Photovoltaic array fault intelligent diagnosis method and system

Abstract

The invention relates to the technical field of new energy, and discloses a photovoltaic array fault intelligent diagnosis method and system, and the method comprises the steps: collecting the historical operation data, assembly temperature and solar irradiance data of a photovoltaic string in a normal state and a fault state, and taking the data as an original sample feature data set; preprocessing the original sample data, and randomly dividing the original sample data into a training data set, a verification data set and a test data set; based on an LSTM-CNN-ArcLoss fusion network algorithm, constructing an intelligent fault diagnosis model of the photovoltaic array; and inputting the training data set, the verification data set and the test data set into a photovoltaic array intelligent fault diagnosis model, and outputting a photovoltaic string fault classification result by the photovoltaic array intelligent fault diagnosis model. The method not only can effectively reduce the operation and maintenance difficulty, save the operation and maintenance cost and improve the operation efficiency, but also assists the photovoltaic power station to continuously, stably and efficiently operate, and has certain engineering application value and economic and social benefits.

Description

Technical Field

[0001] The present invention relates to the field of new energy technology, and in particular to a photovoltaic array fault intelligent diagnosis method and system. Background Art

[0002] In recent years, with the rapid development of the new energy sector, my country's photovoltaic industry has leapt to the top of the world, from manufacturing to application. However, due to the widespread problems of scattered sites, harsh environments, and simple monitoring systems for photovoltaic power plants, photovoltaic array inspections are difficult and require extensive manpower and resources, resulting in low overall power generation efficiency and negatively impacting profitability.

[0003] Photovoltaic power stations are outdoor, open-air power plants. PV modules operate in open-air environments for extended periods of time, often unattended. The PV array is the fundamental unit of a photovoltaic power generation system, and its operating parameters are the most essential parameters for the operation of a PV power station. Due to the large number of PV arrays, array failures are a common cause of operational failures in PV power stations. Exposure to wind, sun, rain, snow, and other harsh natural conditions makes PV modules susceptible to failures such as open circuits, short circuits, abnormal aging, module breakage, and shadowing. The large number and dispersion of PV modules in a PV power station, coupled with the extensive and complex information collection, makes PV module fault diagnosis difficult.

[0004] Currently, the operation and maintenance of photovoltaic power plants mostly relies on traditional networked monitoring systems and inspection personnel to conduct regular inspections of the plant's electrical equipment. This method not only requires a large amount of manpower and material resources, but also presents significant operational and maintenance challenges, making it difficult for maintenance personnel to detect faults promptly and accurately. Failure to promptly detect faults can, over time, shorten the lifespan of the electrical equipment, leading to power generation losses and, in severe cases, even safety incidents. In the era of grid parity for photovoltaic power generation, this traditional operation and maintenance approach poses a significant obstacle to increasing the profitability of photovoltaic power plants.

[0005] The health of photovoltaic (PV) modules directly impacts the profitability of a power plant throughout its lifecycle. Traditional PV array fault detection typically requires on-site testing using IV testing equipment. Operations and maintenance personnel must remove modules and connect them to specialized IV diagnostic instruments for testing. Only one module can be tested at a time, and manual statistics and analysis of current and voltage curve data are then required to determine module failure, resulting in low efficiency. Finally, based on the analysis results, a second visit to the station is required to complete the fault closure loop. Traditional PV array fault detection not only requires extensive human and material resources, but also suffers from inefficient and poor diagnostic results. Improving PV array monitoring capabilities, quickly and accurately locating PV array faults and identifying their type, and then implementing effective repair and management measures, is crucial for improving PV power plant operation and maintenance efficiency, minimizing power generation losses, and ensuring the safe and reliable operation of PV power plants.

[0006] The output characteristics of photovoltaic arrays vary nonlinearly with solar radiation and ambient temperature. If a photovoltaic array fails, its output characteristics become more complex, making it difficult to directly observe and analyze the general rules of the failure from its output data. Summary of the Invention

[0007] In response to the above-mentioned technical deficiencies, the technical problem to be solved by the present invention is to provide a photovoltaic array fault intelligent diagnosis method and system, aiming to solve the problem of large number of on-site inspection personnel and low inspection efficiency in photovoltaic power stations, and reduce the manual operation and maintenance costs of photovoltaic power stations.

[0008] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a method for intelligent diagnosis of photovoltaic array faults, comprising: collecting historical operating data, component temperature, and solar irradiance data of photovoltaic strings in normal and fault states as original sample feature data sets; After preprocessing, the original sample data is randomly divided into a training data set, a verification data set, and a test data set; Constructing a photovoltaic array intelligent fault diagnosis model based on the LSTM-CNN-ArcLoss fusion network algorithm; The training data set, the validation data set and the test data set are input into the photovoltaic array intelligent fault diagnosis model, and the photovoltaic array intelligent fault diagnosis model outputs the photovoltaic string fault classification result.

[0009] Furthermore, the original sample feature data set includes: solar panel temperature, solar radiation, string operating current value, string operating voltage value, and parallel string operating current value in the same MPPT.

[0010] Further, based on k The time complexity of the mean clustering algorithm for filling missing values in the original sample data is as follows: ; in, l is the number of data attribute values contained in each photovoltaic array, t is the number of iterations for computing values within the cluster, n is the total dataset of the PV array, m is the total set of missing value data for photovoltaic arrays, k is the number of clusters in the clustering result.

[0011] Furthermore, the data of the training dataset are input into the LSTM-CNN-ArcLoss fusion network algorithm for training, and the model is preliminarily evaluated using the validation dataset to prevent overfitting. The classification accuracy of the photovoltaic array intelligent fault diagnosis model is verified using the test dataset.

[0012] Furthermore, the photovoltaic array intelligent fault diagnosis model is composed of an LSTM layer, a feature enhancement layer, a convolution layer, an adaptive pooling layer and a classification layer connected in sequence; Among them, the time series and nonlinear characteristics of the input data are mined through the LSTM layer to obtain one-dimensional data with time series characteristics; Use feature enhancement layers to expand the one-dimensional data output into two-dimensional data; The CNN network mines the spatial information of two-dimensional data through multi-scale fusion of convolutional layers, extracts data features, obtains feature maps containing multi-scale features, and adjusts the feature maps to a size suitable for ArcLoss classifier processing through adaptive pooling layers; The extracted feature maps are classified using the ArcLoss classifier, and the fault classification results of the photovoltaic strings are output.

[0013] Furthermore, it also includes: adding incremental learning function to the photovoltaic array intelligent fault diagnosis model, regularly using new or expanded data to retrain the photovoltaic array intelligent fault diagnosis model, so that the photovoltaic array intelligent fault diagnosis model can be iteratively upgraded in the continuous learning process.

[0014] Furthermore, the process of iterative upgrading of the photovoltaic array intelligent fault diagnosis model during continuous learning is as follows: The current PV array intelligent fault diagnosis model determines whether a string is faulty, writes the fault diagnosis results into the diagnostic history database of the fault management system, and periodically reads quantitative recent diagnostic data from the diagnostic history database. The read feature data set with the diagnosis results is written into the current photovoltaic array intelligent fault diagnosis model training database to replace the original training data set; The new training data set is then used to train the current photovoltaic array intelligent fault diagnosis model to generate a new photovoltaic array intelligent fault diagnosis model.

[0015] Furthermore, the photovoltaic array intelligent fault diagnosis model is periodically retrained quarterly based on the fault diagnosis situation.

[0016] A photovoltaic array fault intelligent diagnosis system, comprising: The acquisition module is used to collect historical operating data, component temperature and solar irradiance data of photovoltaic strings under normal and fault conditions as original sample feature data sets; A preprocessing module preprocesses the original sample data and randomly divides it into a training data set, a verification data set, and a test data set; The diagnosis module builds a photovoltaic array intelligent fault diagnosis model based on the LSTM-CNN-ArcLoss fusion network algorithm, and inputs the training data set, verification data set, and test data set into the photovoltaic array intelligent fault diagnosis model; An output module, configured to output a photovoltaic string fault classification result through the photovoltaic array intelligent fault diagnosis model; The upgrade module is used to determine whether there is a fault in the photovoltaic array string through the current photovoltaic array intelligent fault diagnosis model, write it into the diagnostic history database of the fault management system and regularly randomly read a certain amount of recent diagnostic data to replace the original training data set. The fault diagnosis model is then trained with the new training data set to generate a new photovoltaic array intelligent fault diagnosis model.

[0017] The beneficial effects of the present invention are: through intelligent means, remote fault diagnosis is realized, and the skill requirements of on-site operation and maintenance personnel are greatly reduced. It can not only effectively reduce the difficulty of operation and maintenance, save operation and maintenance costs, and improve operating efficiency, but also help photovoltaic power stations to operate continuously, stably and efficiently, and has certain engineering application value and economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a photovoltaic array fault intelligent diagnosis method provided in Example 1 of the present invention.

[0020] Figure 2 This is a network structure diagram of the photovoltaic array intelligent fault diagnosis model provided in Example 1 of the present invention.

[0021] Figure 3 This is a flowchart of the iterative upgrade of the photovoltaic array intelligent fault diagnosis model provided in Example 1 of the present invention.

[0022] Figure 4 A longitudinal comparison diagram of the operating currents of a faulty string and its parallel strings provided in Example 2 of the present invention DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example

[0024] like Figures 1 to 3 As shown, this embodiment provides a method for intelligent diagnosis of photovoltaic array faults, which specifically includes: collecting historical operating data, component temperature and solar irradiance data of photovoltaic strings in normal and fault states as original sample feature data sets. It should be noted that: Based on engineering data collection and analysis, it was determined that when a photovoltaic string fails, one or more of the string operating current, string operating voltage, and the operating current of parallel strings within the same MPPT will significantly change. Furthermore, considering the impact of solar panel temperature and solar radiation on the photovoltaic array, the present invention uses solar panel temperature, solar radiation, string operating current, operating voltage, and the operating current of parallel strings within the same MPPT as characteristic parameters for the intelligent diagnosis method for photovoltaic array faults.

[0025] After preprocessing, the original sample data is randomly divided into training data set, verification data set and test data set. It should be noted that: The preprocessing of sample data is to use efficient and fast methods to process data such as missing original values, incorrect original attribute values, and repeated original sample data, so as to obtain higher quality original data. Because in actual engineering applications, missing values account for a high proportion, and the K-means algorithm has the advantages of simplicity, efficiency, and ease of implementation, considering the processing quality and the amount of power station data, the present invention uses a clustering algorithm based on K-means for data preprocessing; and uses the K-means algorithm to preprocess photovoltaic array fault data. The time complexity of the missing value filling method based on the K-means algorithm is ;in, l is the number of data attribute values contained in each photovoltaic array, t is the number of iterations for computing values within the cluster, n is the total dataset of the PV array, m is the total set of missing value data for photovoltaic arrays, k is the number of clusters in the clustering result.

[0026] A photovoltaic array intelligent fault diagnosis model is constructed based on the LSTM-CNN-ArcLoss fusion network algorithm; the training data set, validation data set, and test data set are input into the photovoltaic array intelligent fault diagnosis model, and the photovoltaic array intelligent fault diagnosis model outputs the photovoltaic string fault classification results. It should be noted that: The data of the training dataset is input into the LSTM-CNN-ArcLoss fusion network algorithm for training, and the model is preliminarily evaluated using the validation dataset to prevent overfitting. For example, in the figure, Val_acc refers to the accuracy of the model calculated on the validation set, and in the figure, train_acc refers to the accuracy of the model calculated on the training set. Check whether the accuracy of the model on the validation set is higher than its accuracy on the training set; use the test dataset to verify the classification accuracy of the photovoltaic array intelligent fault diagnosis model.

[0027] like Figure 2 As shown in Figure 1, the photovoltaic array intelligent fault diagnosis model is composed of an LSTM layer, a feature enhancement layer, a convolution layer, an adaptive pooling layer, and a classification layer connected in sequence.

[0028] Specifically, the long short-term memory network (LSTM) in deep learning is used to mine the time series and nonlinear features of the input parameters to obtain a one-dimensional data output with time series features. The input data sequence length in the LSTM network is 5, and the output data sequence length is 32. Then, the feature enhancement is performed by copying the one-dimensional output data to obtain two-dimensional output data, which is then input into the CNN network. The CNN network mines the spatial information of the two-dimensional data through multi-scale fusion of the convolutional layer, extracts data features, obtains a feature map containing multi-scale features, and adjusts the feature map to a size suitable for processing by the ArcLoss classifier through the adaptive pooling layer. Finally, the extracted feature map is put into the ArcLoss classifier for classification, thereby outputting the fault classification result of the photovoltaic string.

[0029] The method provided in this embodiment combines the respective advantages of LSTM and CNN networks. Through hierarchical feature extraction, the performance of feature extraction is continuously enhanced, the generalization ability of the model is improved, and highly recognizable features can be extracted through a small number of network layers, reflecting the high-quality feature mining ability of this method. In the LSTM-CNN-ArcLoss fusion network, the LSTM layer can well complete the temporal feature extraction, and the convolution layer can well complete the spatial feature extraction and feature overlap in the time dimension, reducing the data scale. The ArcLoss function is used instead of the traditional Softmax function to improve the classification effect of the feature vector, further improving the model diagnosis accuracy.

[0030] Since the performance of photovoltaic modules is affected by the season and service life, data distribution varies at different times. For example, at noon, the solar irradiance in summer is generally greater than that in winter, and the temperature of photovoltaic modules is also significantly higher. As the operating time increases, photovoltaic modules will experience light-induced degradation and aging degradation due to differences in their own technical characteristics. Therefore, when the photovoltaic array intelligent fault diagnosis model is deployed in a production environment, as the external environment changes and time passes, the data parallax gradually expands, and the prediction accuracy of the photovoltaic array intelligent fault diagnosis model often decreases to varying degrees, causing the fault diagnosis model to gradually become invalid. In order to make the photovoltaic array intelligent fault diagnosis model unaffected by the season and power attenuation, and to ensure that the photovoltaic array intelligent fault diagnosis model has a high accuracy over time, this embodiment further iteratively upgrades the photovoltaic array intelligent fault diagnosis model, including: adding an incremental learning function to the photovoltaic array intelligent fault diagnosis model, regularly retraining the photovoltaic array intelligent fault diagnosis model with new or expanded data, so that the photovoltaic array intelligent fault diagnosis model can be iteratively upgraded during the continuous learning process.

[0031] like Figure 3 As shown in Figure 1, the specific process is as follows: the inverter reports the collected PV string feature data to the current PV array intelligent fault diagnosis model of the fault management system. The current PV array intelligent fault diagnosis model determines whether the string is faulty and writes the fault diagnosis results to the fault management system's diagnostic history database. A certain amount of recent diagnostic data is periodically read from the diagnostic history database. The read feature dataset containing the diagnostic results is written to the current PV array intelligent fault diagnosis model training database, replacing the original training dataset. The current PV array intelligent fault diagnosis model is then trained using the new training dataset to generate a new PV array intelligent fault diagnosis model. This enables automatic updating of the PV array intelligent fault diagnosis model, addressing the issue of fault diagnosis model failure caused by changes in the external environment and the passage of time. The PV array intelligent fault diagnosis model can be periodically retrained on a quarterly basis based on the fault diagnosis results.

[0032] This embodiment also provides a photovoltaic array fault intelligent diagnosis system, including: an acquisition module, which is used to collect historical operating data, component temperature and solar irradiance data of photovoltaic strings in normal and fault states as original sample feature data sets; a preprocessing module, which preprocesses the original sample data and randomly divides it into a training data set, a verification data set and a test data set; a diagnosis module, which constructs a photovoltaic array intelligent fault diagnosis model based on an LSTM-CNN-ArcLoss fusion network algorithm, and inputs the training data set, the verification data set and the test data set into the photovoltaic array intelligent fault diagnosis model; an output module, which is used to output the photovoltaic string fault classification result through the photovoltaic array intelligent fault diagnosis model; and an upgrade module, which is used to determine whether there is a fault in the photovoltaic array through the current photovoltaic array intelligent fault diagnosis model, write it into the diagnosis history database of the fault management system, and regularly randomly read a certain amount of recent diagnostic data to replace the original training data set, and then use the new training data set to train the fault diagnosis model to generate a new photovoltaic array intelligent fault diagnosis model.

[0033] The working principle of the present invention is as follows: the photovoltaic array fault intelligent diagnosis method proposed in the present invention combines the respective advantages of LSTM and CNN networks, fully mines the time and space information of the original data, and increases the class distance through the ArcLoss function, thereby improving the accuracy of fault diagnosis. Compared with the single LSTM network diagnosis model and the LSTM-CNN network diagnosis model, this method has faster convergence speed and higher accuracy, is more suitable for processing complex power station operation data, and has good engineering applicability. At the same time, in order to eliminate the influence of MPPT on photovoltaic string fault diagnosis, the parallel string working current value in the same MPPT is introduced as a new input parameter of the photovoltaic array intelligent fault diagnosis model, which optimizes the input parameters of the photovoltaic array intelligent fault diagnosis model and improves the accuracy of fault diagnosis. Example

[0034] This embodiment is the second embodiment of the present invention. Different from the first embodiment, this embodiment provides a verification test of a photovoltaic array fault intelligent diagnosis method to verify and illustrate the technical effects adopted in this method.

[0035] PV array fault analysis reveals that the primary characteristic parameters influencing PV array output characteristics are module temperature, solar irradiance, maximum power point current, short-circuit current, maximum power point voltage, and open-circuit voltage. However, due to equipment limitations in engineering practice, it is impossible to collect the short-circuit current and open-circuit voltage values of PV strings in real time. Therefore, in practical engineering applications, it is necessary to optimize the input characteristic parameters of the PV array intelligent fault diagnosis model.

[0036] With advances in photovoltaic inverter technology, string-type grid-connected photovoltaic inverters now generally feature real-time monitoring of each string's operating voltage, current, and generated power. Under consistent external environmental conditions (module temperature, light intensity), the power output trends of each input string of the same inverter are consistent, and the initial power ratios of the strings are constant. In the event of a string fault, horizontal and vertical comparisons between strings can be used to quickly identify faults such as open circuits, short circuits, shadowing, and abnormal aging.

[0037] This embodiment collects characteristic data under eight states: normal T1, open circuit fault T2, abnormal aging T3, short circuit fault T4, severe short circuit fault T5, slight shadow occlusion T6, heavy shadow occlusion T7, and severe shadow occlusion T8. The data are selected from data of a photovoltaic power station in Shandong from 6:00 to 18:00, and the data sampling interval is 1 minute.

[0038] During the construction of a photovoltaic power station, in order to ensure that the photovoltaic strings can always maintain the optimal output state, the photovoltaic strings are usually connected in parallel and then connected to the maximum power point tracking system (MPPT). This experimental power station uses two photovoltaic strings connected in parallel and then connected to one MPPT. Under the regulation of the MPPT system, when one of the strings fails, the operating current value Ia of the string connected in parallel with the faulty string will also change. Under the same irradiance and ambient temperature, the vertical comparison analysis of the operating current of the faulty string and its parallel string is as follows: Figure 4 shown.

[0039] pass Figure 4 It can be seen that when a string experiences a short-circuit fault, the operating current of the parallel strings remains essentially unchanged compared to normal conditions. When a string experiences an abnormal aging fault, the operating current of the parallel strings increases slightly compared to normal conditions. When a string experiences a short-circuit fault, the operating current of the parallel strings remains essentially unchanged compared to normal conditions. When a string experiences a severe short-circuit fault, the operating current of the parallel strings increases slightly compared to normal conditions. When a string experiences varying degrees of shadowing, the operating current of the parallel strings increases significantly. Table 1 shows the changes in the electrical characteristic parameters of strings in different states and their parallel connections under the same irradiance and ambient temperature.

[0040] Table 1: Changes in electrical characteristic parameters of strings in different states and their parallel strings under the same irradiance and ambient temperature.

[0041] Running status String operating current String operating voltage Parallel string operating current normal -- -- -- Circuit breaker fault disappear constant constant Abnormal aging Slight decline decline constant Short circuit fault decline decline constant Serious short circuit fault decline decline Slight rise Shadow occlusion glitch Slight decline Slight decline Slight rise Heavy shadow occlusion failure Significant decline Significant decline A significant increase Severe shadow occlusion failure Significant decline Significant decline Significant increase In summary, according to the analysis of measured data, under different fault conditions, the parallel string operating current value Ia will change to varying degrees. Therefore, this paper introduces the parallel string operating current value Ia on the basis of selecting component temperature T, light intensity S, operating current I, and operating voltage U as the direct input parameters of the photovoltaic array intelligent fault diagnosis system. A total of five characteristic parameters are used as input parameters of the fault diagnosis model.

[0042] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A photovoltaic array fault intelligent diagnosis method, characterized in that: include: Collect historical operating data of photovoltaic strings in normal and fault conditions, component temperature, and solar irradiance data as original sample feature data sets; After preprocessing, the original sample data is randomly divided into a training data set, a verification data set, and a test data set; Constructing a photovoltaic array intelligent fault diagnosis model based on the LSTM-CNN-ArcLoss fusion network algorithm; The training data set, the validation data set and the test data set are input into the photovoltaic array intelligent fault diagnosis model, and the photovoltaic array intelligent fault diagnosis model outputs the photovoltaic string fault classification result.

2. The photovoltaic array fault intelligent diagnosis method according to claim 1, characterized in that: The original sample feature data set includes: solar panel temperature, solar radiation, string operating current value, string operating voltage value, and parallel string operating current value in the same MPPT.

3. The photovoltaic array fault intelligent diagnosis method according to claim 1, characterized in that: based on k The time complexity of the mean clustering algorithm for filling missing values in the original sample data is as follows: ； in, l is the number of data attribute values contained in each photovoltaic array, t is the number of iterations for computing values within the cluster, n is the total dataset of the PV array, m is the total set of missing value data for photovoltaic arrays, k is the number of clusters in the clustering result.

4. The photovoltaic array fault intelligent diagnosis method according to claim 1, characterized in that: The data of the training dataset is input into the LSTM-CNN-ArcLoss fusion network algorithm for training. The model is preliminarily evaluated using the validation dataset to prevent overfitting. The classification accuracy of the photovoltaic array intelligent fault diagnosis model is verified using the test dataset.

5. The photovoltaic array fault intelligent diagnosis method according to claim 1, characterized in that: The photovoltaic array intelligent fault diagnosis model is composed of an LSTM layer, a feature enhancement layer, a convolution layer, an adaptive pooling layer and a classification layer connected in sequence; Among them, the time series and nonlinear characteristics of the input data are mined through the LSTM layer to obtain one-dimensional data with time series characteristics; Use feature enhancement layers to expand the one-dimensional data output into two-dimensional data; The CNN network mines the spatial information of two-dimensional data through multi-scale fusion of convolutional layers, extracts data features, obtains feature maps containing multi-scale features, and adjusts the feature maps to a size suitable for ArcLoss classifier processing through adaptive pooling layers; The extracted feature maps are classified using the ArcLoss classifier, and the fault classification results of the photovoltaic strings are output.

6. The photovoltaic array fault intelligent diagnosis method according to claim 1, characterized in that: Also includes: An incremental learning function is added to the photovoltaic array intelligent fault diagnosis model, and the photovoltaic array intelligent fault diagnosis model is regularly retrained with new or expanded data, so that the photovoltaic array intelligent fault diagnosis model can be iteratively upgraded in the continuous learning process.

7. The photovoltaic array fault intelligent diagnosis method according to claim 6, characterized in that: The process of iterative upgrading of the photovoltaic array intelligent fault diagnosis model during continuous learning is as follows: The current PV array intelligent fault diagnosis model determines whether a string is faulty, writes the fault diagnosis results into the diagnostic history database of the fault management system, and periodically reads quantitative recent diagnostic data from the diagnostic history database. The read feature data set with the diagnosis results is written into the current photovoltaic array intelligent fault diagnosis model training database to replace the original training data set; The new training data set is then used to train the current photovoltaic array intelligent fault diagnosis model to generate a new photovoltaic array intelligent fault diagnosis model.

8. The photovoltaic array fault intelligent diagnosis method according to claim 7, characterized in that: The photovoltaic array intelligent fault diagnosis model is periodically retrained quarterly according to the fault diagnosis situation.

9. A diagnostic system based on the photovoltaic array fault intelligent diagnosis method according to any one of claims 1 to 8, characterized in that: include: The acquisition module is used to collect historical operating data, component temperature and solar irradiance data of photovoltaic strings under normal and fault conditions as original sample feature data sets; A preprocessing module preprocesses the original sample data and randomly divides it into a training data set, a verification data set, and a test data set; The diagnosis module builds a photovoltaic array intelligent fault diagnosis model based on the LSTM-CNN-ArcLoss fusion network algorithm, and inputs the training data set, verification data set, and test data set into the photovoltaic array intelligent fault diagnosis model; An output module, configured to output a photovoltaic string fault classification result through the photovoltaic array intelligent fault diagnosis model; The upgrade module is used to determine whether there is a fault in the photovoltaic array string through the current photovoltaic array intelligent fault diagnosis model, write it into the diagnostic history database of the fault management system and regularly randomly read a certain amount of recent diagnostic data to replace the original training data set. The fault diagnosis model is then trained with the new training data set to generate a new photovoltaic array intelligent fault diagnosis model.

Images