Fault prediction method and device, computer equipment and medium

Abstract

The invention relates to the technical field of photovoltaic power generation, and discloses a fault prediction method and device, computer equipment and a medium. The method comprises the following steps: firstly, correcting I-V characteristics of a bypass diode by introducing equivalent resistors connected in series, and determining parameters in a photovoltaic module model by adopting a non-invasive parameter estimation method; then, screening data simulated by the photovoltaic module model through an information gain algorithm to obtain training data, and training to obtain a fault prediction model based on the training data; thirdly, performing fault prediction on the target photovoltaic module in combination with the self-encoder and the fault prediction model; then, fault inversion is carried out on parameters in the photovoltaic module model, and the parameters in the photovoltaic module model are optimized; and finally, based on electrical data obtained by simulation of the optimized photovoltaic module model, establishing a life prediction model, and accurately predicting the life of the photovoltaic module. According to the invention, the capabilities of fault detection, diagnosis and life prediction of the photovoltaic power station are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and particularly relates to a fault prediction method, device, computer device and medium. Background Art

[0002] A photovoltaic power station is an energy facility that converts solar energy into electrical energy. With the increasing global demand for renewable energy, the scale of photovoltaic power stations is continuously expanding. However, during the long-term operation of photovoltaic modules in a photovoltaic power station, affected by various factors, faults occur frequently, seriously restricting the stability and power generation efficiency of the photovoltaic power generation system. Therefore, how to predict faults of photovoltaic modules has become the current focus of attention. Summary of the Invention

[0003] In view of this, the present invention provides a fault prediction method, device, computer device and medium to solve the problem of fault prediction of photovoltaic modules in a photovoltaic power station.

[0004] In a first aspect, the present invention provides a fault prediction method, which includes:

[0005] Obtain electrical data of a target photovoltaic module; the target photovoltaic module includes at least one target bypass diode;

[0006] Based on the electrical data and a fault prediction model, obtain a fault prediction result of the target photovoltaic module; multiple training data of the fault prediction model are obtained by simulating a photovoltaic module based on a pre-constructed photovoltaic module model; the photovoltaic module model includes an equivalent bypass diode and an equivalent resistance connected in series with the equivalent bypass diode; the equivalent bypass diode is used to simulate the one-way conduction characteristic of the bypass diode in the photovoltaic module; the equivalent resistance is used to correct the working characteristic of the equivalent bypass diode.

[0007] Through the method provided in this embodiment, considering that in the actual operation of a photovoltaic module, the bypass diode in the photovoltaic module has an inherent resistance due to factors such as its internal structure, in this application embodiment, an equivalent resistance is introduced to simulate this characteristic, so that when generating training data using the photovoltaic module model, the working characteristic of the bypass diode under complex working conditions such as different light intensities, temperatures, and partial shading can be accurately reproduced. That is to say, the training data obtained based on the photovoltaic module model is highly consistent with the operating state of the photovoltaic module in the real scenario. Further, the fault prediction model trained based on the training data can accurately predict the faults of the photovoltaic module, significantly improving the accuracy and reliability of fault diagnosis.

[0008] In an alternative embodiment, the photovoltaic module model includes a bypass diode sub-model; the bypass diode sub-model is used to simulate the operation of the bypass diode in the photovoltaic module; the bypass diode sub-model is obtained based on the Schottky equation corresponding to the equivalent bypass diode and the equivalent resistance.

[0009] Through the above embodiment, the Schottky equation can accurately describe the current conduction mechanism of the bypass diode. Further, by constructing the bypass diode sub-model in combination with the series equivalent resistance, the current-voltage characteristics of the bypass diode under various working conditions (such as different light intensities, temperatures, and partial shading and other complex environments) can be accurately simulated.

[0010] In an alternative embodiment, the model parameters in the photovoltaic module model are obtained by fitting the current values and voltage values of the photovoltaic module under multiple preset irradiation conditions.

[0011] Through the above embodiment, different preset irradiation conditions simulate various light intensity and angle changes faced by the photovoltaic module in actual applications. Thus, the model parameters obtained by fitting can accurately reflect the true working characteristics of the bypass diode under different light environments.

[0012] In an alternative embodiment, the photovoltaic module model further includes a single diode sub-model; the single diode sub-model is used to simulate the operation of the photovoltaic cell in the photovoltaic module; the method further includes:

[0013] Based on the current values and voltage values of the photovoltaic module under multiple preset irradiation conditions, determine the current value and voltage value when the bypass diode in the photovoltaic module is in the conducting state;

[0014] Based on the current value and voltage value when the bypass diode is in the conducting state, determine the first parameter in the bypass diode sub-model;

[0015] Based on the current value and voltage value when the bypass diode is in the conducting state, determine the second parameter in the single diode sub-model.

[0016] In an alternative embodiment, determining the first parameter in the bypass diode sub-model based on the current value and voltage value when the bypass diode is in the conducting state includes:

[0017] Based on the current value and voltage value when the bypass diode is in the conducting state, the trust-region algorithm is used for nonlinear least squares fitting to obtain the first parameter; the first parameter enables the error of the bypass diode sub-model simulating the operation of the bypass diode to meet the preset conditions. Through the above implementation, when the trust-region algorithm performs nonlinear least squares fitting, a "trust region" will be constructed near the current parameter estimate value in each iteration. Within this region, by approximating the objective function (i.e., the error function between the simulated value and the actual value of the bypass diode sub-model) as a quadratic function, and then solving the minimum value of this quadratic function under the constraints of the trust region, the update direction and step size of the parameter can be obtained. In this way, the information in the current value and voltage value under multiple preset illumination conditions can be fully utilized to continuously adjust the parameters in the bypass diode sub-model, so that the finally obtained first parameter can accurately match the operating characteristics of the bypass diode under various illumination conditions, greatly improving the accuracy of the bypass diode sub-model in simulating the bypass diode.

[0018] In an alternative implementation, the current value and voltage value of the photovoltaic module when it is in a preset fault are obtained;

[0019] Based on the current value and voltage value of the photovoltaic module when it is in a preset fault, the particle swarm optimization algorithm is used to optimize the second parameter in the single-diode sub-model to obtain the second parameter under the preset fault.

[0020] Through the above implementation, when the photovoltaic module is in a preset fault, the change rules of the internal current and voltage of the photovoltaic module are different from those in the normal operating state. Based on the electrical data in these fault states and using the particle swarm optimization algorithm to correct the second parameter of the single-diode sub-model, the electrical characteristics of the photovoltaic cell under fault conditions can be accurately described by the single-diode sub-model. For example, when a local short-circuit fault occurs in the photovoltaic cell, the second parameter in the single-diode sub-model is optimized, so that the single-diode sub-model can accurately simulate the change trend of abnormal current increase and voltage drop, greatly improving the simulation accuracy of the single-diode sub-model for the photovoltaic cell in the fault state and providing a reliable basis for subsequent fault analysis.

[0021] In an alternative implementation, the first loss function of the fault prediction model is obtained by weighted summation of multiple second loss functions; among them, the second loss function corresponds to the preset fault one by one, and the second loss function is used to indicate the prediction error of the fault prediction model for the preset fault; the weights of the multiple second loss functions are determined based on at least one of the severity, occurrence times, and detection difficulty of the corresponding faults.

[0022] Through the above implementation, different faults have different degrees of impact on photovoltaic modules. The weight of the second loss function is determined by the severity of the fault, the number of faults that occur, and the difficulty of detection, that is, different weights are assigned to different faults. In this way, the fault prediction model can prioritize the optimization of the prediction accuracy of high-weight faults (such as high severity and frequent faults). For example, if a fault causes huge losses to the photovoltaic power station, the weight of the first loss function corresponding to it can be significantly higher than the weight of the first loss function corresponding to low-impact faults, thereby guiding the fault prediction model to pay more attention to the prediction ability of such faults and improving the pertinence of fault prediction. For another example, a higher weight is assigned to the second loss function corresponding to faults with high detection difficulty, so that the fault prediction model can mine the weak features of these difficult-to-detect faults during the training process, thereby improving the detection ability of complex faults. For another example, a higher weight is assigned to the second loss function corresponding to frequent faults, so that the fault prediction model can focus on learning the features of high-frequency faults during training, thereby improving the prediction accuracy of high-frequency faults.

[0023] In an optional embodiment, obtaining a fault prediction result of a target photovoltaic module based on electrical data and a fault prediction model includes:

[0024] The electrical data is input into a pre-built autoencoder to obtain reconstructed data of the target photovoltaic module; based on the reconstructed data and the electrical data, it is determined whether the target photovoltaic module has failed;

[0025] In the event of a failure of the target photovoltaic module, a fault prediction result is obtained based on the electrical data and the fault prediction model.

[0026] Through the above implementation, the autoencoder's reconstructed data is compared with the electrical data to determine whether the target PV module has failed. If a fault is confirmed, the fault prediction model is then used to predict the failure. This complementary approach forms a dual verification mechanism, significantly reducing the risk of misjudgments and missed detections and improving the reliability of the fault prediction results. Furthermore, the autoencoder is capable of deep feature extraction and reconstruction of the input electrical data. During the encoding phase, it compresses the high-dimensional electrical data into a low-dimensional latent space, extracting the information that best represents the data's essential characteristics. During the decoding phase, it attempts to reconstruct this feature information back to the original data's dimensions. This process helps uncover hidden patterns and anomalies in the electrical data; even subtle changes can be reflected in reconstruction errors. Consequently, the autoencoder can accurately determine whether the target PV module has failed.

[0027] In an optional embodiment, determining whether a target photovoltaic module has failed based on the reconstruction data and the electrical data includes:

[0028] In the case where the deviation between the reconstructed data and the electrical data is greater than a preset threshold, it is determined that the target photovoltaic module has failed.

[0029] Through the above embodiments, the purpose of the autoencoder to reconstruct data is to restore the input electrical data as much as possible. When the photovoltaic module fails, its internal electrical characteristics will change, resulting in abnormalities in the original electrical data. This abnormality will be reflected in the difference between the reconstructed data and the original data. Even if the changes in the initial stage of the failure are very subtle, by comparing the difference with the preset threshold, these changes can be captured in a timely manner to achieve the detection of early failures. For example, when the photovoltaic cells begin to show slight performance degradation, there will be slight changes in electrical parameters such as the output current and voltage. The autoencoder will generate a difference from the original data due to these changes during the reconstruction of the data. Therefore, potential failure risks can be detected through difference comparison.

[0030] In an alternative embodiment, the fault prediction model is a neural network optimized based on a pruning algorithm; in the case where the target photovoltaic module fails, based on the electrical data and the fault prediction model, a fault prediction result is obtained, including:

[0031] In the case where the target photovoltaic module fails, the encoded data corresponding to the electrical data is input into the fault prediction model to obtain a fault prediction result; the encoded data corresponding to the electrical data is obtained by inputting the electrical data into the encoder in the autoencoder.

[0032] Through the above embodiments, on the one hand, the fault prediction model is a neural network optimized using a pruning algorithm, which removes redundant connections and neurons in the original neural network, reduces unnecessary computational complexity, and improves the operating efficiency of the fault prediction model. On the other hand, the encoder of the autoencoder converts the electrical data into encoded data. During this process, the encoder extracts the key features in the electrical data and performs dimensionality reduction, removing redundant information in the electrical data and retaining the key features related to the fault, further reducing the computational complexity of the fault prediction model while maintaining a high accuracy of the fault prediction model.

[0033] In an alternative embodiment, the method further includes:

[0034] Obtain a plurality of initial data; the plurality of initial data correspond to multiple fault types;

[0035] Based on the information gain of each initial data, screen the plurality of initial data to obtain a plurality of training data.

[0036] Through the above embodiments, the information gain of the initial data can characterize the amount of information provided by the initial data for fault prediction. Screening the initial data based on the information gain can retain the data that carries key information and is helpful for accurately judging the fault type, and remove the data with low information content and little effect on fault classification. In this way, on the one hand, since the training data obtained after screening contains the key data for distinguishing different fault types, when using this data to train the model, the characteristics of various fault types can be learned more accurately, enhancing the accuracy of the fault prediction model in predicting faults. On the other hand, screening the initial data based on the information gain reduces the dimension of the training data, enabling the fault prediction model to converge more quickly and saving training time.

[0037] In an alternative embodiment, screening a plurality of initial data based on the information gain of each initial data to obtain a plurality of training data includes:

[0038] Based on the information gain of each initial data, using a sliding window test algorithm to determine an information gain threshold;

[0039] Based on the information gain threshold and the information gain of each initial data, obtain a plurality of training data.

[0040] In an alternative embodiment, obtaining the plurality of training data based on the information gain threshold and the information gain of each initial data includes:

[0041] Based on the information gain threshold and the information gain of each initial data, obtain initial training data;

[0042] Using the principal component analysis method and the initial training data, obtain the plurality of training data.

[0043] In an alternative embodiment, the photovoltaic module model includes a single diode sub-model; the single diode sub-model is used to simulate the operation of the photovoltaic cells in the photovoltaic module; the method further includes:

[0044] Obtain the environmental data of the target photovoltaic module;

[0045] Based on the environmental data, determine the operating temperature of the target photovoltaic module;

[0046] Based on the operating temperature and the second parameters of the single diode sub-model, determine the degradation rate of the target photovoltaic module;

[0047] Based on the degradation rate, predict the lifespan of the target photovoltaic module.

[0048] Through the above embodiments, based on the environmental data, the operating temperature of the target photovoltaic module is determined, taking into account the self-heating effect of the solar module due to solar radiation during actual operation. Compared with directly using the environmental temperature as the operating temperature of the photovoltaic module, the operating temperature of the target photovoltaic module determined in the embodiments of the present application is more accurate and can more realistically reflect the thermal state of the photovoltaic module during actual operation, providing a basis for accurately determining the degradation rate of the target photovoltaic module and predicting its lifespan.

[0049] In an alternative embodiment, predicting the lifespan of the target photovoltaic module based on the degradation rate includes:

[0050] Based on the degradation rate, determining the time when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold;

[0051] Based on the time, determining the lifespan of the target photovoltaic module.

[0052] In an alternative embodiment, determining the time when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold based on the degradation rate includes:

[0053] Based on the degradation rate and a pre-constructed corrosion degradation model, determining the time when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold.

[0054] In an alternative embodiment, the corrosion degradation model includes:

[0055]

[0056] where ΔP(t) is the power attenuation at time t, ΔP ∞ is the saturated power attenuation after long-term exposure, ΔP(t i ) is the initial power loss, R D is the degradation rate, t i is the starting degradation time.

[0057] In an alternative embodiment, determining the degradation rate of the target photovoltaic module based on the operating temperature and the second parameter of the single-diode sub-model includes:

[0058]

[0059] where R D is the degradation rate of the target photovoltaic module, T is the operating temperature of the target photovoltaic module, RH is the relative humidity, E A is the activation energy of the sealing material, k B is the Boltzmann constant, n is the humidity coefficient, A is an empirical constant, R s,new is the second parameter in the single-diode sub-model.

[0060] In an alternative embodiment, the method further includes:

[0061] Based on the severity, occurrence times, and detection difficulty of the faults corresponding to each of the second loss functions, using a risk priority numbering mechanism, determine the risk priority of the faults corresponding to each of the second loss functions;

[0062] Based on the risk priority of the faults corresponding to each of the second loss functions, determine the weight of each of the second loss functions.

[0063] In a second aspect, the present invention provides a fault prediction device, which includes:

[0064] A first acquisition module, configured to acquire electrical data of a target photovoltaic module; the target photovoltaic module includes at least one target bypass diode;

[0065] A first prediction module, configured to obtain a fault prediction result of the target photovoltaic module based on the electrical data and a fault prediction model; multiple training data of the fault prediction model are obtained by simulating a photovoltaic module based on a pre-constructed photovoltaic module model; the photovoltaic module model includes an equivalent bypass diode and an equivalent resistor connected in series with the equivalent bypass diode; the equivalent bypass diode is used to simulate the one-way conduction characteristic of the bypass diode in the photovoltaic module; the equivalent resistor is used to correct the operating characteristic of the equivalent bypass diode.

[0066] Through the device provided in this embodiment, considering that in the actual operation of a photovoltaic module, the bypass diode in the photovoltaic module has an inherent resistance due to factors such as its internal structure, in this application embodiment, an equivalent resistor is introduced to simulate this characteristic, so that when generating training data using the photovoltaic module model, the electrical behavior of the bypass diode under complex working conditions such as different light intensities, temperatures, and partial shading can be accurately reproduced. That is to say, the training data obtained based on the photovoltaic module model is highly consistent with the operating state of the photovoltaic module in the real scenario. In this way, the fault prediction model trained based on the training data can accurately predict the faults of the photovoltaic module, significantly improving the accuracy and reliability of fault diagnosis.

[0067] In a third aspect, the present invention provides a computer device, including: a memory and a processor, which are communicatively connected to each other, and the memory stores computer instructions, and the processor executes the computer instructions to execute the fault prediction method according to the first aspect or any corresponding embodiment thereof.

[0068] In a fourth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the fault prediction method according to the first aspect or any corresponding embodiment thereof.

[0069] In a fifth aspect, the present invention provides a computer program product, including computer instructions for causing a computer to execute the fault prediction method according to the first aspect or any corresponding embodiment thereof above. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0071] Figure 1 is a schematic flowchart of a fault prediction method according to an embodiment of the present invention;

[0072] Figure 2 is a schematic diagram of obtaining a fault prediction result of a target photovoltaic module by using a fault prediction model according to an embodiment of the present invention;

[0073] Figure 3 is a schematic diagram of a reconstruction error of using an autoencoder for fault detection according to an embodiment of the present invention;

[0074] Figure 4 is an equivalent circuit diagram of a photovoltaic module according to an embodiment of the present invention;

[0075] Figure 5 is an equivalent circuit diagram of a photovoltaic module under uniform illumination conditions according to an embodiment of the present invention;

[0076] Figure 6 is an equivalent circuit diagram of the photovoltaic module under partial shading conditions according to an embodiment of the present invention;

[0077] Figure 7 is an equivalent circuit diagram in the case of a short circuit of a photovoltaic cell in a photovoltaic module according to an embodiment of the present invention;

[0078] Figure 8 is an I-V curve diagram of a photovoltaic module according to an embodiment of the present invention;

[0079] Figure 9 is a result comparison diagram of verifying a bypass diode sub-model through a simulation experiment according to an embodiment of the present invention;

[0080] Figure 10 is a schematic diagram of the information gain of initial data according to an embodiment of the present invention;

[0081] Figure 11Schematic diagram of classifier performance with different information gain thresholds according to an embodiment of the present invention;

[0082] Figure 12 Schematic diagram of a confusion matrix according to an embodiment of the present invention;

[0083] Figure 13 Schematic diagram of the fault diagnosis accuracy of a neural network under different pruning ratios according to an embodiment of the present invention;

[0084] Figure 14 Schematic diagram of determining training data based on feature extraction using information gain and principal component analysis according to an embodiment of the present invention;

[0085] Figure 15 Schematic diagram of a fitting curve according to an embodiment of the present invention;

[0086] Figure 16 I-V curve change diagram of a photovoltaic module due to conductive finger corrosion under a damp heat test according to an embodiment of the present invention;

[0087] Figure 17 Overall flow schematic diagram of a fault prediction method according to an embodiment of the present invention;

[0088] Figure 18 Structural block diagram of a fault prediction device according to an embodiment of the present invention;

[0089] Figure 19 Hardware structure schematic diagram of a computer device according to an embodiment of the present invention. Specific embodiments

[0090] To make the objectives, 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 with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0091] First, an exemplary introduction to the application scenarios of the embodiments of the present application is provided.

[0092] A photovoltaic power station is an energy facility that converts solar energy into electrical energy. With the increasing global demand for renewable energy, the scale of photovoltaic power stations is constantly expanding. However, the efficient operation of a photovoltaic power station depends on the continuous monitoring and maintenance of photovoltaic modules (also known as photovoltaic components) and systems. During actual operation and maintenance, common challenges include faults in photovoltaic modules, such as partial shading, hot spot effect, wiring problems, and component aging, which can all lead to a decrease in the power generation efficiency of the photovoltaic power station. Since photovoltaic power stations usually cover a large area, traditional manual inspection methods are difficult to detect problems in a timely and comprehensive manner, and the detection cost is high. In addition, the life management of photovoltaic cells is not perfect, and there is a lack of accurate life prediction models, resulting in the inability to reasonably plan the replacement time of photovoltaic modules, thereby increasing the operation and maintenance costs.

[0093] In view of this, an embodiment of the present application provides a fault prediction method to accurately predict faults in photovoltaic modules. It should be noted that for the fault prediction method provided by the embodiment of the present invention, the execution subject can be a fault prediction device, and this fault prediction device can be implemented as part or all of an electronic device through software, hardware, or a combination of software and hardware. Among them, the electronic device can be a server or a terminal. Among them, the server in the embodiment of the present application can be a single server or a server cluster composed of multiple servers, and the terminal in the embodiment of the present application can be other intelligent hardware devices such as a smart phone, a personal computer, a tablet computer, a wearable device, and a smart robot. In the following method embodiments, the execution subject is taken as an electronic device for illustration.

[0094] According to an embodiment of the present invention, an embodiment of a fault prediction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0095] The following embodiments of the present application will exemplarily introduce the solution of the fault prediction method in three parts.

[0096] The first part, in combination with Figure 1 、 Figure 2 、 Figure 3 , introduce the fault prediction method provided by the embodiment of the present application, aiming to introduce the overall implementation method of fault prediction for the target photovoltaic module.

[0097] The second part, in combination with Figures 4 to 9 , introduce the fault prediction method provided by the embodiment of the present application, aiming to introduce the specific implementation method of constructing a photovoltaic module model.

[0098] The third part, in combination with Figures 10 to 14, introduce the fault prediction method provided by the embodiments of the present application, aiming to introduce the specific implementation manner of constructing a fault prediction model.

[0099] Part Four, in combination with Figure 15 , Figure 16 , introduce the fault prediction method provided by the embodiments of the present application, aiming to introduce the specific implementation manner of predicting the lifespan of a target photovoltaic module.

[0100] The following, in combination with Figure 1 , Figure 2 , Figure 3 , introduce the first part of the embodiments of the present application, that is, introduce the overall implementation manner of using a fault prediction model to perform fault prediction on a target photovoltaic module.

[0101] In this embodiment, a fault prediction method is provided, which can be used for the above-mentioned electronic devices, such as servers, etc. Figure 1 is a schematic flowchart of a fault prediction method provided according to an embodiment of the present invention. As Figure 1 shown, this process includes the following S101 - S102:

[0102] S101: Obtain the electrical data of the target photovoltaic module.

[0103] Among them, the target photovoltaic module includes at least one target bypass diode.

[0104] In a possible implementation manner, the photovoltaic module is used to convert solar energy into electrical energy and is the basic power generation unit of a photovoltaic power station. The photovoltaic module contains multiple photovoltaic cells, and each photovoltaic cell is connected in parallel with a bypass diode. Each photovoltaic cell and the corresponding bypass diode are connected in parallel to form a branch (also called a photovoltaic sub-module). The bypass diodes in the photovoltaic module are connected in parallel across the two ends of the photovoltaic cells. When the photovoltaic cells in the photovoltaic module are blocked or malfunction, the bypass diodes will conduct, providing a bypass channel for the current to avoid the blocked or malfunctioned photovoltaic cells from consuming the electrical energy generated by other photovoltaic modules. It should be noted that the embodiments of the present application do not specifically limit the number of target bypass diodes in the target photovoltaic module.

[0105] In a possible implementation manner, the electrical data of the target photovoltaic module includes but is not limited to the current, voltage, etc. of the target photovoltaic module. Exemplarily, the electrical data of the target photovoltaic module can be measured by sensors, such as current sensors and voltage sensors.

[0106] In another possible implementation, the electrical data of the target photovoltaic module includes the time-domain characteristics of the current, voltage, etc. of the target photovoltaic module. Exemplarily, the electrical data includes the average value, standard deviation, root mean square, variance value, kurtosis value, skewness value, square root of amplitude, crest factor, impulse factor, marginal factor, form factor, and peak value of the current, voltage, etc.

[0107] In yet another possible implementation, the electrical data of the target photovoltaic module can also be the frequency-domain characteristics of the current, voltage, etc. of the target photovoltaic module. Exemplarily, the electrical data includes the average frequency, root variance frequency, and root mean square frequency of the current, voltage, etc.

[0108] S102: Based on the electrical data and the fault prediction model, obtain the fault prediction result of the target photovoltaic module.

[0109] Among them, the multiple training data of the fault prediction model are obtained by simulating the photovoltaic module based on a pre-constructed photovoltaic module model; the photovoltaic module model includes an equivalent bypass diode and an equivalent resistor connected in series with the equivalent bypass diode; the equivalent bypass diode is used to simulate the one-way conduction characteristic of the bypass diode in the photovoltaic module; the equivalent resistor is used to correct the working characteristic of the equivalent bypass diode.

[0110] In one possible implementation, the fault prediction result refers to the type of fault predicted by the fault prediction model for the target photovoltaic module based on the electrical data, such as dust accumulation fault, shading fault, degradation fault, short-circuit fault, etc. Among them, the dust accumulation fault refers to the reduction of the light absorption rate of the photovoltaic module due to the long-term accumulation of dust, dirt and other impurities on the surface of the photovoltaic module, thereby affecting the power generation efficiency of the photovoltaic module. The shading fault refers to the partial photovoltaic cells of the photovoltaic module being blocked by objects such as trees, buildings, bird droppings, etc., so that the blocked part cannot receive sunlight normally, thus affecting the performance of the entire photovoltaic module. The degradation fault refers to the gradual decline of the performance of the photovoltaic module with the increase of the use time and the influence of environmental factors, mainly manifested as the reduction of the photoelectric conversion efficiency and the decrease of the output power. The short-circuit fault refers to the fault that a low-resistance path appears between the positive and negative poles in the internal or external circuit of the photovoltaic module, resulting in an abnormal increase in current and a decrease in the output voltage of the photovoltaic module.

[0111] In a possible implementation, the photovoltaic module model is used to simulate the operating characteristics of the photovoltaic module to analyze the behavior of the photovoltaic module under different operating conditions. In the photovoltaic module model, the equivalent resistance and the equivalent bypass diode are used to simulate the operation of the bypass diode in the photovoltaic module. Specifically, the equivalent bypass diode is used to simulate the one-way conduction characteristic of the bypass diode in the actual photovoltaic module, where the one-way conduction characteristic is that the bypass diode in the photovoltaic module conducts when the negative potential of the photovoltaic cell connected in parallel with it is greater than the positive potential. For example, when the photovoltaic cell in the photovoltaic module generates a reverse voltage due to being shaded or other reasons, the bypass diode conducts to prevent the photovoltaic cell from being reversely broken down and plays a protective role. The equivalent resistance is used to correct the operating characteristics of the equivalent bypass diode so that the photovoltaic module model can more accurately simulate the actual operating conditions of the photovoltaic module. This is because the bypass diode is not an ideal one-way conduction device and has non-ideal characteristics such as certain conduction resistance and forward voltage drop, and the equivalent resistance is used to simulate the influence of these characteristics on the operation of the bypass diode.

[0112] In a possible implementation, the fault prediction model can be a deep learning model, a reinforcement learning model, etc. The fault prediction model is trained based on multiple training data. It should be noted that the construction process of the fault prediction model will be described in the second part of the following embodiments and will not be elaborated here.

[0113] In a possible implementation, by using the photovoltaic module model, different simulation conditions are set, such as different light intensities, temperatures, loads, etc., to simulate the operation of the photovoltaic module under various actual operating conditions, so as to obtain the training data of the fault prediction model.

[0114] In a possible implementation, the electrical data is input into the fault prediction model to obtain the fault prediction result. The fault prediction model realizes the fault prediction of the photovoltaic module based on the features in the electrical data. Of course, it is also possible to first use the electrical data for fault judgment, and when it is determined that the target photovoltaic module has a fault, use the fault prediction model to further determine the fault prediction result.

[0115] Through the method provided in this embodiment, considering that in the actual operation of the photovoltaic module, the bypass diode in the photovoltaic module has an inherent resistance due to factors such as its internal structure, in this application embodiment, an equivalent resistance is introduced to simulate this characteristic, so that when generating training data using the photovoltaic module model, the operating characteristics of the bypass diode under complex operating conditions such as different light intensities, temperatures, and partial shading can be accurately reproduced. That is to say, the training data obtained based on the photovoltaic module model is highly consistent with the operating state of the photovoltaic module in the real scenario. Further, the fault prediction model trained based on the training data can accurately predict the faults of the photovoltaic module, significantly improving the accuracy and reliability of fault diagnosis.

[0116] In some examples, fault judgment can be first performed on the target photovoltaic module based on electrical data. In the case of a fault in the target photovoltaic module, a fault prediction model is used to obtain a fault prediction result.

[0117] In this way, on the one hand, first judge whether the target photovoltaic module fails by comparing the reconstructed data of the autoencoder with the electrical data. In the case of confirming the occurrence of a fault, then use the fault prediction model for fault prediction, complementing the autoencoder and the fault prediction model to form a double verification mechanism, which greatly reduces the risks of misjudgment and missed judgment and improves the reliability of the fault prediction result. On the other hand, the autoencoder can perform deep feature extraction and reconstruction on the input electrical data. In the encoding stage, it compresses the high-dimensional electrical data into a low-dimensional latent space and extracts the information that best represents the essential features of the data; in the decoding stage, it attempts to reconstruct these feature information back to the original data dimension. This process helps to discover hidden patterns and anomalies in the electrical data, and even very subtle changes may be reflected in the reconstruction error. Therefore, based on the autoencoder, it can accurately judge whether the target photovoltaic module fails.

[0118] In a possible implementation, as Figure 2 shown, the above S102 obtains a fault prediction result through the following S1021 - S1023:

[0119] S1021: Input the electrical data into a pre-constructed autoencoder to obtain the reconstructed data of the target photovoltaic module.

[0120] Optionally, the autoencoder is a machine learning model based on a neural network, consisting of an encoder and a decoder. The encoder maps the input data to a low-dimensional latent space, and the decoder then reconstructs the low-dimensional data into a high-dimensional output, that is, the reconstructed data. It can be understood that the pre-constructed autoencoder is a model that has been trained in advance and can be used to process the electrical data of photovoltaic modules.

[0121] Exemplarily, the formulas for the encoder and the decoder are as follows:

[0122] Z = f(X) = σ(W1X + b1) (Formula One)

[0123] where Z is the output data of the encoder, X is the input data of the encoder, that is, the electrical data, W1 is the weight matrix of the encoder, b1 is the bias, and σ is the activation function.

[0124]

[0125] where is the output data of the decoder, that is, the reconstructed data, W2 is the weight matrix of the decoder, b2 is the bias, and σ is the activation function.

[0126] S1022: Determine whether the target photovoltaic module has failed based on the reconstructed data and electrical data.

[0127] Optionally, when the deviation between the reconstructed data and the electrical data is greater than a preset threshold, it is determined that the target photovoltaic module has failed.

[0128] Wherein, the deviation can be the difference between the reconstructed data and the electrical data, or the mean square error between the reconstructed data and the electrical data, etc. The preset threshold can be defined according to the actual situation.

[0129] Exemplarily, when the mean square error between the reconstructed data and the electrical data is greater than the preset threshold, it is determined that the target photovoltaic module has failed. The calculation formula for the mean square error between the reconstructed data and the electrical data is as follows:

[0130]

[0131] Wherein, is the mean square error between the reconstructed data and the electrical data, n is the number of data, X i is the electrical data, is the output data after being reconstructed by the encoder-decoder, that is, the reconstructed data.

[0132] In the embodiments of the present application, the purpose of the autoencoder to reconstruct the data is to restore the input electrical data as much as possible. When the photovoltaic module fails, its internal electrical characteristics will change, resulting in abnormalities in the original electrical data. This abnormality will be reflected in the difference between the reconstructed data and the original data. Even if the changes in the initial stage of the failure are very subtle, by comparing the difference with the preset threshold, these changes can be captured in time to achieve the detection of early failures. For example, when the photovoltaic module begins to show slight performance degradation, its output electrical parameters such as current and voltage will have slight changes. The autoencoder will generate a difference from the electrical data when reconstructing the data. Therefore, this potential failure risk can be discovered through the deviation.

[0133] Figure 3 Shows the reconstruction error of using the autoencoder for fault detection. When the mean square error between the electrical data and the reconstructed data of the photovoltaic module in the fault state (fault class) is relatively high, while the mean square error between the electrical data and the reconstructed data of the photovoltaic module in normal operation (STC) is relatively low.

[0134] Optionally, determine whether the target photovoltaic module has failed based on the correlation between the reconstructed data and the electrical data.

[0135] Wherein, the correlation can be characterized by a correlation coefficient, etc., and the embodiments of the present application do not make specific limitations on this.

[0136] S1023: In the case of a fault in the target photovoltaic module, based on electrical data and a fault prediction model, obtain a fault prediction result.

[0137] In the above S1023, the fault prediction result is determined in the following manner: In the case of a fault in the target photovoltaic module, input the encoded data corresponding to the electrical data into the fault prediction model to obtain the fault prediction result. Among them, the encoded data corresponding to the electrical data is obtained after inputting the electrical data into the encoder in the autoencoder.

[0138] During the process of the encoder in the autoencoder converting the electrical data into encoded data, the encoder extracts the key features in the electrical data and performs dimensionality reduction, removes the redundant information in the electrical data, retains the key features related to the fault, further reduces the computational complexity of the fault prediction model, and at the same time maintains a high accuracy rate of the fault prediction model.

[0139] Optionally, the fault prediction model is a neural network optimized based on a pruning algorithm. In this way, the redundant connections and neurons in the original neural network are removed, the unnecessary computational amount is reduced, and the operating efficiency of the fault prediction model is improved.

[0140] The above is the first part of the embodiments of this application. Next, in combination with Figures 4 to 9 , introduce the fault prediction method provided by the embodiments of this application, aiming to introduce the specific implementation manner of constructing the photovoltaic module model.

[0141] In some embodiments, the photovoltaic module model includes a bypass diode sub-model. The bypass diode sub-model is used to simulate the operation of the bypass diode in the photovoltaic module; the bypass diode sub-model is obtained based on the Schottky equation corresponding to the equivalent bypass diode and the equivalent resistance.

[0142] In this way, the Schottky equation can accurately describe the current conduction mechanism of the bypass diode. Further, by constructing the bypass diode sub-model in combination with the series equivalent resistance, the current-voltage characteristics of the bypass diode under various working conditions (such as complex environments with different light intensities, temperatures, and partial shading, etc.) can be accurately simulated.

[0143] In some embodiments, the photovoltaic module model includes a single diode sub-model. The single diode sub-model is used to simulate the operation of the photovoltaic cell in the photovoltaic module.

[0144] Figure 4 is an equivalent circuit diagram of a photovoltaic module. In this photovoltaic module, there is a photovoltaic cell and a bypass diode connected in parallel with the photovoltaic cell. In Figure 4 the left figure, the photocurrent source I ph , diode D, series resistance R s, parallel resistor R h Used to describe the operating characteristics of photovoltaic cells in a photovoltaic module, equivalent bypass diode D', equivalent resistor R bd Used to describe the operating characteristics of the bypass diode in a photovoltaic module. Specifically, the photocurrent source I ph Used to simulate the current generated by a photovoltaic cell under illumination. Diode D is used to reflect the non-linear electrical characteristics of the photovoltaic cell, series resistor R s Represents the ohmic resistance inside the photovoltaic cell, parallel resistor R h Reflects the leakage phenomenon inside the photovoltaic cell. The equivalent bypass diode D' is used to simulate the one-way conduction characteristic of the bypass diode in the photovoltaic module, equivalent resistor R bd Used to correct the operating characteristics of the bypass diode D'. In Figure 4 , the photocurrent source I in the left figure ph , diode D, series resistor R s , parallel resistor R h Is simplified to the photovoltaic cell in the right figure.

[0145] In this photovoltaic module, an equivalent resistor in series with the bypass diode is introduced to more accurately simulate the correspondence between the current and voltage of the bypass diode. Especially when the photovoltaic module is operating under partial shading or low current conditions, it provides a more accurate description of the non-linear current-voltage relationship. The I-V relationship expression of the bypass diode sub-model (also known as the improved bypass diode model) is:

[0146]

[0147] Among them, I bd Is the current of the bypass diode, I sbd Is the saturation current of the bypass diode, V sm Is the voltage of the photovoltaic module, R bd Is the series resistance of the bypass diode, that is, the equivalent resistance, η bd Is the ideality factor of the bypass diode, V tbd Is the thermal voltage of the bypass diode.

[0148] V tbd The calculation formula is:

[0149]

[0150] Among them, k is the Boltzmann constant, T bd Is the temperature of the bypass diode, q is the electron charge.

[0151] By using the Lambert W function to make the formula four explicit, the following expression is obtained:

[0152]

[0153] In the embodiments of the present application, the photovoltaic cells in the photovoltaic module describe the behavior of the photovoltaic cells through the single-diode model, while the bypass diode model is used to protect the performance of the sub-module under partial shading. In actual operation, when the voltage of the photovoltaic cell becomes negative, the bypass diode will conduct, and the current of the sub-module will be dynamically adjusted according to the state of the photovoltaic cell and the state of the bypass diode. The I-V output relationship of the single-diode sub-model is as follows:

[0154]

[0155] where I is the output current of the branch formed by the photovoltaic cell and the shunt bypass diode in parallel, I ph is the photocurrent, I s is the saturation current, V is the voltage across the branch formed by the photovoltaic cell and the shunt bypass diode in parallel, I bd is the current of the bypass diode, R s is the series resistance, R h is the parallel resistance, η is the ideality factor of the diode, V t is the thermal voltage of the diode, N s is the number of cells in the photovoltaic cell, I sm is the current of the photovoltaic cell.

[0156] In some embodiments, the model parameters in the photovoltaic module model are obtained by fitting based on the current values and voltage values of the photovoltaic module under multiple preset illumination conditions. Different preset illumination conditions simulate various light intensity and angle changes faced by the photovoltaic module in actual applications. The model parameters obtained in this way can accurately reflect the true operating characteristics of the bypass diode under different illumination environments. Specifically, the bypass diode sub-model and the single-diode sub-model in the photovoltaic module model are obtained by fitting based on the current values and voltage values of the photovoltaic module under multiple preset illumination conditions.

[0157] In a possible implementation, the first parameter in the bypass diode sub-model and the second parameter in the single-diode sub-model are determined by the following methods a1 - a3:

[0158] a1: Based on the current values and voltage values of the photovoltaic module under multiple preset illumination conditions, determine the current value and voltage value when the bypass diode in the photovoltaic module is in the conducting state.

[0159] Among them, the equivalent bypass diode being in the conducting state can be understood as that the photovoltaic cell in parallel with the equivalent bypass diode is not irradiated. When it is completely blocked, the equivalent bypass diode is activated.

[0160] Optionally, based on the current value and voltage value of the photovoltaic module under uniform illumination (i.e., all photovoltaic cells in the photovoltaic module are irradiated), and the current value and voltage value of the photovoltaic module under partial shading (i.e., at least one photovoltaic cell in the photovoltaic module is not irradiated and is completely blocked), determine the current value and voltage value when the bypass diode in the photovoltaic module is in the conducting state.

[0161] Exemplarily, subtract the current value and voltage value of the photovoltaic module under uniform illumination (i.e., all photovoltaic cells in the photovoltaic module are irradiated) from the current value and voltage value when one photovoltaic cell in the photovoltaic module is not irradiated and is in a completely blocked condition, to obtain the current value and voltage value when the bypass diode in the photovoltaic module is in the conducting state.

[0162] In the embodiment of the present application, in the above a1, a non-invasive parameter estimation method is adopted, and through experimental measurement and mathematical fitting, the first parameter in the bypass diode submodel is accurately estimated.

[0163] Taking the photovoltaic module including three photovoltaic cells, with each photovoltaic cell being connected in parallel with a bypass diode as an example, Figure 5 is the equivalent circuit diagram of the photovoltaic module under uniform illumination conditions. Under uniform illumination conditions, all photovoltaic cells in the photovoltaic module operate normally. Figure 6 is the equivalent circuit diagram of the photovoltaic module under partial shading conditions. In Figure 6 , photovoltaic cell 1 and photovoltaic cell 3 operate normally, the bypass diodes corresponding to photovoltaic cell 1 and photovoltaic cell 2 are in the non-conducting state, photovoltaic cell 2 is completely blocked, and the bypass diode corresponding to photovoltaic cell 2 is in the conducting state. Figure 7 is the equivalent circuit diagram when one photovoltaic cell in the photovoltaic module is short-circuited. In Figure 7 , photovoltaic cell 1 and photovoltaic cell 3 operate normally, photovoltaic cell 2 is in a short-circuit state, and the bypass diodes corresponding to photovoltaic cell 1 and photovoltaic cell 2 are in the non-conducting state.

[0164] Measure the I-V curve of the entire photovoltaic module under uniform illumination conditions (all photovoltaic cells operate normally), such as Figure 8 the curve A in. Due to the existence of the measuring device and connecting wires, the influence of voltage drop needs to be considered, and the corrected voltage formula is calculated as follows:

[0165] V A =V Am -I A ×R wire (Formula VIII)

[0166] Among them, V Ais the corrected voltage value of the photovoltaic module under uniform illumination conditions, I A is the current value of the photovoltaic module under uniform illumination conditions, V Am is the measured voltage value of the photovoltaic module under uniform illumination conditions, R wire is the connecting wire resistance.

[0167] Under partial shading conditions (completely shading a single photovoltaic cell in the photovoltaic module to make its current close to zero), re-measure the I-V curve of the photovoltaic module, as shown in Figure 8 Curve B in. Repeat the above process of correcting voltage data to obtain V B . Curve B corresponds to the I-V relationship when the bypass diode is activated.

[0168] Based on Curve A and Curve B, calculate the equivalent I-V Curve C, that is, Figure 8 Curve C in. Curve C represents the I-V characteristics (which can also be understood as the correspondence between I and V) of the combination of other photovoltaic cells and bypass diodes except photovoltaic cell 2. The calculation formula is as follows:

[0169]

[0170] where, V A is the voltage value of the photovoltaic module corresponding to Curve C, I A is the current value of the photovoltaic module corresponding to Curve C, N m is the number of photovoltaic cells in the photovoltaic module.

[0171] Calculate the current value and voltage value when the bypass diode is in the conducting state. The calculation formula is as follows:

[0172]

[0173] where, I bd and V bd are the current value and voltage value of the bypass diode respectively, V C is the voltage value of the photovoltaic module corresponding to Curve C, V B is the voltage value of the photovoltaic module corresponding to Curve B, I B is the current value of the photovoltaic module corresponding to Curve B.

[0174] a2: Based on the current value and voltage value when the bypass diode is in the conducting state, determine the first parameter in the bypass diode sub-model.

[0175] In a possible implementation, the first parameter in the bypass diode sub-model includes but is not limited to saturation current, ideality factor, equivalent resistance, etc.

[0176] Optionally, in the above a2, the first parameter in the bypass diode sub-model is determined as follows:

[0177] Based on the current value and voltage value when the bypass diode is in the conduction state, the trust-region algorithm is used for non-linear least squares fitting to obtain the first parameter.

[0178] Wherein, the first parameter enables the error of the bypass diode sub-model simulating the operation of the bypass diode to meet the preset condition.

[0179] Exemplarily, after obtaining the current value and voltage value when the bypass diode is in the conduction state, that is, the I-V curve of the bypass diode, the first parameters (saturation current, ideality factor, equivalent resistance) in the bypass diode sub-model are estimated. The initial values are estimated according to the diode parameters at a given temperature, and the formula is as follows:

[0180]

[0181] Where, I sbd,0 is the initial saturation current, η bd,0 is the initial ideality factor, R bd,0 is the initial equivalent resistance, I r is the reverse current of the diode, V f is the forward voltage of the diode, I f is the forward current of the diode, V tbd,0 is the thermal voltage.

[0182] Use non-linear least squares method to fit the data near the maximum power point in the bypass diode I-V curve to calculate the first parameters saturation current I sbd , ideality factor η bd and equivalent resistance R bd in the bypass diode sub-model. In order to improve the accuracy of parameter estimation, the trust-region algorithm (Trust-Region Algorithm, TRA) is used for non-linear least squares fitting. Through the trust-region algorithm, the first parameters in the bypass diode sub-model are optimized within the set parameter range, so that the error between the I-V curve predicted by the bypass diode sub-model and the I-V curve extracted from the experiment is minimized. When the error after parameter update is within the preset error range or reaches the preset number of iterations, the iteration stops, and the first parameters in the bypass diode sub-model are determined. The embodiments of the present application do not limit the preset error range and the preset number of iterations, which can be limited according to the actual situation.

[0183] In the embodiments of the present application, when the trust region algorithm performs non-linear least squares fitting, a "trust region" is constructed around the current parameter estimate in each iteration. Within this region, by approximating the objective function (i.e., the error function between the simulated value and the actual value of the bypass diode sub-model) as a quadratic function, and then solving the minimum value of this quadratic function under the constraints of the trust region, the update direction and step size of the parameters can be obtained. In this way, the information in the current values and voltage values under multiple preset illumination conditions can be fully utilized to continuously adjust the first parameter in the bypass diode sub-model, so that the finally obtained first parameter can accurately match the operating characteristics of the bypass diode under various illumination conditions, greatly improving the accuracy of the bypass diode sub-model in simulating the bypass diode.

[0184] a3: Based on the current value and voltage value when the bypass diode is in the conducting state, determine the second parameter in the single-diode sub-model.

[0185] Among them, the second parameter in the single-diode sub-model includes, but is not limited to, the series resistance, parallel resistance, and photocurrent of the photovoltaic cell.

[0186] Optionally, in the above a3, the second parameter in the single-diode sub-model is determined in the following manner:

[0187] First, based on the current value and voltage value when the bypass diode is in the conducting state, and the current value and voltage value of the photovoltaic module under partial shading conditions, determine the current value and voltage value when the photovoltaic cell is in the normal operating state.

[0188] Exemplarily, when one photovoltaic cell in the photovoltaic module is shaded and the other photovoltaic cells are operating normally, the current value and voltage value of the photovoltaic module are respectively used as the first voltage value. The difference between the first voltage value and the voltage value when the bypass diode is in the conducting state is obtained to get the voltage value when the photovoltaic cell is in the normal operating state. In the photovoltaic module, each photovoltaic cell is connected in series with a bypass diode, and a branch is formed between each photovoltaic cell and its corresponding bypass diode. The currents in each branch are the same. Therefore, the current value when the bypass diode in the photovoltaic module is in the conducting state can be used as the current value of the photovoltaic cell in the normal operating state in the photovoltaic module.

[0189] Then, based on the current value and voltage value of the photovoltaic cell, determine the second parameter of the single-diode sub-model.

[0190] Exemplarily, similar to the method for determining the first parameter in the bypass diode sub-model in the above a2, the second parameter in the single-diode sub-model can also be obtained by the non-linear least squares method, which will not be elaborated here.

[0191] Figure 9 This is a comparison chart of the results of verifying the bypass diode sub-model through simulation experiments. InFigure 9 The I-V curve of the bypass diode obtained based on the bypass diode sub-model (improved model) provided in the embodiments of the present application, the I-V curve of the bypass diode obtained based on the bypass diode model (traditional model) in the related art, and the I-V curve of the bypass diode obtained based on the measured data are shown. As Figure 9 shown, the bypass diode sub-model provided in the embodiments of the present application has higher fitting accuracy than the bypass diode model in the related art. Especially under low-current conditions, the bypass diode sub-model provided in the embodiments of the present application can more accurately reflect the working characteristics of the bypass diode.

[0192] Table 1 shows the I-V curve reproduction accuracy of the bypass diode under different irradiance conditions. The mean absolute percentage error (MAPE) is used as the evaluation criterion, and each of the three branches (also called photovoltaic sub-modules) is analyzed. Each branch includes a photovoltaic cell and a parallel-connected bypass diode. The performance of the traditional model (EM) in the related art and the bypass diode sub-model (PM) proposed in the embodiments of the present application is compared. It can be seen from Table 1 that the MAPE value of the bypass diode sub-model (PM) proposed in the embodiments of the present application in the I-V curve reproduction is significantly lower than that of the traditional model (EM) in the related art. This indicates that the bypass diode sub-model proposed in the embodiments of the present application is more accurate than the traditional model (EM) under various irradiances, especially the error reduction is more significant under lower irradiances. By integrating the improved bypass diode sub-model into the photovoltaic module model, the performance of the photovoltaic module under complex lighting conditions can be more accurately simulated. The improved bypass diode sub-model can more precisely fit the actual I-V curve. Its non-invasive parameter estimation method does not require opening the junction box of the module, avoiding damage to the photovoltaic module. This method is simple and has strong robustness, and can be widely applied to actual photovoltaic systems, especially showing significant advantages under partial shadow conditions.

[0193] Table 1

[0194]

[0195] In a possible implementation manner, when the photovoltaic module is in a preset fault, the change rules of the internal current and voltage of the photovoltaic module are different from those when the photovoltaic module is in a normal operating state. Based on the electrical data under different faults (such as dust coverage, local short circuit, shadow occlusion, and component aging, etc.), for example, obtain the I-V characteristic curve of the photovoltaic module under the fault, and analyze the influence of different faults on the second parameter in the single-diode sub-model of the photovoltaic module. For example, dust coverage is manifested as a decrease in the photocurrent I ph decrease, resulting in a decrease in the overall current of the curve; the local short-circuit fault is manifested as a parallel resistance (also called bypass resistance) R hSignificantly reduced, leading to the collapse of the curve in the low-voltage region; the shielding effect is manifested as an increase in the series resistance Rs, and the curve shows multi-step distortion. The aging failure is manifested as an increase in Rs and a decrease in I ph decrease, and the fill factor FF significantly decays. It can be seen that the second parameter in the single-diode submodel when the photovoltaic module is in a fault state is not the same as the corresponding second parameter when it is in a normal operating state. Therefore, in the case where the photovoltaic module is in a fault state, the second parameter in the single-diode submodel is optimized so that the single-diode submodel can more accurately describe the operating characteristics of the photovoltaic module under fault conditions.

[0196] In the embodiment of the present application, based on the electrical data in these fault states and using the particle swarm optimization algorithm to correct the second parameter of the single-diode submodel, the single-diode submodel can accurately describe the electrical characteristics of the photovoltaic cell under fault conditions.

[0197] The method provided in the embodiment of the present application further includes the following b1-b2:

[0198] b1: Obtain the current value and voltage value when the photovoltaic module is in a preset fault.

[0199] Here, the preset fault can be a short-circuit fault, a shielding fault, a degradation fault, a dust accumulation fault, etc.

[0200] Optionally, the current value and voltage value when the photovoltaic module is in a preset fault are normalized data to eliminate the interference of the environment (such as light intensity, temperature).

[0201] Exemplarily, the calculation formulas for the current value and voltage value when the photovoltaic module is in a preset fault are as follows:

[0202]

[0203] Among them, I norm is the current value when the photovoltaic module is in a preset fault, that is, the normalized current value, I is the current value before normalization; V norm is the voltage value when the photovoltaic module is in a preset fault, that is, the normalized current value, V is the voltage value before normalization; G is the actually measured light intensity; T cell is the actually measured cell temperature; β is the voltage temperature coefficient (typical value -0.3% / °C); G STC = 1000W / m 2 , T STC = 25°C.

[0204] b2: Based on the current value and voltage value when the photovoltaic module is in a preset fault, the particle swarm optimization (PSO) algorithm is used to optimize the second parameter in the single-diode sub-model to obtain the second parameter under the preset fault.

[0205] Optionally, during the process of optimizing the second parameter using the particle swarm optimization algorithm, the mean squared error (MSE) is used as the fitness function to quantify the deviation between the measured data and the predicted data of the single-diode sub-model and establish an inversion equation. Exemplarily, the fitness function is expressed as follows:

[0206]

[0207] where MSE is the mean squared error between the measured current value of the photovoltaic cell and the predicted current value of the photovoltaic cell using the single-diode sub-model, N is the amount of data; I meas is the actual current value when the voltage of the photovoltaic cell is V i ; I model is the predicted current value using the single-diode sub-model when the voltage of the photovoltaic cell is V i ; R s , R h , I ph are the second parameters in the single-diode sub-model; R s is the series resistance, R h is the parallel resistance, I ph is the photocurrent.

[0208] Randomly generate a particle swarm in the parameter space, set the search range, and dynamically adjust the particle positions according to the individual optimum (p best ) and the global optimum (g best ):

[0209]

[0210] where is the velocity of the i-th particle at the (k + 1)-th iteration, is the position of the i-th particle at the (k + 1)-th iteration, the inertial weight w decays with the number of iterations, the learning factors c1 = c2 = 2, and r1, r2 are random numbers.

[0211] Termination condition: Stop when MSE < 2% or when the preset number of iterations (e.g., 100 times) is reached, and output the optimized second parameter (i.e., the second parameter under the preset fault) R s * , R h * , I ph *。

[0212] Based on the second parameter under a preset fault, calculate the I-V curve of the single-diode submodel and the actual I-V curve, and calculate the goodness of fit:

[0213]

[0214] If the goodness of fit > 95%, determine that the inversion result is valid, that is, it is effective to optimize the second parameter using the particle swarm optimization algorithm.

[0215] For high-confidence results (MSE < 3%), update the second parameter in the single-diode submodel when the photovoltaic module is in the preset state:

[0216]

[0217] Among them, R s,new is the second parameter under the preset fault, R s,old is the second parameter of the single-diode submodel when the photovoltaic module is in the normal operating state, and α is the forgetting factor (usually taken as 0.9), which is used to balance historical data and new results. R s,new can be used for the life prediction of the photovoltaic module in the fourth part of the subsequent embodiments.

[0218] In another possible implementation manner, the method provided in the embodiment of the present application further includes: judging the severity of the fault of the photovoltaic cell under the preset fault according to the second parameter under the preset fault.

[0219] Exemplarily, the second parameter includes the series resistance in the photovoltaic cell. If R s * > 2R s , judge that the photovoltaic cell is in an emergency fault. In this case, it is necessary to immediately stop the machine and replace the photovoltaic cell. If 1.5R s < R s * ≤ 2R s , judge that the photovoltaic cell needs to increase the inspection frequency and record the deterioration trend. Among them, R s is the second parameter of the photovoltaic module in the normal operating state.

[0220] The above is the second part of the embodiment of the present application. Next, in combination with Figures 10 to 14 , introduce the fault prediction method provided in the embodiment of the present application, aiming to introduce the specific implementation manner of constructing the fault prediction model.

[0221] In some embodiments, the training data used to train the fault prediction model can be obtained through the following c1 - c2:

[0222] c1: Obtain a plurality of initial data.

[0223] Among them, multiple initial data correspond to multiple types of faults. Fault types such as dust accumulation fault, shielding fault, degradation fault, short - circuit fault, etc. Multiple initial data include the initial data corresponding to the dust accumulation fault, the initial data corresponding to the shielding fault, the initial data corresponding to the degradation fault, the initial data corresponding to the short - circuit fault, etc.

[0224] In one possible implementation, the initial data can be electrical data such as the current value and voltage value of the photovoltaic module. The initial data can also be the time - domain characteristics and frequency - domain characteristics of the electrical data of the photovoltaic module. Among them, the time - domain characteristics include but are not limited to the average value, standard deviation, root mean square, variance value, kurtosis value, skewness value, square root of amplitude, crest factor, impulse factor, margin factor, form factor, and peak value. The frequency - domain characteristics include but are not limited to the average frequency, root variance frequency, and root mean square frequency, as shown in Table 2. In the embodiments of the present application, the initial data includes the time - domain characteristics and frequency - domain characteristics of the current value of the photovoltaic module.

[0225] Table 2

[0226]

[0227] In one possible implementation, the initial data can be obtained by simulating the photovoltaic module model in the first part of the above - mentioned embodiments. In this way, the scarcity problem of the actual electrical data of the photovoltaic module is effectively alleviated.

[0228] In another possible implementation, the initial data can also be the electrical data during the actual operation of the photovoltaic module. The data during the actual operation can more truly reflect the characteristics and behaviors of the photovoltaic module under actual working conditions, and can accurately evaluate the actual performance of the photovoltaic module and analyze the actual fault conditions, etc.

[0229] Of course, the training data of the fault prediction model can also include the electrical data during the actual operation of the photovoltaic module, and the electrical data simulated by using the photovoltaic module model, such as the current value, voltage value, etc. In the embodiments of the present application, among the multiple training data of the fault prediction model, 80% is obtained through the above - mentioned photovoltaic module model simulation, and 20% comes from the electrical data during the actual operation of the photovoltaic module. Table 3 is the initial data used in the embodiments of the present application.

[0230] Table 3

[0231]

[0232] In one possible implementation, the initial data is the data after being standardized. Exemplarily, the mean of the initial data obtained after being standardized is 0, and the variance is 1. The standardization process uses the Z - score normalization method, and the formula is as follows:

[0233]

[0234] Among them, Z is the initial data, and x i represents the initial data before normalization processing, μ is the mean of the initial data before normalization processing, and σ is the standard deviation of the initial data before normalization processing.

[0235] c2: Based on the information gain (Information Gain, IG) of each initial data, multiple initial data are screened to obtain multiple training data.

[0236] In a possible implementation manner, the information gain of the initial data can characterize how much information the initial data provides for fault prediction. Screening the initial data based on the information gain can retain the data that carries key information and is helpful for accurately judging the fault type, and remove the data with low information content and little effect on fault classification. In this way, on the one hand, since the training data obtained after screening contains the key data for distinguishing different fault types, when using these data to train the model, the characteristics of various fault types can be learned more accurately, enhancing the accuracy of the fault prediction model in predicting faults. On the other hand, screening the initial data based on the information gain reduces the dimension of the training data, enabling the fault prediction model to converge more quickly and saving training time.

[0237] In the embodiments of the present application, the information gain of the initial data is determined by the following formula:

[0238] The calculation formula of the entropy of the data set is:

[0239]

[0240] Among them, E(S) is the entropy of the data set S composed of multiple initial data, C is the number of fault types, and p i is the probability of the i-th type of fault.

[0241] The information gain of the initial data A is:

[0242]

[0243] Among them, IG(S,A) is the information gain of the initial data A, Values(A) are all possible values of the initial A, and S v is the subset of the initial data A when taking the value v, and |S v | and |S| are the sizes of the subset and the total set respectively.

[0244] Figure 10 shows the information gain of the initial data. In Figure 10 the initial data includes the frequency domain characteristics and time domain characteristics of the current value of the photovoltaic module. From Figure 10It can be seen that the information gain (IG score) of feature Vf 12 is the largest, and the information gain (IG score) of feature If 15 is the smallest.

[0245] In a possible implementation, c2 screens multiple initial data in the following manner:

[0246] First, based on the information gain of each initial data, the information gain threshold is determined using the sliding window test algorithm.

[0247] In the embodiments of the present application, the information gains of each initial data are sorted to obtain {IG1, IG2, …, IG n}, and a sliding window W(t) with a length of w is defined, which contains w consecutive scores starting from the t-th IG score:

[0248] W(t) = {IG t , IG t+1 , …, IG t+w-1} (Formula XXI)

[0249] All possible sliding windows are traversed to find the window W(t * ) that optimizes the classification performance, and the minimum IG score within this window is defined as the information gain threshold:

[0250]

[0251] where Optimal Threshold is the information gain threshold.

[0252] Then, based on the information gain threshold and the information gain of each initial data, multiple training data are obtained.

[0253] In the embodiments of the present application, the initial data with an information gain greater than the information gain threshold are used as the multiple training data.

[0254] In some embodiments, the training data of the fault prediction model are the data after principal component analysis. For example, the initial data screened by the above information gain are subjected to principal component analysis, and the data after principal component analysis are used as the training data.

[0255] In a possible implementation, obtaining the multiple training data based on the information gain threshold and the information gain of each initial data includes the following content:

[0256] First, based on the information gain threshold and the information gain of each initial data, initial training data are obtained.

[0257] Then, using the principal component analysis method and the initial training data, the multiple training data are obtained.

[0258] By combining information gain and principal component analysis techniques and introducing a sliding window test algorithm, while retaining key information, the feature dimension can be significantly reduced, the computational efficiency can be improved, and support can be provided for subsequent training of the fault prediction model.

[0259] In the embodiment of the present application, the principal component analysis method is used to further reduce the dimension of the initial data after information gain screening, and the principal components with a cumulative explained variance greater than 90% are retained.

[0260] Among them, the covariance matrix of the initial data after information gain screening is expressed as follows:

[0261]

[0262] Among them, X i is the i-th initial data after information gain screening, μ is the mean vector of the initial data after information gain screening, and n is the number of initial data after information gain screening;

[0263] Perform eigenvalue decomposition on the covariance matrix Σ to obtain the eigenvector V j (i.e., the principal component) and the eigenvalue λ j , sort them in descending order of eigenvalues, and select the first k principal components that satisfy a cumulative explained variance greater than 90%. The cumulative explained variance calculation formula is as follows:

[0264]

[0265] Among them, CEV(k) is the cumulative explained variance corresponding to the first k principal components, and m is the total number of initial data after information gain screening.

[0266] Use the matrix composed of the selected principal components to convert the initial data after information gain screening into the data after dimension reduction:

[0267] Y = XW (Formula XXV)

[0268] Among them, Y is the data after dimension reduction, X is the initial data after information gain screening, and W is the matrix composed of the selected principal components.

[0269] In the embodiments of the present application, the dimension-reduced data is used as training data to train multiple classifiers, and their performance in fault detection is evaluated. The initial data after information gain screening is obtained based on sliding window screening. Different starting points and lengths of the sliding window lead to different classifier performances. Three classifiers, namely ClassificationAnd Regression Tree (CART), K-Nearest Neighbor (KNN), and RandomForest (RF), are respectively tested, and their Prediction Accuracy (PACC) and Learning Computation Time (LCT) are compared. Figure 11 Table 4 shows the classifier performances under different information gain thresholds, verifying the effectiveness of the sliding window test algorithm in selecting the initial data. By selecting an appropriate sliding window (such as the sliding window interval being [1.3, 1.5]), the accuracy of the classifier can be significantly improved and the computation time can be reduced, thereby optimizing the overall performance of fault detection.

[0270] Table 4

[0271]

[0272] In the embodiments of the present application, the fault prediction model uses a three-layer Full ConnectNeural Network (FCNN), with 50 neurons in each hidden layer, and adopts Dropout regularization to reduce the risk of overfitting. The labeled fault data is used as the input to train the neural network with the Dropout mechanism. The Dropout mechanism prevents the model from overfitting by randomly discarding neuron connections and ensures the classification accuracy. In the Dropout mechanism, the output of the forward propagation is:

[0273] z (l) =σ(W (l) ·(r (l) ⊙h (l-l) )+b (l) (Formula 26)

[0274] where z (l) is the output of the forward propagation, σ is the activation function, W (l) is the weight matrix from the (l - 1)-th layer to the l-th layer, r (l) is the mask vector sampled from the Bernoulli distribution, h (l-1) is the output of the previous layer, ⊙ represents the element-wise multiplication operation, and b (l) is the bias vector from the (l - 1)-th layer to the l-th layer.

[0275] The Dropout ratio is usually a manually set hyperparameter that represents the probability of discarding neurons in each layer. To optimize the Dropout ratio, the Concrete Dropout mechanism is adopted, which automatically adjusts the Dropout ratio p of each layer:

[0276]

[0277] where is the adjusted Dropout ratio, p is the Dropout ratio, G is a random variable from the Gumbel distribution, and τ is the temperature parameter.

[0278] The Dropout mechanism can dynamically adjust the p of each layer and optimize the parameters through the Gumbel-Softmax distribution, making the fault prediction model have an adaptive ability.

[0279] Table 5 compares the accuracy and running time of different classification models, demonstrating the superiority of the neural network model proposed in the embodiments of this application in the fault classification task. In particular, Concrete Dropout performs best in terms of classification accuracy and efficiency. Figure 12 The confusion matrix in shows the specific performance of the neural network model in the actual application of classifying categories such as aging, shadow occlusion, ash layer coverage, and local short circuits. It can be seen that the Concrete Dropout neural network model has good results in photovoltaic fault classification, especially in the accurate classification of short circuit faults.

[0280] Table 5

[0281]

[0282] In some embodiments, the fault prediction model is a neural network optimized based on a pruning algorithm. In this way, redundant connections and neurons in the original neural network are removed, reducing unnecessary computational complexity and improving the running efficiency of the fault prediction model.

[0283] Figure 13 shows the fault diagnosis accuracy of the neural network under different pruning ratios. When the pruning ratio is 50%, the accuracy performance is still very good. When the pruning ratio is further increased, the rate of decrease in accuracy accelerates. The neural network is pruned and optimized through the Lottery Ticket Hypothesis (LTH), and the weights of the trained neural network are pruned to generate a pruning mask for neurons with less contribution to classification:

[0284]

[0285] where M i,jis the pruning mask, W i,j is the weight of the neural network, i and j are the index of the neurons in the previous layer and the current layer respectively.

[0286] The weights of the neural network after pruning are updated as follows:

[0287] W t+1 =M⊙W t (Formula 29)

[0288] Among them, W t+1 is the weight of the neural network after pruning, W t is the weight of the neural network before pruning, and M is the pruning mask.

[0289] Moderate pruning can significantly reduce the complexity and computational cost of the neural network, while having little impact on classification accuracy. This shows that the pruned neural network can still maintain good classification performance and is very suitable for embedded systems with limited resources.

[0290] In some embodiments, different faults may have varying degrees of impact on photovoltaic modules. Different faults are assigned different weights based on fault severity, number of occurrences (or frequency), and detection difficulty. This allows the fault prediction model to prioritize the prediction accuracy of high-weighted faults (e.g., those with high severity or frequent occurrence).

[0291] In a possible implementation, the first loss function of the fault prediction model is obtained by weighted summation of multiple second loss functions.

[0292] Among them, the second loss function corresponds one-to-one to the preset fault, and the second loss function is used to indicate the prediction error of the fault prediction model for the preset fault; the weights of multiple second loss functions are determined based on at least one of the severity, number of occurrences, and difficulty of detection of their corresponding faults.

[0293] For example, if a fault causes a huge loss to a photovoltaic power station, the weight of the corresponding first loss function can be significantly higher than the weight of the first loss function corresponding to a low-impact fault, thereby guiding the fault prediction model to pay more attention to the prediction ability of such faults and improving the targeted nature of fault prediction. For another example, assigning a higher weight to the second loss function corresponding to faults that are difficult to detect allows the fault prediction model to mine the subtle features of these difficult-to-detect faults during training, thereby improving the detection ability of complex faults. For another example, assigning a higher weight to the second loss function corresponding to frequent faults allows the fault prediction model to focus on learning the features of high-frequency faults during training, thereby improving the prediction accuracy of high-frequency faults.

[0294] In the embodiments of the present application, a Risk Priority Number (RPN) mechanism is introduced into the fault prediction model. According to the severity, occurrence times (or occurrence frequency), and detection difficulty of each fault, the risk priority (RPN value) is calculated as follows:

[0295] RPN = S × O × D (Formula 30)

[0296] Where RPN is the risk priority, S is the severity of the fault, O is the occurrence times (or occurrence frequency) of the fault, and D is the detection difficulty of the fault. The larger the value of RPN, the higher the risk of this type of fault, and it requires priority attention during processing.

[0297] Table 6 lists the risk priorities of different faults in the photovoltaic module, which is used to evaluate the risk levels of different faults and helps the fault prediction model to give priority attention to high-risk faults during classification to reduce potential safety hazards.

[0298] Table 6

[0299]

[0300] To optimize the fault classification results, it is necessary to perform weighted processing on different types of faults in combination with RPN, so that faults with higher RPN values occupy higher weights in the learning and classification processes of the model. When training the classification model, the loss of each type of fault is weighted according to its RPN value. Suppose there are K types of preset faults, and the RPN value of each type of fault is RPN k , and the corresponding sample classification loss is L k . The first loss function of the fault prediction model can be defined as:

[0301]

[0302] Where L weighted is the first loss function, w k is the weight of the preset fault k, which is obtained by normalizing its corresponding RPN value:

[0303]

[0304] In the evaluation stage of the fault prediction model, the classification effect of the model on high-risk faults can be measured by the RPN Weighted Accuracy (RWA). Its calculation formula is as follows:

[0305]

[0306] Where RWA is used to characterize the classification effect of the model, A k is the classification accuracy of the model for the preset fault k; RPNk is the risk priority of the preset fault.

[0307] Through the RPN, it is possible to identify which faults have higher priorities, so that more weights can be assigned to these high-priority faults during the classification process of the fault prediction model, in order to improve the overall robustness and practicality of the fault prediction model. The RWA is used to measure the classification effect of the fault prediction model on high-risk faults, ensuring that in actual applications, the fault prediction model can give priority to handling high-risk faults.

[0308] Figure 14 Schematic diagram for determining training data for feature extraction based on information gain and principal component analysis. In Figure 14 , first, through data preprocessing (such as the Z-score normalization method), multiple initial data are obtained; then, the information gain of each initial data is determined; next, the information gain threshold is determined based on the sliding window test method, and the initial data after information gain screening are determined based on the information gain threshold; finally, using the principal component analysis method, the principal component analysis of the initial data after information gain screening is performed to obtain the training data.

[0309] The above is the third part of the embodiments of the present application. Next, in combination with Figure 15 , Figure 16 , the fault prediction method provided by the embodiments of the present application is introduced, aiming to introduce the specific implementation manner of predicting the lifespan of the target photovoltaic module.

[0310] In some embodiments, the photovoltaic module model includes a single diode sub-model. The single diode sub-model is used to simulate the operation of the photovoltaic cells in the photovoltaic module. The construction of the single diode sub-model has been described in the second part of the above embodiments and will not be elaborated here.

[0311] The method provided by the embodiments of the present application further includes the following steps d1 - d4:

[0312] d1: Obtain the environmental data of the target photovoltaic module.

[0313] In a possible implementation manner, the environmental data includes but is not limited to meteorological parameters such as temperature and humidity.

[0314] d2: Determine the operating temperature of the target photovoltaic module based on the environmental data.

[0315] In a possible implementation manner, based on the environmental data and a pre-constructed photovoltaic module temperature model, the operating temperature of the target photovoltaic module is determined. Among them, the photovoltaic module temperature model is expressed as follows:

[0316]

[0317] Among them, T Dis the operating temperature of the target photovoltaic module, T amb is the ambient temperature, E is the solar radiation intensity, WS is the wind speed, ΔT is the temperature difference inside the component, E0 is the reference radiation intensity, and a and b are empirical parameters.

[0318] In this photovoltaic module temperature model, the solar component, that is, the photovoltaic module, will generate self-heating effect due to solar radiation during actual operation. Therefore, the actual temperature of the photovoltaic module is usually higher than the ambient temperature. Using this photovoltaic module temperature model, the operating temperature of the photovoltaic module in the corresponding installation area can be calculated based on actual meteorological data, so as to more accurately predict the impact of temperature on the corrosion degradation of the photovoltaic module.

[0319] In the embodiment of the present application, based on environmental data, the operating temperature of the target photovoltaic module is determined, considering the self-heating effect generated by the solar component due to solar radiation during actual operation. Compared with directly using the ambient temperature as the operating temperature of the photovoltaic module, the operating temperature of the target photovoltaic module determined in the embodiment of the present application is more accurate and can more truly reflect the thermal state of the photovoltaic module during actual operation, providing a basis for accurately determining the degradation rate of the target photovoltaic module and predicting the lifespan.

[0320] d3: Based on the operating temperature and the second parameter of the single-diode submodel, determine the degradation rate of the target photovoltaic module.

[0321] In a possible implementation, in the above d3, the degradation rate of the target photovoltaic module is determined through a corrosion degradation rate model, and the formula of the corrosion degradation rate model is as follows:

[0322]

[0323] where, R D is the degradation rate of the target photovoltaic module, T is the operating temperature of the target photovoltaic module, RH is the relative humidity, E A is the activation energy of the sealing material, k B is the Boltzmann constant, n is the humidity coefficient, A is an empirical constant; R s,new is the second parameter in the single-diode submodel. It should be noted that when the photovoltaic module is in a normal operating state, R s,new is the second parameter in the diode submodel. When the photovoltaic module is in a fault state, R s,new is the second parameter under this fault.

[0324] In the embodiment of the present application, the determination process of the above formula (35) includes: obtaining the meteorological data of the installation area where the photovoltaic module is located from the meteorological database, mainly including the environmental temperature T, relative humidity RH, etc. Integrate these meteorological data with the laboratory accelerated test data, and calculate the acceleration effect of the environment on corrosion degradation by substituting the dynamic changes of temperature and humidity into the Peck formula to obtain the corrosion degradation rate model.

[0325] d4: Predict the life of the target photovoltaic module based on the degradation rate.

[0326] In a possible implementation manner, based on the degradation rate, a life prediction model of the solar component under different environmental conditions is constructed. Use the S-type function to fit the power attenuation, and the fitting curve is as Figure 15 shown, simulate the power attenuation curve of the component at different positions to obtain the life prediction model. The above d4 predicts the life of the target photovoltaic module through the following life prediction model:

[0327]

[0328] where L is the predicted life of the target photovoltaic module, and R D (T, RH) is the degradation rate, and the degradation rate can be obtained through the above formula (35) and formula (34).

[0329] In another possible implementation manner, in the above d4, the life of the target photovoltaic module is determined in the following manner: based on the degradation rate, determine the moment when the power attenuation of the target photovoltaic module reaches the preset power attenuation threshold, and determine the life of the target photovoltaic module based on this moment.

[0330] Optionally, based on the degradation rate and the pre-constructed corrosion degradation model, determine the moment when the power attenuation of the target photovoltaic module reaches the preset power attenuation threshold, and determine the life of the target photovoltaic module based on this moment.

[0331] The following is an exemplary introduction to the construction process of the corrosion degradation model. Conduct a standard Damp-Heat (DH) test on the photovoltaic module, place the photovoltaic module in an environment with a temperature of 85°C and a relative humidity of 85%, and test the electrical performance of the photovoltaic module under accelerated degradation conditions, including the changes in short-circuit current (Isc), open-circuit voltage (Voc), fill factor (FF), etc. over time. Through these test data, record the power attenuation of the photovoltaic module and determine the performance degradation rate caused by wire or busbar corrosion. Figure 16It is a graph showing the change of the I-V curve of a photovoltaic module due to the corrosion of conductive fingers under humid heat testing. As the corrosion occurs and progresses, the I-V curve begins to sink, indicating that the power output and efficiency are gradually decreasing. Based on laboratory tests, a corrosion degradation model based on physical principles is established. This corrosion degradation model describes the chemical reactions triggered after moisture penetrates into the component and the corrosion process of the conductive fingers, and uses a sigmoidal degradation function to model the power attenuation. Through the corrosion degradation model, the power attenuation of the target photovoltaic module at a certain moment is determined. The corrosion degradation model is expressed as follows:

[0332]

[0333] where ΔP(t) is the power attenuation at time t, ΔP ∞ is the saturated power attenuation after long-term exposure, ΔP(t i ) is the initial power loss, R D ∝R s,new is the degradation rate, and t i is the starting degradation time.

[0334] When the power attenuation reaches a preset attenuation threshold, it is determined that the life of the target photovoltaic module ends, that is, the performance degradation of the target photovoltaic module reaches an irreversible failure state. Exemplarily, the preset attenuation threshold is 20%, that is, when the power attenuation of the target photovoltaic module reaches 20%, it is determined that the life of the target photovoltaic module ends. Based on the above formula (37), the moment when the target photovoltaic module reaches the preset attenuation threshold can be determined, so as to determine the life of the target photovoltaic module.

[0335] Figure 17 This is a schematic diagram of the overall process of a fault prediction method provided by an embodiment of the present application. In Figure 17 , the method includes the following content:

[0336] First, based on the Schottky equation, the I-V characteristics of the bypass diode are corrected by introducing a series equivalent resistance, and a non-invasive parameter estimation method is used to fit the I-V curve. Using the nonlinear least squares method and the trust region algorithm, the parameters in the photovoltaic module model are determined.

[0337] It should be noted that the determination of the parameters in the photovoltaic module model, including the first parameter in the bypass diode sub-model and the second parameter in the single diode sub-model, has been described in the second part of the above embodiment, and will not be elaborated here.

[0338] Then, the data simulated by the photovoltaic module model is screened by the information gain algorithm and the sliding window test algorithm to obtain the initial training data. Then, the principal component analysis method is used to further process the initial training data to obtain the training data, and the fault prediction model is trained using the training data. The specific implementation manner of obtaining the training data can refer to the description in the third part of the embodiments of the present application, and will not be elaborated here.

[0339] Next, the autoencoder is used to determine whether the target photovoltaic module has a fault. In the case where the target photovoltaic module has a fault, the fault prediction model (such as the pruned neural network model) is used for fault prediction to obtain the fault type of the target photovoltaic module.

[0340] It should be noted that for how to combine the autoencoder and the fault prediction model to perform fault prediction on the target photovoltaic module, reference can be made to the description in the first part of the above embodiments, and will not be elaborated here.

[0341] Subsequently, a fault inversion equation is established using the deviation between the actually measured electrical data and the electrical data simulated by the photovoltaic module model, and the particle swarm optimization algorithm is used to optimize the second parameter in the single diode submodel.

[0342] It should be noted that the specific implementation manner of performing fault inversion on the parameters in the photovoltaic module model can refer to the description in the second part of the embodiments of the present application, and will not be elaborated here.

[0343] Finally, based on the electrical data simulated by the optimized photovoltaic module model, a photovoltaic module temperature model, a corrosion degradation rate model, and a life prediction model are established to accurately predict the life of the photovoltaic module.

[0344] Among them, the construction methods and usage methods of the photovoltaic module temperature model, the corrosion degradation rate model, and the life prediction model can refer to the fourth part of the above embodiments, and will not be elaborated here.

[0345] In the embodiments of the present application, a data-driven intelligent operation and maintenance method for a photovoltaic power station is provided. Through the data-driven model, the automated operation and maintenance and fault management of the photovoltaic power station are realized to improve the efficiency and service life of the photovoltaic power station.

[0346] The above mainly introduces the solution provided by the embodiments of the present application from the perspective of the method.

[0347] In an embodiment of the present application, a fault prediction device is further provided. This device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0348] Figure 18 It is a schematic structural diagram of a fault prediction device. As Figure 18 shown, the device includes:

[0349] A first acquisition module 1801, configured to acquire electrical data of a target photovoltaic module; the target photovoltaic module includes at least one target bypass diode;

[0350] A first prediction module 1802, configured to obtain a fault prediction result of the target photovoltaic module based on the electrical data and a fault prediction model; multiple training data of the fault prediction model are obtained by simulating a photovoltaic module based on a pre-constructed photovoltaic module model; the photovoltaic module model includes an equivalent bypass diode and an equivalent resistance connected in series with the equivalent bypass diode; the equivalent bypass diode is used to simulate the one-way conduction characteristic of the bypass diode in the photovoltaic module; the equivalent resistance is used to correct the operating characteristic of the equivalent bypass diode.

[0351] With the device provided in this embodiment, considering that in the actual operation of a photovoltaic module, the bypass diode in the photovoltaic module has an inherent resistance due to factors such as its internal structure, in the embodiment of the present application, an equivalent resistance is introduced to simulate this characteristic, so that when generating training data using the photovoltaic module model, the electrical behavior of the bypass diode under complex working conditions such as different light intensities, temperatures, and partial shading can be accurately reproduced. That is to say, the training data obtained based on the photovoltaic module model is highly consistent with the operating state of the photovoltaic module in the real scenario. In this way, the fault prediction model trained based on the training data can accurately predict the faults of the photovoltaic module, significantly improving the accuracy and reliability of fault diagnosis.

[0352] In some alternative implementation manners, in this device, the photovoltaic module model includes a bypass diode sub-model; the bypass diode sub-model is used to simulate the operation of the bypass diode in the photovoltaic module; the bypass diode sub-model is obtained based on the Schottky equation corresponding to the equivalent bypass diode and the equivalent resistance.

[0353] In some alternative implementation manners, in this device, the model parameters in the photovoltaic module model are obtained by fitting the current values and voltage values of the photovoltaic module under multiple preset irradiation conditions.

[0354] In some alternative embodiments, the photovoltaic module model further includes a single diode sub-model; the single diode sub-model is used to simulate the operation of the photovoltaic cells in the photovoltaic module; the apparatus further includes:

[0355] A first determination module, configured to determine the current value and voltage value when the bypass diode in the photovoltaic module is in a conducting state based on the current values and voltage values of the photovoltaic module under multiple preset illumination conditions;

[0356] A second determination module, configured to determine a first parameter in the bypass diode sub-model based on the current value and voltage value when the bypass diode is in a conducting state;

[0357] A third determination module, configured to determine a second parameter in the single diode sub-model based on the current value and voltage value when the bypass diode is in a conducting state.

[0358] In some alternative embodiments, the second determination module is specifically configured to perform non-linear least squares fitting using a trust region algorithm based on the current value and voltage value when the bypass diode is in a conducting state to obtain the first parameter; the first parameter enables the error of the bypass diode sub-model simulating the operation of the bypass diode to meet a preset condition.

[0359] In some alternative embodiments, the apparatus further includes:

[0360] A second acquisition module, configured to acquire the current value and voltage value of the photovoltaic module when a preset fault occurs;

[0361] An optimization module, configured to optimize the second parameter in the single diode sub-model based on the current value and voltage value of the photovoltaic module when a preset fault occurs using a particle swarm optimization algorithm to obtain the second parameter under the preset fault.

[0362] In some alternative embodiments, the first loss function of the fault prediction model in the apparatus is obtained by weighted summation of multiple second loss functions; wherein, the second loss function corresponds to a preset fault one by one, and the second loss function is used to indicate the prediction error of the fault prediction model for the preset fault; the weights of the multiple second loss functions are determined based on at least one of the severity, occurrence times, and detection difficulty of the respective corresponding faults.

[0363] In some alternative embodiments, the first prediction module 1802 is specifically configured to input the electrical data into a pre-constructed autoencoder to obtain the reconstructed data of the target photovoltaic module; based on the reconstructed data and the electrical data, determine whether the target photovoltaic module has a fault; in the case where the target photovoltaic module has a fault, based on the electrical data and the fault prediction model, obtain a fault prediction result.

[0364] In some alternative embodiments, the first prediction module 1802 is specifically configured to determine that the target photovoltaic module fails when the deviation between the reconstructed data and the electrical data is greater than a preset threshold.

[0365] In some alternative embodiments, the fault prediction model is a neural network optimized based on a pruning algorithm; the first prediction module 1802 is specifically configured to, when the target photovoltaic module fails, input the encoded data corresponding to the electrical data into the fault prediction model to obtain a fault prediction result; the encoded data corresponding to the electrical data is obtained by inputting the electrical data into the encoder in the autoencoder.

[0366] In some alternative embodiments, the apparatus further includes:

[0367] A third acquisition module, configured to acquire a plurality of initial data; the plurality of initial data correspond to multiple types of faults;

[0368] A screening module, configured to screen the plurality of initial data based on the information gain of each initial data to obtain a plurality of training data.

[0369] In some alternative embodiments, the screening module is specifically configured to determine an information gain threshold by using a sliding window test algorithm based on the information gain of each initial data; and obtain a plurality of training data based on the information gain threshold and the information gain of each initial data.

[0370] In some alternative embodiments, the screening module is specifically configured to obtain initial training data based on the information gain threshold and the information gain of each initial data; and obtain the plurality of training data by using a principal component analysis method and the initial training data.

[0371] In some alternative embodiments, the photovoltaic module model includes a single-diode submodel; the single-diode submodel is used to simulate the operation of photovoltaic cells in the photovoltaic module; the apparatus further includes:

[0372] A fourth acquisition module, configured to acquire environmental data of the target photovoltaic module;

[0373] A fourth determination module, configured to determine the operating temperature of the target photovoltaic module based on the environmental data;

[0374] A fifth determination module, configured to determine the degradation rate of the target photovoltaic module based on the operating temperature and the second parameter of the single-diode submodel;

[0375] A second prediction module, configured to predict the lifespan of the target photovoltaic module based on the degradation rate.

[0376] In some alternative embodiments, the second prediction module is specifically configured to determine, based on the degradation rate, the time when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold; and determine the lifespan of the target photovoltaic module based on the time.

[0377] In some alternative embodiments, the second prediction module is specifically configured to determine, based on the degradation rate, the time when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold, including:

[0378] Determining, based on the degradation rate and a pre-constructed corrosion degradation model, the time when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold.

[0379] In some alternative embodiments, the corrosion degradation model includes:

[0380]

[0381] where ΔP(t) is the power attenuation at time t, ΔP ∞ is the saturated power attenuation after long-term exposure, ΔP(t i ) is the initial power loss, R D is the degradation rate, and t i is the starting degradation time.

[0382] In some alternative embodiments, determining the degradation rate of the target photovoltaic module based on the operating temperature and the second parameter of the single-diode sub-model includes:

[0383]

[0384] where R D is the degradation rate of the target photovoltaic module, T is the operating temperature of the target photovoltaic module, RH is the relative humidity, E A is the activation energy of the sealing material, x B is the Boltzmann constant, n is the humidity coefficient, A is an empirical constant, and R s,new is the second parameter in the single-diode sub-model.

[0385] In some alternative embodiments, the device further includes:

[0386] A risk determination module, configured to determine the risk priority of the fault corresponding to each second loss function based on the severity, occurrence frequency, and detection difficulty of the fault corresponding to each second loss function, using a risk priority numbering mechanism;

[0387] A weight determination module, configured to determine the weight of each second loss function based on the risk priority of the fault corresponding to each second loss function.

[0388] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding above embodiments, and will not be repeated here.

[0389] The fault prediction device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0390] The embodiment of the present invention also provides a computer device having the above-mentioned Figure 18 shown fault prediction device.

[0391] Please refer to Figure 19 , Figure 19 is a schematic structural diagram of a computer device provided by an optional embodiment of the present invention. As shown in Figure 19 , the computer device includes: one or more processors 10, a memory 20, and an interface for connecting each component, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a set of blade servers, or a multi-processor system). Figure 19 In

[0392] , a single processor 10 is taken as an example.

[0393] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0394] The memory 20 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device and the like. In addition, the memory 20 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely disposed relative to the processor 10, and these remote memories may be connected to the computer device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0395] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, a hard disk, or a solid-state drive; the memory 20 may further include a combination of the above types of memory.

[0396] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0397] An embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processed on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above types of memory. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiment is implemented.

[0398] A part of the present invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the present invention through the operations of the computer. Those skilled in the art should understand that the forms of existence of computer program instructions in a computer-readable medium include but are not limited to source files, executable files, installation package files, etc. Correspondingly, the ways for a computer to execute computer program instructions include but are not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.

[0399] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A fault prediction method, characterized in that, The method includes: Obtaining electrical data of a target photovoltaic module; the target photovoltaic module includes at least one target bypass diode; Based on the electrical data and a fault prediction model, obtaining a fault prediction result of the target photovoltaic module; multiple training data of the fault prediction model are obtained by simulating a photovoltaic module based on a pre-constructed photovoltaic module model; the photovoltaic module model includes an equivalent bypass diode and an equivalent resistor connected in series with the equivalent bypass diode; the equivalent bypass diode is used to simulate the one-way conduction characteristic of the bypass diode in the photovoltaic module; the equivalent resistor is used to correct the operating characteristic of the equivalent bypass diode.

2. The method according to claim 1, wherein The photovoltaic module model includes a bypass diode sub-model; the bypass diode sub-model is used to simulate the operation of the bypass diode in the photovoltaic module; the bypass diode sub-model is obtained based on the Schottky equation corresponding to the equivalent bypass diode and the equivalent resistor.

3. The method according to claim 2, wherein The model parameters in the photovoltaic module model are obtained by fitting the current values and voltage values of the photovoltaic module under multiple preset irradiation conditions.

4. The method according to claim 3, wherein The photovoltaic module model further includes a single diode sub-model; the single diode sub-model is used to simulate the operation of the photovoltaic cells in the photovoltaic module; the method further includes: Based on the current values and voltage values of the photovoltaic module under multiple preset irradiation conditions, determining the current value and voltage value when the bypass diode in the photovoltaic module is in a conducting state; Based on the current value and voltage value when the bypass diode is in a conducting state, determining a first parameter in the bypass diode sub-model; Based on the current value and voltage value when the bypass diode is in a conducting state, determining a second parameter in the single diode sub-model.

5. The method according to claim 4, characterized in that The determining the first parameter in the bypass diode sub-model based on the current value and voltage value when the bypass diode is in a conducting state includes: Based on the current value and voltage value when the bypass diode is in a conducting state, performing non-linear least squares fitting using a trust region algorithm to obtain the first parameter; the first parameter enables the error of the bypass diode sub-model simulating the operation of the bypass diode to meet a preset condition.

6. The method according to claim 4 or 5, characterized in that, The method further includes: Obtaining the current value and voltage value of the photovoltaic module when in a preset fault; Based on the current value and voltage value of the photovoltaic module when in a preset fault, using a particle swarm optimization algorithm to optimize the second parameter in the single diode sub-model to obtain the second parameter under the preset fault.

7. The method according to claim 1, wherein The first loss function of the fault prediction model is obtained by weighted summation of multiple second loss functions; wherein, the second loss function corresponds to a preset fault one by one, and the second loss function is used to indicate the prediction error of the fault prediction model for the preset fault; the weights of the multiple second loss functions are determined based on at least one of the severity, occurrence frequency, and detection difficulty of the respective corresponding faults.

8. The method according to claim 1 or 7, characterized in that, The obtaining the fault prediction result of the target photovoltaic module based on the electrical data and the fault prediction model includes: Input the electrical data into a pre - constructed auto - encoder to obtain the reconstructed data of the target photovoltaic module; Based on the reconstructed data and the electrical data, determine whether the target photovoltaic module has a fault; In the case where the target photovoltaic module has a fault, based on the electrical data and the fault prediction model, obtain the fault prediction result.

9. The method according to claim 8, wherein The determining whether the target photovoltaic module has a fault based on the reconstructed data and the electrical data includes: In the case where the deviation between the reconstructed data and the electrical data is greater than a preset threshold, determine that the target photovoltaic module has a fault.

10. The method according to claim 8, characterized in that, The fault prediction model is a neural network optimized based on a pruning algorithm; the obtaining the fault prediction result based on the electrical data and the fault prediction model in the case where the target photovoltaic module has a fault includes: In the case where the target photovoltaic module has a fault, input the encoded data corresponding to the electrical data into the fault prediction model to obtain the fault prediction result; the encoded data corresponding to the electrical data is obtained by inputting the electrical data into the encoder in the auto - encoder.

11. The method according to claim 1, characterized in that, The method further includes: Obtain a plurality of initial data; the plurality of initial data correspond to multiple fault types; Based on the information gain of each initial data, screen the plurality of initial data to obtain the plurality of training data.

12. The method according to claim 11, wherein The screening the plurality of initial data based on the information gain of each initial data to obtain the plurality of training data includes: Based on the information gain of each initial data, use a sliding window test algorithm to determine an information gain threshold; Based on the information gain threshold and the information gain of each initial data, obtain the plurality of training data.

13. The method according to claim 12, characterized in that, The obtaining the plurality of training data based on the information gain threshold and the information gain of each initial data includes: Based on the information gain threshold and the information gain of each initial data, obtain initial training data; Use the principal component analysis method and the initial training data to obtain the plurality of training data.

14. The method according to claim 1, characterized in that, The photovoltaic module model includes a single - diode sub - model; the single - diode sub - model is used to simulate the operation of the photovoltaic cells in the photovoltaic module; the method further includes: Obtain the environmental data of the target photovoltaic module; Based on the environmental data, determine the operating temperature of the target photovoltaic module; Based on the operating temperature and the second parameter of the single - diode sub - model, determine the degradation rate of the target photovoltaic module; Based on the degradation rate, predict the lifespan of the target photovoltaic module.

15. The method according to claim 14, wherein The predicting the lifespan of the target photovoltaic module based on the degradation rate includes: Based on the degradation rate, determine the moment when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold; Based on the moment, determine the lifespan of the target photovoltaic module.

16. The method according to claim 15, wherein The determining the moment when the power attenuation of the target photovoltaic module reaches a preset power attenuation threshold based on the degradation rate includes: Based on the degradation rate and a pre-constructed corrosion degradation model, determine the moment when the power attenuation of the target photovoltaic module reaches the corresponding preset power attenuation threshold.

17. The method according to claim 16, wherein The corrosion degradation model includes: where, ΔP(t) is the power attenuation at time t, ΔP ∞ is the saturated power attenuation after long-term exposure, ΔP(t i ) is the initial power loss, R D is the degradation rate, t i is the starting degradation time.

18. The method according to claim 14, wherein The determining of the degradation rate of the target photovoltaic module based on the operating temperature and the second parameter of the single diode sub-model includes: where, R D is the degradation rate of the target photovoltaic module, T is the operating temperature of the target photovoltaic module, RH is the relative humidity, E A is the activation energy of the sealing material, k B is the Boltzmann constant, n is the humidity coefficient, A is the empirical constant, R s,new is the second parameter in the single-diode sub-model.

19. The method according to claim 7, wherein The method further includes: Based on the severity, occurrence frequency, and detection difficulty of the faults corresponding to each of the second loss functions, use a risk priority numbering mechanism to determine the risk priority of the faults corresponding to each of the second loss functions. Based on the risk priority of the faults corresponding to each of the second loss functions, determine the weight of each of the second loss functions.

20. A fault prediction device, characterized in that, The device includes: A first acquisition module, configured to acquire electrical data of the target photovoltaic module; the target photovoltaic module includes at least one target bypass diode. A first prediction module, configured to obtain a fault prediction result of the target photovoltaic module based on the electrical data and a fault prediction model; a plurality of training data of the fault prediction model are obtained by simulating a photovoltaic module based on a pre-constructed photovoltaic module model; the photovoltaic module model includes an equivalent bypass diode and an equivalent resistor connected in series with the equivalent bypass diode; the equivalent bypass diode is used to simulate the one-way conduction characteristic of the bypass diode in the photovoltaic module; the equivalent resistor is used to correct the operating characteristic of the equivalent bypass diode.

21. A computer device, characterized in that, Includes: A memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the fault prediction method according to any one of claims 1 to 19.

22. A computer-readable storage medium, characterized in that, Computer instructions are stored on the computer-readable storage medium, and the computer instructions are used to cause a computer to execute the fault prediction method according to any one of claims 1 to 19.

23. A computer program product, characterized in that, Includes computer instructions, and the computer instructions are used to cause a computer to execute the fault prediction method according to any one of claims 1 to 19.

Images