A photovoltaic inverter fault diagnosis system optimization method based on deep learning

The photovoltaic inverter fault diagnosis system based on deep learning solves the problems of time-consuming and labor-intensive methods and fluctuating detection data in traditional methods, and achieves efficient and accurate fault diagnosis, thus ensuring the stability and reliability of the photovoltaic system.

CN118826642BActive Publication Date: 2024-11-22KUNMING AUTOMATION WHOLE SET OF EQUIP BUSINESS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411288353.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-14
Publication Date
2024-11-22
Estimated Expiration
2044-09-14

AI Technical Summary

Technical Problem

Traditional photovoltaic inverter fault diagnosis methods are time-consuming, labor-intensive, and prone to missing problems. Intelligent diagnosis methods suffer from fluctuations in detection data equipment, which affect the accuracy of diagnostic results and consume excessive computational resources.

Method used

A photovoltaic inverter fault diagnosis system based on deep learning is adopted. By periodically collecting operating parameters, constructing a time series set, setting judgment thresholds and sliding windows, and combining deep learning models for secondary judgment, the system selects data with high accuracy for fault prediction.

Benefits of technology

It enables intelligent and timely fault detection of photovoltaic inverters, improves the accuracy and reliability of diagnosis, saves computing resources, and ensures stable system operation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118826642B_ABST
    Figure CN118826642B_ABST
Patent Text Reader

Abstract

The present invention provides a photovoltaic inverter fault diagnosis system optimization method based on deep learning, which relates to the photovoltaic inverter fault diagnosis technology field, and aims to optimize the intelligent diagnosis method of photovoltaic inverters while considering the data impact caused by the fluctuation or abnormality of the detection equipment, including: periodically collecting multiple operating parameters of photovoltaic inverters and forming operating parameter sequences; constructing a time series operating parameter sequence set with a length of N; making a judgment based on the time series operating parameter sequence set to obtain the accuracy of the operating parameters; setting a judgment threshold based on the accuracy of the operating parameters and judging whether the photovoltaic inverter is suspected of having a fault based on the judgment threshold; making a secondary judgment based on the time series operating parameter sequence set and the accuracy of the operating parameters based on deep learning to predict the fault type of the photovoltaic inverter. The present invention has the advantages of more reliable and accurate fault diagnosis and saving computing resources.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic inverter fault diagnosis, and in particular to a photovoltaic inverter fault diagnosis system optimization method based on deep learning. Background Art

[0002] A photovoltaic inverter is a device used to convert direct current generated by solar photovoltaic panels into alternating current. In photovoltaic systems, photovoltaic inverters play a very important role.

[0003] Since the photovoltaic inverter is the core component of the photovoltaic power generation system, its normal operation directly affects the performance and power generation efficiency of the entire photovoltaic system. Traditional fault diagnosis methods rely on manual inspection and maintenance, which is not only time-consuming and labor-intensive, but also easy to miss problems. The intelligent fault diagnosis system can monitor the operating status of the inverter around the clock, detect potential faults in a timely manner, reduce the frequency and intensity of manual inspections, and thus reduce maintenance costs. However, when performing intelligent fault diagnosis, the accuracy of the diagnosis results may be affected by fluctuations or failures in the detection data equipment, and it is easy to have problems with excessive computing consumption.

[0004] Therefore, it is necessary to optimize the intelligent diagnosis method of photovoltaic inverters while considering the data impact caused by fluctuations or anomalies of the detection equipment to achieve more accurate and reliable fault diagnosis. Summary of the invention

[0005] The purpose of the present invention is to provide a photovoltaic inverter fault diagnosis system optimization method based on deep learning, which can optimize the intelligent diagnosis method of the photovoltaic inverter while considering the data impact caused by the fluctuation or abnormality of the detection equipment.

[0006] The present invention is achieved through the following technical solutions:

[0007] A photovoltaic inverter fault diagnosis system optimization method based on deep learning, comprising the following steps:

[0008] Periodically collect various operating parameters of the photovoltaic inverter and form an operating parameter sequence;

[0009] Construct a time series operation parameter sequence set with a length of N;

[0010] Perform a judgment based on the time series operation parameter sequence set to obtain the accuracy of the operation parameters, wherein the accuracy of the operation parameters ranges from [0, 1], and a larger value of the accuracy of the operation parameters represents a more accurate operation parameter;

[0011] A judgment threshold is set according to the accuracy of the operating parameters and based on the judgment threshold, it is judged whether the photovoltaic inverter is suspected of having a fault, if the judgment result is no, no operation is performed, if the judgment result is yes, proceed to the next step;

[0012] Based on deep learning, a secondary judgment is made according to the timing operation parameter sequence set and the accuracy of the operation parameters to predict the fault type of the photovoltaic inverter;

[0013] A method for determining the accuracy of the operating parameters by performing a judgment based on the timing operating parameter sequence set is as follows:

[0014] Setting a detection window, wherein the detection window is used to select M consecutive operation parameter sequences in the timing operation parameter sequence set;

[0015] Starting from the first operating parameter sequence in the timing operating parameter sequence set, a first detection parameter sequence set is obtained by selecting through a detection window;

[0016] Initialize the slide number , set the sliding step size ;

[0017] With the sliding step length Slide the detection window back to select Detect parameter sequence set and update the sliding number ;

[0018] like Go to the next step, otherwise return to the previous step. represents round down;

[0019] Calculate the floating parameters of all detection parameter sequence sets and select the detection parameter sequence set with the smallest floating parameter as the standard detection parameter sequence set;

[0020] The accuracy of the operating parameters is obtained based on the standard detection parameter sequence set.

[0021] Preferably, the method of calculating the floating parameters of all detection parameter sequence sets and selecting the detection parameter sequence set with the smallest floating parameter as the standard detection parameter sequence set is:

[0022] Compute the floating parameters for the set of all detection parameter sequences:

[0023] ;

[0024] ;

[0025] ;

[0026] in, For the detecting said floating parameter of a set of parameter sequences, For the The first The first of the operating parameter sequences operating parameters, For the The detection parameter sequence set The average value of the operating parameters, is the total number of types of operating parameters, Based on The operation of finding the maximum value of the value change is Based on The operation of finding the minimum value of the value change is and are weights respectively;

[0027] Select The minimum value in The detection parameter sequence set corresponding to the minimum value in is used as the standard detection parameter sequence set.

[0028] Preferably, the method for obtaining the accuracy of the operating parameters based on the standard detection parameter sequence is:

[0029] ;

[0030] ;

[0031] in, and represent the exponential function and logarithmic function with the natural logarithm e as the base, respectively. is the intermediate parameter, is the upper limit empirical value of the intermediate parameter, is the lower limit empirical value of the intermediate parameter, For the The minimum value in The accuracy of operating parameters.

[0032] Preferably, the value range of M is: .

[0033] Preferably, the method of setting a judgment threshold according to the accuracy of the operating parameter and judging whether the photovoltaic inverter is suspected of having a fault based on the judgment threshold is:

[0034] Obtaining a standard threshold value for each of the operating parameters in the current operating parameter sequence;

[0035] A judgment threshold is set for each of the operating parameters:

[0036] ;

[0037] ;

[0038] ;

[0039] in, The first the upper limit of the standard threshold value of the operating parameter, and are respectively the first the upper and lower limits of the judgment thresholds of the operating parameters, is the accuracy of the operating parameters, The total number of types of operating parameters;

[0040] It is determined whether each operating parameter in the current operating parameter sequence exceeds the corresponding standard threshold value. If any operating parameter exceeds the corresponding standard threshold value, it is determined that the photovoltaic inverter is suspected of having a fault; otherwise, it is determined that the photovoltaic inverter is not suspected of having a fault.

[0041] Preferably, the method for performing secondary judgment based on the set of timing operation parameter sequences and the accuracy of the operation parameters based on deep learning is to perform secondary judgment through a trained network model, wherein the network model includes an input layer, a feature extraction layer, a hidden layer and an output layer connected in sequence.

[0042] Preferably, the feature extraction layer works as follows:

[0043] Receive input data from the input layer, the input data is:

[0044] ;

[0045] ;

[0046] ;

[0047] , ;

[0048] in, represents the set of timing operation parameter sequences, is the accuracy of the operating parameters, As of the current time of A sequence of operating parameters, Representative time The The average value of the operating parameters, , The total number of types of operating parameters;

[0049] Perform feature extraction:

[0050]

[0051]

[0052] ;

[0053] The output of the feature extraction layer is

[0054] in, Representative characteristic parameters of the operating parameters, is the sign function, Based on The function that finds the maximum value of the change of Based on The function that finds the minimum value of the change of Represents the exponential function with the natural logarithm e as the base.

[0055] Preferably, the expression of the hidden layer is:

[0056] ;

[0057] ;

[0058] in, Represents the hidden layer The output of the layer, For the hidden layer The weight of the layer, For the hidden layer The offset parameters of the layer;

[0059] The output layer adopts the softmax function.

[0060] Preferably, the operating parameters include DC side current, DC side voltage, AC side current, AC side voltage, AC frequency, operating temperature and operating humidity.

[0061] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0062] The present invention realizes intelligent monitoring of the status of the photovoltaic inverter, which helps to timely discover and solve the fault of the inverter, prevent the fault from expanding, ensure the stable operation of the system, and improve the reliability of the photovoltaic system;

[0063] When the present invention performs monitoring, it first filters out data that is determined to be fault-free through simple threshold comparison, and does not require further fault identification, which helps save computing resources;

[0064] The present invention takes into account the influence of the stable state of monitoring equipment such as various sensors on the monitoring results when filtering out data determined to be fault-free and when performing fault identification, thereby improving the reliability and accuracy of the results;

[0065] When considering the reliability of the monitoring equipment, the present invention eliminates the influence of the fluctuation of the data itself as much as possible, selects the minimum fluctuation segment to make reliability judgment, and further improves the accuracy of the judgment result;

[0066] The present invention is reasonably designed and easy to implement, and can also be applied to status monitoring and fault identification of other equipment, thus being easy to promote and implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A schematic flow chart of a photovoltaic inverter fault diagnosis system optimization method based on deep learning provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0069] Example 1

[0070] This embodiment provides a photovoltaic inverter fault diagnosis system optimization method based on deep learning, see Figure 1 , including the following steps:

[0071] Step S1: periodically collecting various operating parameters of the photovoltaic inverter and forming an operating parameter sequence;

[0072] Step S2: construct a time series operation parameter sequence set with a length of N;

[0073] Step S3: Perform a judgment according to the time series operating parameter sequence set to obtain the accuracy of the operating parameters, wherein the accuracy of the operating parameters ranges from [0, 1], and the larger the value of the accuracy of the operating parameters, the more accurate the operating parameters;

[0074] Step S4: setting a judgment threshold according to the accuracy of the operating parameters and judging whether the photovoltaic inverter is suspected of having a fault based on the judgment threshold, if the judgment result is no, no operation is performed, if the judgment result is yes, proceeding to the next step;

[0075] Step S5: Perform a secondary judgment based on the timing operation parameter sequence set and the accuracy of the operation parameters based on deep learning to predict the fault type of the photovoltaic inverter.

[0076] In specific implementation, the operating parameters may include DC side current, DC side voltage, AC side current, AC side voltage, AC frequency, operating temperature and operating humidity. Comprehensive detection of internal electrical status and external environmental status helps to obtain fault information characteristics in all aspects and find the fault more accurately.

[0077] In this embodiment, a method for determining the accuracy of the operating parameters is as follows:

[0078] Setting a detection window, wherein the detection window is used to select M consecutive operation parameter sequences in the timing operation parameter sequence set;

[0079] Starting from the first operating parameter sequence in the timing operating parameter sequence set, a first detection parameter sequence set is obtained by selecting through a detection window;

[0080] Initialize the slide number , set the sliding step size ;

[0081] With the sliding step length Slide the detection window back to select Detect parameter sequence set and update the sliding number ;

[0082] like Go to the next step, otherwise return to the previous step. represents round down;

[0083] Calculate the floating parameters of all detection parameter sequence sets and select the detection parameter sequence set with the smallest floating parameter as the standard detection parameter sequence set;

[0084] The accuracy of the operating parameters is obtained based on the standard detection parameter sequence set.

[0085] Intelligent monitoring of photovoltaic relays can improve the efficiency, stability and sustainability of monitoring. In order to save computing power resources for model identification, a threshold range can usually be set for each operating parameter first. If all operating parameters are within the threshold range, the subsequent fault identification process can be omitted. However, there are some problems in filtering normal data through the threshold range and subsequently identifying faults through the model. For example, the data problem directly detected by the sensor may be caused by the unstable fluctuation of the sensor itself or the data fluctuation of the photovoltaic inverter itself. In other words, the monitoring data of the monitoring equipment cannot guarantee 100% reliability at all times. Because this embodiment tries to obtain the accuracy of the operating parameters of the relevant monitoring equipment, different threshold ranges are set based on the accuracy of the operating parameters when filtering normal data through the threshold range. In the subsequent fault identification through the model, the accuracy of the operating parameters is also included in the characteristic parameters, which improves the accuracy and reliability of intelligent identification of photovoltaic inverter faults and saves computing power resources.

[0086] Example 2

[0087] This embodiment is based on the technical solution of Embodiment 1, and mainly further illustrates the specific implementation method of steps S3-S4.

[0088] Furthermore, the method of calculating the floating parameters of all detection parameter sequence sets and selecting the detection parameter sequence set with the smallest floating parameter as the standard detection parameter sequence set is:

[0089] Compute the floating parameters for the set of all detection parameter sequences:

[0090] ;

[0091] ;

[0092] ;

[0093] in, For the detecting said floating parameter of a set of parameter sequences, For the The first The first of the operating parameter sequences operating parameters, For the The detection parameter sequence set The average value of the operating parameters, is the total number of types of operating parameters, Based on The operation of finding the maximum value of the value change is Based on The operation of finding the minimum value of the value change is and are weights respectively;

[0094] Select The minimum value in The detection parameter sequence set corresponding to the minimum value in is used as the standard detection parameter sequence set.

[0095] As a preferred solution, the method for obtaining the accuracy of the operating parameters based on the standard detection parameter sequence is:

[0096] ;

[0097] ;

[0098] in, and represent the exponential function and logarithmic function with the natural logarithm e as the base, respectively. is the intermediate parameter, is the upper limit empirical value of the intermediate parameter, is the lower limit empirical value of the intermediate parameter, For the The minimum value in The accuracy of operating parameters.

[0099] The larger the value of The smaller, the Reduce, that is to say The lower the value, the less accurate the operating parameters. As a variable parameter, the amount of data used for calculation is considered to affect the calculation results. When the amount of data used is small, In the interval less than 1, the value is large. The value of is larger in the negative range (negative means smaller absolute value), and thus The absolute value of the interval in the negative interval and the absolute value less than 1 is larger. The subtracted value is larger, which increases and , which means that when the amount of data is small, a certain accuracy bonus will be given. This is because actual experiments have found that the fluctuation of the monitoring equipment itself is limited, so when the amount of data is small, try not to reduce the accuracy of the operating parameters.

[0100] Finally, the value range of M is: .

[0101] As a preferred solution of this embodiment, a method for setting a judgment threshold according to the accuracy of the operating parameters and judging whether the photovoltaic inverter is suspected of having a fault based on the judgment threshold is:

[0102] Obtaining a standard threshold value for each of the operating parameters in the current operating parameter sequence;

[0103] A judgment threshold is set for each of the operating parameters:

[0104] ;

[0105] ;

[0106] ;

[0107] in, The first one in the current operating parameter sequence the upper limit of the standard threshold value of the operating parameter, and are respectively the first the upper and lower limits of the judgment thresholds of the operating parameters, is the accuracy of the operating parameters, The total number of types of operating parameters;

[0108] It is determined whether each operating parameter in the current operating parameter sequence exceeds the corresponding standard threshold value. If any operating parameter exceeds the corresponding standard threshold value, it is determined that the photovoltaic inverter is suspected of having a fault; otherwise, it is determined that the photovoltaic inverter is not suspected of having a fault.

[0109] The purpose of obtaining the accuracy of the operating parameters in this embodiment is to capture the values ​​of the monitored operating parameters in the stage where the photovoltaic inverter is in a relatively stable state as much as possible. Therefore, a small section is selected through a small window frame in the acquired operating parameter sequence set to make a judgment based on the degree of data fluctuation. Selecting the data with the smallest fluctuation as the final judgment standard can eliminate the influence of the photovoltaic inverter's own fluctuation to a certain extent. For the case of low accuracy of the operating parameters, the range of the judgment threshold will be narrowed, and stricter judgment standards will be implemented to avoid missing data that may have problems due to inaccurate parameter acquisition. On the other hand, if the fluctuation is still caused by the photovoltaic inverter, strict judgment will be performed due to the narrowing of the judgment threshold range, and suspected problem data will not be missed. Therefore, this method has higher reliability. The accuracy of the operating parameters is standardized to a numerical range of 0-1, which is convenient for optimizing the judgment threshold.

[0110] Example 3

[0111] This embodiment is based on the technical solution of Embodiment 1, and mainly further illustrates the specific implementation method of step S5.

[0112] In this embodiment, the method for performing secondary judgment based on the timing operation parameter sequence set and the accuracy of the operation parameters based on deep learning is to perform secondary judgment through a trained network model, and the network model includes an input layer, a feature extraction layer, a hidden layer and an output layer connected in sequence.

[0113] As a preferred solution, the feature extraction layer works as follows:

[0114] Receive input data from the input layer, the input data is:

[0115] ;

[0116] ;

[0117] ;

[0118] , ;

[0119] in, represents the set of timing operation parameter sequences, is the accuracy of the operating parameters, As of the current time of A sequence of operating parameters, Representative time The The average value of the operating parameters, , The total number of types of operating parameters;

[0120] Perform feature extraction:

[0121]

[0122]

[0123] ;

[0124] The output of the feature extraction layer is

[0125] in, Representative characteristic parameters of the operating parameters, is the sign function, Based on The function that finds the maximum value of the change of Based on The function that finds the minimum value of the change of Represents the exponential function with the natural logarithm e as the base.

[0126] Furthermore, the expression of the hidden layer is:

[0127] ;f

[0128] ;

[0129] in, Represents the hidden layer The output of the layer, For the hidden layer The weight of the layer, For the hidden layer The offset parameters of the layer;

[0130] The output layer uses a softmax function. The softmax function converts the values ​​of multiple nodes in the output layer into a probability distribution, each node corresponds to a category, and the sum of the probabilities of all categories is 1. Here, each category is a fault type.

[0131] This embodiment can also provide an example of a method using Softmax classification. For example, the fault types may include normal fault-free, input side fault, output side fault and abnormal environment fault. The operating parameters may include DC side current, DC side voltage, AC side current, AC side voltage, AC frequency, working temperature and working humidity. Each fault type corresponds to some strongly correlated operating parameters. For example, input side fault corresponds to DC side current and DC side voltage, output side fault corresponds to AC side current, AC side voltage and AC frequency, and abnormal environment fault corresponds to working temperature and working humidity. When using the softmax function, when calculating the probability of a certain fault, the numerator is selected as the sum of the hidden layer outputs corresponding to the corresponding operating parameters, and the output corresponding to the accuracy of the operating parameters can be used as an additional weight. For example, the final output of the hidden layer is , assuming that the first two values ​​are the DC side current and DC side voltage:

[0132] ;

[0133] in, Represents the probability of input side failure. The probability of other failures can also be obtained by referring to the above, just replace the numerator with the relevant parameters. and All are training parameters. Fault-free will be treated as a special category. When the value is less than a preset threshold, it is directly defined as fault-free. The Softmax function classification task can be automatically established by inputting data and labels using a deep learning framework. This is a relatively mature technology, so I will not go into details here.

[0134] In this embodiment, the model adopts conventional training methods to obtain the parameters of the model itself through training data and labels. When extracting features, the overall parameter features and the parameter features at the current moment can be captured, and the hidden layer can better display the feature situation through changes to facilitate the subsequent output layer to output the classification results. The feature data of this embodiment takes into account the accuracy of the operating parameters, and focuses on the data of the photovoltaic inverter itself as much as possible to improve the reliability of fault identification. Therefore, any fault of the photovoltaic inverter will not only have unilateral data abnormalities, so it is not possible to simply determine which link has a fault through a threshold, so fault judgment can be more reliably performed through model learning, and the final fault types judged may include normal no fault, input side fault, output side fault, and environmental abnormal fault.

[0135] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A photovoltaic inverter fault diagnosis system optimization method based on deep learning, characterized in that: The following steps are involved: Periodically collect various operating parameters of the photovoltaic inverter and form an operating parameter sequence; Construct a time series operation parameter sequence set with a length of N; Perform a judgment based on the time series operating parameter sequence set to obtain the accuracy of the operating parameters, where the accuracy of the operating parameters ranges from [0, 1], and a larger value of the accuracy of the operating parameters represents a more accurate operating parameter; A judgment threshold is set according to the accuracy of the operating parameters and whether the photovoltaic inverter is suspected of having a fault is judged based on the judgment threshold, if the judgment result is no, no operation is performed, if the judgment result is yes, proceed to the next step; Based on deep learning, a secondary judgment is made according to the timing operation parameter sequence set and the accuracy of the operation parameters to predict the fault type of the photovoltaic inverter; A method for determining the accuracy of the operating parameters by performing a judgment based on the timing operating parameter sequence set is as follows: Setting a detection window, wherein the detection window is used to select M consecutive operation parameter sequences in the timing operation parameter sequence set; Starting from the first operating parameter sequence in the timing operating parameter sequence set, a first detection parameter sequence set is obtained by selecting through a detection window; Initialize the slide number , set the sliding step size ; With the sliding step length Slide the detection window back to select Detect parameter sequence set and update the sliding number ; like Go to the next step, otherwise return to the previous step. represents round down; Calculate the floating parameters of all detection parameter sequence sets and select the detection parameter sequence set with the smallest floating parameter as the standard detection parameter sequence set; Acquiring the accuracy of operating parameters based on the standard detection parameter sequence set; The method of calculating the floating parameters of all detection parameter sequence sets and selecting the detection parameter sequence set with the smallest floating parameter as the standard detection parameter sequence set is: Compute the floating parameters for the set of all detection parameter sequences: ; ; ; in, For the detecting said floating parameter of a set of parameter sequences, For the The first The first of the operating parameter sequences operating parameters, For the The detection parameter sequence set The average value of the operating parameters, is the total number of types of operating parameters, Based on The operation of finding the maximum value of the value change, Based on The operation of finding the minimum value of the value change is and are weights respectively; Select The minimum value in The detection parameter sequence set corresponding to the minimum value in is used as the standard detection parameter sequence set; The method for obtaining the accuracy of operating parameters based on the standard detection parameter sequence is: ; ; in, and represent the exponential function and logarithmic function with the natural logarithm e as the base, respectively. is the intermediate parameter, is the upper limit empirical value of the intermediate parameter, is the lower limit empirical value of the intermediate parameter, For the The minimum value in The accuracy of the operating parameters; The value range of M is: ; The method of setting a judgment threshold according to the accuracy of the operating parameters and judging whether the photovoltaic inverter is suspected of having a fault based on the judgment threshold is: Obtaining a standard threshold value for each operating parameter in the current operating parameter sequence; A judgment threshold is set for each of the operating parameters: ; ; ; in, The first one in the current operating parameter sequence the upper limit of the standard threshold value of the operating parameter, and are respectively the first the upper and lower limits of the judgment thresholds of the operating parameters, is the accuracy of the operating parameters, The total number of types of operating parameters; Determine whether each operating parameter in the current operating parameter sequence exceeds the corresponding standard threshold value, and if any operating parameter exceeds the corresponding standard threshold value, determine that the photovoltaic inverter is suspected of having a fault, otherwise determine that the photovoltaic inverter is not suspected of having a fault; The method for performing secondary judgment based on the time series operation parameter sequence set and the accuracy of the operation parameters based on deep learning is to perform secondary judgment through a trained network model, wherein the network model includes an input layer, a feature extraction layer, a hidden layer and an output layer connected in sequence; The feature extraction layer works as follows: Receive input data from the input layer, the input data is: ; ; ; , ; in, represents the set of timing operation parameter sequences, is the accuracy of the operating parameters, As of the current time of A sequence of operating parameters, Representative time The The average value of the operating parameters, , The total number of types of operating parameters; Perform feature extraction: ; The output of the feature extraction layer is in, Representative characteristic parameters of the operating parameters, is the sign function, Based on The function that finds the maximum value of the change of Based on The function that finds the minimum value of the change of represents the exponential function with the natural logarithm e as the base; The expression of the hidden layer is: ; ; in, Represents the hidden layer The output of the layer, For the hidden layer The weight of the layer, For the hidden layer The offset parameters of the layer; The output layer adopts the softmax function.

2. The photovoltaic inverter fault diagnosis system optimization method based on deep learning according to claim 1, characterized in that: The operating parameters include DC side current, DC side voltage, AC side current, AC side voltage, AC frequency, operating temperature and operating humidity.