bridge type inverter circuit is formed by six insulated gate bipolar transistors with anti-parallel diodes, and grid-connected control is realized; in a double closed-loop control strategy of the photovoltaic grid-connected inverter, the voltage of the direct-current bus is kept stable by the voltage outer loop control; the inner ring controls the output current, and adjusts the output grid current to be in phase with the grid voltage in the same frequency, so that the unit power factor grid connection is realized; currently, an aluminum electrolytic capacitor with larger capacity is mostly adopted as a direct current bus capacitor C of a photovoltaic grid-connected inverter _dc Thereby absorbing a large amount of higher harmonic signals generated by the inverter in a high-frequency working mode, playing roles in stabilizing the voltage of the side of the direct current bus and storing energy after boosting the inverter, and achieving the purposes of stably conveying and eliminating harmonic waves to a power grid; the components of a photovoltaic inverter are typically in a high frequency modeThe capacitor is subjected to lower operation, so that the Equivalent Series Resistance (ESR) of the capacitor is increased, the capacitance value is reduced, the aging of the DC bus capacitor is accelerated, the performance of the DC bus capacitor is reduced, and finally, the parametric fault is caused, and even the inverter is damaged or the system is broken; the ESR of the electrolytic capacitor is increased to 2-3 times of the initial value or the capacitance value is reduced to 80% of the initial value, which is used as a standard for failure judgment; however, the direct current bus capacitor ESR value of the photovoltaic inverter system is small and is easily influenced by the working condition of the circuit, and the detection precision is not high; therefore, the invention reduces the capacitance value to 80% of the initial value as the criterion of the parametric fault of the direct current bus capacitor of the photovoltaic grid-connected inverter.

3. The method for diagnosing faults of the photovoltaic grid-connected inverter based on the CEEMDAN-SE-IHHO-LSTM model according to claim 1, wherein in the step S2, the voltage signal data of the direct current bus with faults are obtained and decomposed into a plurality of eigenvalue model functions with different frequencies through the CEEMDAN; c (C) _dc The state difference of the photovoltaic grid-connected inverter circuits under different aging degrees is small, so that the method has important significance for extracting the characteristics under the fault state; the CEEMDAN can effectively avoid modal aliasing generated when processing the parametric fault signal, and can better process and analyze complex data; CEEMDAN is capable of decomposing the raw data into IMFs comprising different frequency ranges, each IMF comprising information of the raw data in a different frequency range; it introduces an extra signal-to-noise ratio for white noise to control the noise level during each decomposition; the IMF can be completely reconstructed into the original data, almost no noise exists, compared with Empirical Mode Decomposition (EMD), ensemble Empirical Mode Decomposition (EEMD) and Complementary Ensemble Empirical Mode Decomposition (CEEMD), the IMF has higher decomposition rate, and the adaptive noise can be added to eliminate the modal aliasing, and the CEEMDAN is specifically decomposed as follows:

1) Adding Gaussian white noise with the mean value of 0K times to a sequence x (t) to be decomposed to construct a sequence x to be decomposed for K times _i (t)(i＝1,2,3,…,K)；

x _i (t)＝x(t)+εδ _i (t)

Wherein: epsilon is a Gaussian white noise weight coefficient; delta _i (t) is the ith added white noise sequence;

2) For x as above _i (t) performing EMD decomposition to obtain a first IMF and a residual signal r ₁ (t)；

r ₁ (t)＝x(t)-IMF ₁ (t)

3) Adding specific noise to the residual signal of the j-th stage obtained after decomposition, and continuing EMD decomposition;

r _j (t)＝r _j-1 (t)-IMF _j (t)

repeating the step 3), if the residual signal r obtained by the nth decomposition _n And (t) is a monotone signal, ending the decomposition, otherwise continuing to execute the decomposition.

4. The method for diagnosing faults of the photovoltaic grid-connected inverter based on the CEEMDAN-SE-IHHO-LSTM model according to claim 1, wherein in the step S3, SE values of all sub-sequences are calculated, and all components after CEEMDAN decomposition are reconstructed into high-frequency, low-frequency and trend sequences; the SE can evaluate the complexity of the time sequence data, and compared with the approximate entropy, the SE calculation does not depend on the length of the data, and has better consistency; the specific calculation process is as follows:

1) The time series X is constructed as an m-dimensional vector:

X(i)＝{x(i),x(i+1),...,x(i+m-1)}

where i=1, 2,..

2) Defining the distance between X (i) and X (j) as d [ X (i), X (j) ] (i not equal to j), wherein the distance is the one with the largest difference value in the corresponding elements:

3) Given a threshold r (r > 0), the number of d [ X (i), X (j) ] < r and the ratio to the total vector N-m:

4) All results from the above formula are averaged:

5) When the dimension m+1 is repeated, the sample entropy of the sequence is theoretically:

6) In practice, however, N cannot be infinity, but is a finite value, and then the estimated value of the sample entropy is:

5. the method for diagnosing faults of the photovoltaic grid-connected inverter based on the CEEMDAN-SE-IHHO-LSTM model according to claim 1, wherein in the step S4, a standard Harris eagle optimization algorithm is improved; the standard Harris eagle optimization algorithm is a novel intelligent optimization algorithm for simulating the predation behavior of the Harris eagle, which is proposed by Heidari et al in 2019, and mainly comprises two stages: the global search stage and the local development stage, wherein the harris hawk inhabitation position is randomly determined in the global search stage, and the following strategy is generally adopted for hunting:

wherein X (t+1) and X (t) are the positions of the t+1st and t-th iterations of the harris eagle, X _rd (t) and X _bset (t) represents the random individual and prey positions, r ₁ R2, r3, r4, q are all represented as [0,1]]Random numbers within the interval, xm (t), represent the average position of the population, expressed as follows:

the energy of the prey can change in the escape process, and the energy factor E controls the transition of the global search and the local development stage, and the calculation formula of E is as follows:

wherein T and T are the current iteration number and the maximum iteration number respectively, E0 represents the initial energy, and values are randomly selected in the interval of [ -1,1 ]; when the I E I is <1, the algorithm performs local search, otherwise, global optimization is performed; in the local development stage, after the harris eagle determines an attack target, starting to carry out surrounding attack on a prey, and dividing the possibility of surrounding attack into four strategies by an algorithm to simulate the attack behavior of the harris eagle; the selection of the enclosing and striking strategy mainly refers to the relative size of an energy factor E, and the parameter alpha represents the probability of the prey being killed and is a value in the interval of [0,1 ];

1) When |E| is not less than 0.5 and alpha is not less than 0.5, the harris eagle adopts a soft enclosing attack strategy to kill the hunting, and the hunting escape energy is sufficient at the moment, but the hunting escape energy is captured due to lack of escape opportunities, and the position updating formula is as follows:

X(t+1)＝X _best (t)-X(t)-EPX _best (t)-X(t)

in the formula, PX _r (t) -X (t) is the distance of the prey from the current harris eagle, p=2 (1-r) ₅ ) Representing preyAbility to escape, r5 is [0,1 ]]Random numbers within the interval;

2) When the E is less than 0.5 and the alpha is more than or equal to 0.5, the harris eagle adopts a hard surrounding attack strategy to catch up the hunting, and the hunting is caught due to the lack of energy, and the position updating formula is as follows:

X(t+1)＝X _best (t)-EPX _best (t)-X(t)

3) When the I E I is more than or equal to 0.5 and the alpha is less than 0.5, the harris eagle adopts a progressive rapid diving soft enclosing attack strategy to prey the prey, the prey is in a state with sufficient energy and has larger probability of successfully escaping, for the harris eagle transformation strategy to prey the prey, soft enclosing attack is carried out before formal prey, and the position updating formula is as follows:

Y ₁ ＝X _best (t)-EPX _best (t)-X(t)

Y ₂ ＝Y ₁ +Q×LF(d)

wherein f is an objective function, Q is a d-dimensional random variable, and LF is a Levy flight function;

4) When |E| <0.5 and alpha <0.5, the harris eagle adopts a progressive rapid diving hard surrounding attack strategy to prey on the hunting object, and the hunting object has a larger chance of getting out of the body but lacks energy, so the harris eagle firstly carries out hard surrounding attack to shorten the distance between the harris eagle and the hunting object to achieve the purpose of prey, and the position updating formula is as follows:

Y ₁ ＝X _best (t)-EPX _best (t)-X _m (t)

Y ₂ ＝Y ₁ +Q×LF(d)。

6. The method for diagnosing faults of a photovoltaic grid-connected inverter based on the CEEMDAN-SE-IHHO-LSTM model according to claim 1 and 5, wherein in the step S4, further, a standard Harish eagle optimization algorithm is improved, in order to avoid the problems of local extremum, premature convergence and the like of the algorithm, firstly, chaos initialization is adopted to increase population diversity, so that initial Harish eagle population is uniformly distributed in a solution space, logistic chaotic mapping is adopted to initialize the population, and a Logistic chaotic mapping formula is as follows:

Z _n+1 ＝Z _n ·μ·(1-Z _n )

wherein mu is a Logistic control parameter and takes a value in a [0,4] interval; z is a random number between [0,1 ]; mapping the generated chaotic sequence into a new solution space to obtain a position updating formula of the harris eagle, wherein the position updating formula comprises the following steps:

X(t+1)＝X _l +(X _u -X _l )·Z _n+1

wherein X is _u 、X _l Respectively the upper and lower boundaries of the solution space;

secondly, in the standard HHO algorithm, energy E is the main regulatory factor in the transition phase, generally presenting a linearly decreasing trend, so to more accurately simulate interaction of prey, the use of a random shrinkage exponential function is more suitable for expressing energy variation during prey escape; therefore, an improved energy linear decreasing adjustment method is proposed, a random shrinkage exponential function is combined into the change process of the hunting energy, and the corresponding energy equation after improvement is as follows:

Finally, in order to improve the local optimizing capability of the algorithm, an adaptive weight factor is introduced, and the hunting behavior of the harris eagle in four enclosing strategies is aimed at, so that the hunting updates the position with smaller adaptive weight, and the corresponding formula is as follows:

in the formula, T is the maximum iteration number, and T is the current iteration number.

7. The method for diagnosing faults of the photovoltaic grid-connected inverter based on the CEEMDAN-SE-IHHO-LSTM model as claimed in claim 1, wherein the step S5 is performed by optimizing the super-parameters of the LSTM by IHHO and realizing the fault identification of the output capacitor of the photovoltaic inverter; the cyclic neural network (Recurrent Neural Network, RNN) can better handle time series problems, but cannot solve long-term dependence problems, i.e. as the length of the input sequence increases, the model cannot use earlier data information in the sequence; the LSTM neural network replaces the neurons in the RNN hidden layer with the memory units with long-term memory effect, so that the long-term dependence problem can be effectively solved; the LSTM is very suitable for processing data highly related to a time sequence, has excellent effect on identification and prediction of complex time sequence data, and can be used for solving the problem of insufficient identification precision of the parametric fault diagnosis of the photovoltaic grid-connected inverter; in RNN, because the network layer updates information without limit, the information becomes chaotic and disappears easily, resulting in gradient vanishing problem, while LSTM network adds forgetting unit and memory unit in hidden layer, discards secondary information when inputting new information, retains important information in long-term memory, these units are called gates in LSTM, the gates in memory unit include forgetting gate, input gate and output gate 3 parts, forgetting gate discards irrelevant information, input gate decides new information stored in unit state, output gate controls hidden layer node output, these gate control units make LSTM have the ability to update and control information flow in different area blocks; however, the parameter identification effect of the LSTM is affected by key parameter settings such as hidden layer nodes, training times and the like; therefore, HHO is adopted to optimize LSTM key parameters, and the high-precision identification of the faults of the photovoltaic inverter is realized, and the specific realization steps are as follows:

1) Setting algorithm initial parameters, initializing a population by adopting chaotic mapping, and setting LSTM super-parameter optimizing range;

2) Constructing an IHHO-LSTM photovoltaic inverter fault parameter identification model, and taking the Harris eagle individual fitness value as a judgment standard of prediction precision;

3) Calculating hunting energy E according to a random shrinkage index function formula;

4) When the absolute value of the prey energy E is larger than 1, the algorithm performs global search to generate a global optimal solution, otherwise, the position of the prey is updated according to the self-adaptive weight, and then local search is performed;

5) Updating Harris eagle individuals and global optimal solutions;

6) Step 3, 4, until the maximum iteration times or iteration precision is reached, outputting an optimal solution and an adaptability value, wherein the optimal solution is an optimal network super-parameter of LSTM optimization;

7) And constructing an identification model according to the IHHO optimized optimal super parameters, and outputting an identification result and an error.

8. The method for diagnosing faults of a photovoltaic grid-connected inverter based on the CEEMDAN-SE-IHHO-LSTM model as claimed in claim 1, wherein the model prediction effect is evaluated according to the step S6; using mean absolute percentage error (M _APE ) Determining coefficient (R) ² ) Evaluating the model identification effect, M _APE Smaller values indicate better prediction effect, R ² The value range is [0,1 ]]，R ² The closer to 1, the higher the identification accuracy is; the specific calculation expression is as follows:

wherein y is _i Representing the actual value, y' _i The predicted value is represented by a value of the prediction,

for the average value, n represents the number of predicted samples.

Images