Method and device for determining fault prediction model of photovoltaic inverter, and storage medium

Abstract

The invention discloses a photovoltaic inverter fault prediction model determination method and device, a fault prediction method, a storage medium, an electronic device and a computer program product, and the method comprises the steps: obtaining historical data of a plurality of photovoltaic inverters, and building a data sample set based on the historical data; performing feature extraction on the data sample set by adopting a multi-scale convolution mode, and determining fault feature sets under different time scales; splicing the fault feature set and the data sample set to obtain a training sample, and training a plurality of classifiers in sequence by adopting the training sample; determining weights corresponding to the plurality of classifiers based on the classification error rates of the plurality of classifiers; and combining the plurality of classifiers according to the weights corresponding to the plurality of classifiers, and determining a fault prediction model of the photovoltaic inverter. Through multi-scale convolution feature extraction and weighted classifier combination, the fault prediction accuracy and reliability of the model are improved, and the model can early warn potential faults of the photovoltaic inverter in advance.

Description

Technical Field

[0001] The present application relates to the photovoltaic field, and specifically, to a method and device for determining a fault prediction model of a photovoltaic inverter, a fault prediction method, a storage medium, an electronic device, and a computer program product. Background Art

[0002] In modern energy systems, photovoltaic inverters play a vital role as a bridge connecting solar cells with the power grid or load. They are not only responsible for converting the direct current generated by solar cells into alternating current to meet the needs of the power grid or household electricity, but also undertake important functions such as maximum power point tracking, voltage and current control, and protection circuits. Like other electronic devices, photovoltaic inverters are affected by various internal and external factors, such as the aging of electronic components, fluctuations in ambient temperature, and abnormal power quality, all of which may cause inverter failures. As the core equipment of the photovoltaic system, the inverter's failure will not only affect the operation of the photovoltaic system, but also cause irreparable losses to the inverter and even the power grid. Therefore, accurate prediction of photovoltaic inverter failures is of great significance to ensure the stable operation of the photovoltaic system.

[0003] In the related art, the operation and maintenance of photovoltaic inverters usually rely on regular equipment inspections and post-failure responses after equipment failures occur.

[0004] However, regular inspections can reveal some obvious equipment problems, but it is difficult to judge potential, early faults, which are often not discovered until the fault affects the output efficiency of the inverter or causes more serious problems. Responding after the fact means that the fault has already occurred, which will not only lead to a decrease in the power generation efficiency of the photovoltaic system, but may also cause a chain reaction to the entire energy network, and even bring safety risks and economic losses. Summary of the invention

[0005] Embodiments of the present application provide a method and device for determining a fault prediction model of a photovoltaic inverter, a fault prediction method, a storage medium, an electronic device, and a computer program product.

[0006] According to one aspect of an embodiment of the present application, a method for determining a fault prediction model of a photovoltaic inverter is provided, the method comprising: acquiring historical data of multiple photovoltaic inverters, and establishing a data sample set based on the historical data; extracting features from the data sample set using a multi-scale convolution method, and determining a fault feature set at different time scales; concatenating the fault feature set with the data sample set to obtain a training sample, and using the training sample to train multiple classifiers in sequence; determining weights corresponding to the multiple classifiers based on their respective classification error rates; and combining the multiple classifiers according to their respective weights to determine a fault prediction model for the photovoltaic inverter.

[0007] In one embodiment of the present application, a multi-scale convolution method is used to extract features from a data sample set to determine fault feature sets at different time scales, specifically including: converting the data sample set into a vector form to obtain a data sample vector, wherein the data sample vector includes a time series of data of multiple different categories; using convolution kernels of multiple different scales to extract features from the data sample vector respectively to obtain multiple groups of fault features of different scales; and splicing multiple groups of fault features of different scales to obtain fault feature sets at different time scales.

[0008] In one embodiment of the present application, a fault feature set and a data sample set are concatenated to obtain a training sample, and a plurality of classifiers are trained in sequence using the training sample, specifically including: defining initial weights for a plurality of sample features in the training sample, wherein the initial weight of each sample feature is equal; obtaining a plurality of classifiers to be trained; training a first classifier using the training sample, wherein the first classifier is one of the plurality of classifiers; adjusting the initial weights of the plurality of sample features according to the classification result output by the trained first classifier; using a classifier among the plurality of classifiers that has not been trained as the next first classifier, and repeating the step of training the first classifier using the training sample until the plurality of classifiers are trained.

[0009] In one embodiment of the present application, weights corresponding to multiple classifiers are determined based on their respective classification error rates, specifically including: determining the classification error rates of the multiple classifiers according to the classification results respectively output by the multiple classifiers; determining the weights corresponding to the multiple classifiers according to their respective classification error rates.

[0010] In one implementation of the present application, historical data of multiple photovoltaic inverters are obtained, and a data sample set is established based on the historical data, specifically including: obtaining historical data of multiple photovoltaic inverters; adding labels to the historical data of the multiple photovoltaic inverters for classification according to maintenance records of the multiple photovoltaic inverters; and preprocessing the classified historical data of the multiple photovoltaic inverters to obtain a data sample set.

[0011] One aspect of the present application provides a fault prediction method for a photovoltaic inverter, which applies the fault prediction model of the photovoltaic inverter determined above, and the method includes: obtaining current data of the photovoltaic inverter to be predicted; inputting the current data of the photovoltaic inverter to be predicted into the fault prediction model of the photovoltaic inverter, determining the fault condition of the photovoltaic inverter and recording it; and determining the failure probability of the photovoltaic inverter based on the fault condition of the photovoltaic inverter recorded within a preset time period.

[0012] Another aspect of the present application provides a device for determining a fault prediction model of a photovoltaic inverter, the device comprising: a sample set establishment module, used to obtain historical data of multiple photovoltaic inverters, and establish a data sample set based on the historical data; a feature extraction module, used to extract features from the data sample set using a multi-scale convolution method, and determine the fault feature set at different time scales; a training module, used to splice the fault feature set with the data sample set to obtain training samples, and use the training samples to train multiple classifiers in sequence; a weight determination module, used to determine the weights corresponding to multiple classifiers based on the classification error rates of the multiple classifiers; a combination module, used to combine the multiple classifiers according to the weights corresponding to the multiple classifiers, and determine the fault prediction model of the photovoltaic inverter.

[0013] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned method for determining a fault prediction model of a photovoltaic inverter when running.

[0014] According to another aspect of an embodiment of the present application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the fault prediction model determination method of the photovoltaic inverter through the computer program.

[0015] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps of the method described in each embodiment of the present application are implemented.

[0016] The above-mentioned photovoltaic inverter fault prediction model determination method and device, fault prediction method, storage medium, electronic device, and computer program product use a multi-scale convolutional neural network to extract features from a data sample set, with the purpose of capturing patterns and features at different time scales from time series data. By using convolution kernels of different sizes, various features can be extracted from short time scales to long time scales to generate more comprehensive and complex features, covering signals from small fluctuations to long-term trends. By capturing features at different time scales, the subsequently trained model can more accurately identify early signs of faults and improve the sensitivity and reliability of predictions. Then, multiple classifiers are trained in sequence using training samples. By adding features, the model can analyze data more carefully and improve the accuracy and complexity of classification. The combination of multiple classifiers can improve the stability and accuracy of the model and have a stronger recognition ability for complex fault modes. All trained classifiers are weighted and combined according to their weights to generate a final fault prediction model. In summary, through multi-scale convolution feature extraction and weighted classifier combination, the fault characteristics of photovoltaic inverters at different time scales can be effectively identified, and the accuracy and reliability of model fault prediction can be improved. This method can not only process the time series characteristics in the historical data of photovoltaic inverters, but also optimize the prediction performance of the model by adjusting the weight of the classifier. In practical applications, the model can provide early warning of potential failures of photovoltaic inverters, reduce unplanned downtime, improve the operating efficiency and economy of photovoltaic systems, and provide a scientific basis for the maintenance and management of photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0019] Figure 1 It is a hardware structure block diagram of a fault prediction model determination method for a photovoltaic inverter according to an embodiment of the present application;

[0020] Figure 2 is a flow chart of a method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application;

[0021] Figure 3 This is a second flowchart of a method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application;

[0022] Figure 4 is a flowchart of a method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application;

[0023] Figure 5 This is a fourth flowchart of a method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application;

[0024] Figure 6 is a fifth flowchart of a method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application;

[0025] Figure 7 is a sixth flowchart of a method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application;

[0026] Figure 8 It is a structural block diagram of a fault prediction model determination device for a photovoltaic inverter according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0029] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 1 is a hardware structure block diagram of a computer terminal of a method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application. Figure 1 As shown, the computer terminal may include one or more ( Figure 1Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a microprocessor (Microprocessor Unit, referred to as MPU) or a programmable logic device (Programmable logic device, referred to as PLD)) and a memory 104 for storing data. In an exemplary embodiment, the computer terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Equivalent functions or comparisons shown Figure 1 A different configuration with more features is shown.

[0030] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the fault prediction model determination method of the photovoltaic inverter in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0031] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0032] In this embodiment, a method for determining a fault prediction model of a photovoltaic inverter is provided. Figure 2 1 is a flow chart of an optional method for determining a fault prediction model of a photovoltaic inverter according to an embodiment of the present application, the flow includes the following steps S200-S240:

[0033] Step S200, acquiring historical data of a plurality of photovoltaic inverters, and establishing a data sample set based on the historical data.

[0034] Specifically, the operation history data of each photovoltaic inverter can be collected from the photovoltaic power station, including but not limited to ambient temperature, irradiance, input current, output current, output power, etc. These data are recorded in normal and faulty states of the equipment. By comparing the normal and pre-fault data, a data sample set can be constructed.

[0035] Step S210: extract features from the data sample set using a multi-scale convolution method to determine fault feature sets at different time scales.

[0036] Specifically, a multi-scale convolutional neural network is used to extract features from the data sample set, with the goal of capturing patterns and features at different time scales from the time series data. By using convolution kernels of different sizes, various features can be extracted from short to long time scales, such as the fluctuation pattern of current in a short time, the trend of power decline in the medium and long term, etc. These features help identify potential fault signals of the inverter.

[0037] Step S220: Concatenate the fault feature set and the data sample set to obtain training samples, and use the training samples to train multiple classifiers in sequence.

[0038] Specifically, after feature extraction, the obtained fault feature set is merged with the original data sample set to form a new training sample set for training the classifier. The training process usually adopts an ensemble learning method, such as the AdaBoost algorithm, to gradually build multiple weak classifiers, each of which may focus on a certain feature or pattern in the sample set.

[0039] Step S230: determining weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively.

[0040] Specifically, during the training process, the performance of each weak classifier is measured based on its classification error rate. The lower the error rate of the classifier, the higher its weight, which reflects the importance of the model in the ensemble learning. The weight is determined based on the algorithm's iterative process, such as the weighted error rate in AdaBoost.

[0041] Step S240: combining the multiple classifiers according to the weights corresponding to the multiple classifiers to determine a fault prediction model for the photovoltaic inverter.

[0042] Specifically, all trained classifiers are weighted and combined according to their weights to generate the final fault prediction model. This model is an integrated model that can integrate the prediction results of multiple classifiers and determine the classification of new samples through weighted decision making.

[0043] In this embodiment, a multi-scale convolutional neural network is used to extract features from a data sample set, with the goal of capturing patterns and features at different time scales from time series data. By using convolution kernels of different sizes, various features can be extracted from short time scales to long time scales, generating more comprehensive and complex features, covering signals from small fluctuations to long-term trends. By capturing features at different time scales, the model trained subsequently can more accurately identify early signs of faults and improve the sensitivity and reliability of predictions. Then, multiple classifiers are trained in sequence using training samples, and by adding features, the model can analyze data more carefully and improve the accuracy and complexity of classification. The combination of multiple classifiers can improve the stability and accuracy of the model and have a stronger recognition ability for complex fault patterns. All trained classifiers are weighted and combined according to their weights to generate the final fault prediction model. In summary, through multi-scale convolution feature extraction and weighted classifier combination, the fault characteristics of photovoltaic inverters at different time scales can be effectively identified, and the accuracy and reliability of fault prediction can be improved. This method can not only process the time series features in the historical data of photovoltaic inverters, but also optimize the prediction performance of the model by adjusting the weights of the classifiers. In practical applications, the model can provide early warning of potential failures of photovoltaic inverters, reduce unplanned downtime, improve the operating efficiency and economy of photovoltaic systems, and provide a scientific basis for the maintenance and management of photovoltaic power stations.

[0044] In one embodiment, Figure 3 As shown, step S210 uses a multi-scale convolution method to extract features from the data sample set to determine the fault feature set at different time scales. It includes: steps S300-S320:

[0045] Step S300, converting the data sample set into a vector form to obtain a data sample vector.

[0046] The data sample vector includes time series of data of different categories.

[0047] Specifically, the collected historical operation data of photovoltaic inverters is converted into a vector form suitable for processing by machine learning algorithms. Each data sample vector contains multiple types of time series data, such as the change records of ambient temperature, irradiance, input current, output current and output power over time. This conversion converts the data from the original table or record form into a one-dimensional or high-dimensional vector, which is convenient for subsequent feature extraction and classification learning.

[0048] Step S310, using convolution kernels of multiple different scales to extract features from the data sample vectors respectively, to obtain multiple groups of fault features of different scales.

[0049] Specifically, this step uses the convolution operation in deep learning to automatically extract features, especially the one-dimensional convolution kernel, which is used to slide on the time series data and extract features. The use of convolution kernels of different scales (or widths) is to capture features of different time scales in the data. For example, a smaller convolution kernel may capture short-term fluctuation patterns, while a larger convolution kernel may identify long-term operating trends. This process generates multiple sets of fault features, each set of features corresponding to a specific time scale.

[0050] For example, the operating parameters of a photovoltaic inverter can reflect the operating status of the equipment. In addition to the directly accessible measurement point data, there are also certain correlations between different measurement points, such as the relationship between irradiance and DC current. Using one-dimensional convolution to extract features between different features can generate more complex features, which is of great significance for the subsequent model fault identification. For a data sample vector, that is, a one-dimensional input vector x = (x0, x1, ..., x n-1 ), and a one-dimensional convolution kernel w = (w0, w1, ..., w k-1 ) and the bias term b, for the output y at position i i It can be calculated by the following formula:

[0051]

[0052] The features that can be extracted by a single-scale convolution are limited. In order to improve the diversity and expression ability of features, multi-scale convolution is used to obtain features of different receptive fields. That is, one-dimensional convolution kernels with different k values ​​are selected, that is, convolution kernels of different sizes are defined. i , where i represents the index of the convolution kernel, and each convolution kernel w i The size is k i , the specific form is And define the bias term b i , for each convolution kernel w i , convolution operation is performed with the input signal x by sliding window. For the i-th convolution kernel, its corresponding feature map y i The calculation formula is:

[0053]

[0054] Among them, t represents the position index of the newly generated feature map, k i is the size of the i-th convolution kernel, x[t+j] represents the value of the input signal x at position t+j, and w i [j] is the convolution kernel w i The weight at position j, f is the activation function, which can be linear or non-linear, such as ReLU.

[0055] Repeat the above steps and use multiple convolution kernels w of different sizes i Convolve the input vector x to obtain feature maps y with different receptive fields i Each convolution kernel corresponds to a feature map, and the size of the feature map depends on the size of the convolution kernel and the length of the input vector.

[0056] Step S320, concatenate multiple groups of fault features at different scales to obtain fault feature sets at different time scales.

[0057] Specifically, after using multi-scale convolution kernels to extract fault features separately, these features need to be combined to form a more comprehensive fault feature set. The concatenation operation connects feature vectors of different scales to create a high-dimensional feature space containing all information, providing rich input for subsequent classification tasks. After completing the convolution operation of all convolution kernels, the generated multiple feature maps y i Splicing together to form an enhanced feature representation. Splicing can be performed on the feature dimension, that is, for each time point t, the value y of all feature maps at that time point i [t] are connected into a new vector to form a new feature data set. The concatenated feature vector is combined with the original operating parameter data set to form an updated data set containing more complex features.

[0058] In this embodiment, by converting the data sample set into the form of a vector, the unstructured or semi-structured data is converted into a structured vector, which simplifies the input format of the model and improves the computational efficiency. The converted time series data helps the model capture the dynamic change law of the data, which is crucial for identifying the changing trend of the inverter operating state. Convolution kernels of different scales can extract features in different time windows, which increases the model's ability to understand the diversity of potential faults of the inverter. Through the convolution operation, the model can automatically identify and learn patterns in time series data, which is more efficient and accurate. The splicing feature can integrate the information extracted from different time scales to form a more complete and comprehensive fault feature description. The integration of multi-scale features provides more dimensional information, which helps the classifier distinguish between normal and potential fault states, thereby improving the accuracy and reliability of fault prediction.

[0059] In one embodiment, Figure 4 As shown, in step S220, the fault feature set and the data sample set are spliced ​​to obtain training samples, and the training samples are used to train multiple classifiers in sequence. It includes: steps S400-S450:

[0060] Step S400, defining initial weights for multiple sample features in a training sample.

[0061] The initial weight of each sample feature is equal. Before starting training, the same initial weight needs to be assigned to each sample in the data set. This ensures that in the early stages of the training process, the model gives equal weight to all samples (regardless of their category), thereby avoiding premature bias towards a certain type of sample. In ensemble learning, such as the AdaBoost algorithm, weight adjustment is one of the core processes, used to guide the model to pay more attention to samples that are difficult to classify correctly.

[0062] For example, for the sample feature x used for training i =(x i1 ,x i2 ,…,x in ), the label can be represented as y i ∈{0,1}, in order to adapt to the training of the model, the label is converted from {0,1} to {-1,1}. The label conversion method can be performed by the following formula:

[0063] y i ′=2y i -1

[0064] where y i Indicates the label before conversion, y i ′ represents the converted label.

[0065] Initially, the weight of each sample is equal. Assuming there are L training data, the weight distribution of the initial sample can be expressed as:

[0066]

[0067] Step S410, obtaining multiple classifiers to be trained.

[0068] Specifically, before training, multiple weak classifiers need to be selected or generated, which will be gradually trained and eventually form a strong classifier through weighted combination. Usually, these weak classifiers can be simple models such as decision trees, k-nearest neighbor classification (kNN), and support vector machines (SVM), but they have limited classification capabilities in isolated cases.

[0069] Step S420: training the first classifier using training samples.

[0070] The first classifier is one of the multiple classifiers. The first weak classifier is trained using a training sample set with initial weights. This process usually uses a weighted loss function to ensure that the classifier can take into account the weight of the sample.

[0071] For example, for M weak classifiers m=1, 2, ..., M, the current sample distribution D is used in each iteration. m Train a weak classifier G m (x), the weak classifier is constructed using a decision tree. Then the error rate of the weak classifier is calculated:

[0072]

[0073] Among them, ∈ m It represents the weighted error of the weak classifier based on the current sample distribution. I is the indicator function, which returns 1 when the condition is true and 0 otherwise. i ′ is the actual label of the i-th sample.

[0074] Calculate the weights of the weak classifiers:

[0075]

[0076] Weight α m It can measure the importance of the current weak classifier. m When it is close to 0, the error of the current weak classifier is small, α m The value of is large, when ∈ m When α is close to 0.5, m The value of is close to 0.

[0077] According to the performance of the weak classifier, update the sample distribution:

[0078]

[0079] Where Z m is a normalization factor that ensures that D m+1 is a valid probability distribution, Z m It can be expressed as:

[0080]

[0081] The weight of the currently correctly classified sample is reduced (here is the weight of the sample, not the weight of the weak classifier), and the weight of the incorrectly classified sample is increased, so that the weak classifier pays more attention to difficult samples in subsequent iterations.

[0082] Step S430, adjusting the initial weights of the plurality of sample features according to the classification result output by the trained first classifier.

[0083] Specifically, after the first classifier is trained, the weights of the samples are adjusted based on their classification results. The weights of samples that are misclassified are increased, while the weights of samples that are correctly classified are reduced. This adjustment process is iterative, ensuring that subsequent classifiers pay more attention to samples that are misclassified.

[0084] Step S440, determining whether the multiple classifiers have all been trained. If not all of them have been trained, executing step S450.

[0085] Step S450: use one of the multiple classifiers that has not been trained as the next first classifier, and return to step S420 to train the first classifier using the training sample.

[0086] Specifically, after the first classifier is trained and weighted, the process continues with the next untrained classifier, repeating the same training and weighting process. This cycle continues until all classifiers are trained, and each classifier is optimized based on the previous classifier to better handle samples that are misclassified. The training of each classifier is optimized based on the output of the previous classifier, which helps to gradually enhance the classification ability of the model.

[0087] In this embodiment, all samples are given equal importance at the beginning, which helps the model to learn the features of all samples in a balanced manner in the early stages of training. Avoid the model from being overly dependent on a certain type of sample in the early stages of training, thereby improving the generalization ability of the model. Through training, the first classifier begins to learn the features in the training samples, providing preliminary classification capabilities for subsequent model enhancements. By adjusting the weights, the model pays more attention to samples that are difficult to classify, which helps to improve the final classification performance of the model. The dynamic adjustment of weights enables the model to continuously optimize its learning focus during the training process and gradually improve the classification accuracy of difficult samples. Ultimately, multiple weak classifiers are combined by weighted combination to form a strong classifier, which enables the model to more accurately identify and classify potential fault states of photovoltaic inverters.

[0088] In one embodiment, Figure 5 As shown, step S230, based on the classification error rates of the multiple classifiers, the weights corresponding to the multiple classifiers are determined. It includes: steps S500-S510:

[0089] Step S500, determining the classification error rates of the multiple classifiers respectively according to the classification results respectively output by the multiple classifiers.

[0090] Specifically, during the training process, each weak classifier will classify the training samples and output the classification results. The classification error rate refers to the proportion of training samples that are incorrectly classified by the classifier, that is, in a given weighted sample, the sum of the weights of the samples that the classifier misclassifies divided by the sum of the weights of all samples. By calculating the classification error rate of each classifier, its ability to identify faults can be evaluated. The lower the error rate, the stronger the recognition ability of the classifier. The classification error rate provides an intuitive measure of model performance and helps to identify which classifiers perform better on the current training set.

[0091] Step S510: determining weights corresponding to the multiple classifiers respectively according to the classification error rates of the multiple classifiers respectively.

[0092] Specifically, the weight of the classifier is calculated based on its classification error rate, and the classifier with the lower error rate will be given a higher weight. The calculation method of the weight usually involves the logarithmic transformation of the error rate, and then normalization to ensure that the weight value is within a reasonable range. This process ensures that the classifiers with better performance have a greater influence in subsequent ensemble decisions. By assigning a weight to each classifier, the model can pay more attention to weak classifiers with higher classification accuracy, making it more accurate in ensemble decisions. Dynamic adjustment: The weights of the classifiers are dynamically adjusted as the training process progresses, which allows the model to self-optimize in iterations and gradually improve the overall classification performance.

[0093] Exemplarily, step S240 may specifically include: weighted combination of all weak classifiers, and finally generating a strong classifier G(x):

[0094]

[0095] Among them, sign(·) represents the sign function, which returns 1 or -1 to determine the final category, G m (x) is the mth classifier, α m is the weight of the mth classifier. So far, we have completed the construction of the AdaBoost classifier. According to the classification effect of the model, the optimal model is saved for the subsequent photovoltaic inverter fault prediction task.

[0096] In this embodiment, by determining the classification error rates of multiple classifiers, it is easy to intuitively identify which classifiers perform better on the current training set. The calculation of the error rate is a key link in the iterative learning process. It provides guidance for the training and weight allocation of subsequent classifiers to ensure that the model can be gradually improved. The whole process aims to optimize the decision-making ability of the integrated learning model by evaluating and adjusting the performance of each classifier on the training data set. In each iteration, the model recalculates its weight according to the classification error rate of each classifier, and the weight of the misclassified samples will increase, prompting the subsequent classifiers to pay more attention to these difficult samples. In this way, as the iteration proceeds, the model can gradually learn more complex and comprehensive fault features, and finally form a powerful and accurate fault prediction model that can effectively identify the potential fault state of the photovoltaic inverter. This strategy of adjusting weights based on the classification error rate is the core of the AdaBoost algorithm. It can not only improve the classification accuracy of the model, but also enhance the robustness and generalization ability of the model, and has significant application value for fault warning and preventive maintenance of photovoltaic inverters.

[0097] In one embodiment, Figure 6 As shown, step S200, obtains historical data of multiple photovoltaic inverters, and establishes a data sample set based on the historical data. It includes: steps S600-S620:

[0098] Step S600, obtaining historical data of multiple photovoltaic inverters.

[0099] Specifically, historical operating data is collected from the inverter equipment of the photovoltaic power station. The data usually includes real-time or periodic monitoring parameters of the inverter, such as ambient temperature, irradiance, input and output current, output power, etc. These parameters can reflect the operating status of the inverter under different environments and working conditions. Data collection can be assisted by IoT devices or Supervisory Control And Data Acquisition (SCADA) systems to ensure data quality and real-time performance.

[0100] For example, the photovoltaic inverter fault prediction problem can be regarded as a two-class problem, with the two categories being normal equipment and potential fault equipment. First, the historical operation data of the equipment in the station is collected, including ambient temperature, irradiance, input current, output current, output power, etc. The granularity of data collection is 5 minutes.

[0101] Step S610: adding tags to historical data of the plurality of photovoltaic inverters for classification according to the maintenance records of the plurality of photovoltaic inverters.

[0102] Specifically, after collecting the historical operating data of the inverter, the data needs to be classified and labeled according to the maintenance records of the equipment. Specifically, if the inverter is recorded as having a fault within a week after a certain point in time, then all data within this week will be marked as a "potential fault" state (for example, the label is 1), and the data of those inverters that have not failed during the corresponding time will be marked as a "normal" state (for example, the label is 0). In this way, the data is converted into the format required for supervised learning, that is, each data sample is equipped with a label indicating that it belongs to the "normal" or "potential fault" category.

[0103] Step S620 , preprocessing the classified historical data of the plurality of photovoltaic inverters to obtain a data sample set.

[0104] Specifically, preprocessing is a key step in machine learning and data mining. It includes steps such as data cleaning, feature engineering, and data normalization, which aim to improve the quality of data and the performance of the model. Preprocessing can include removing missing values, outliers, standardizing the data range, and denoising the data. The preprocessed data will be used as training samples for training fault prediction models. Through accurate labeling and careful preprocessing, the model can learn based on clearer and purer data sets, thereby more accurately and reliably predicting potential faults of photovoltaic inverters, providing strong support for preventive maintenance and optimized operation of photovoltaic systems.

[0105] For example, photovoltaic power generation has obvious periodicity. After obtaining the operating data, the data can be filtered according to the irradiance, and only the irradiance>50W / m 2 data. Secondly, due to environmental factors and failures of the data acquisition equipment itself, the historical data obtained contains some dirty data, which needs to be cleaned. Dirty data can include missing values, constant values, and outliers. When there are missing values ​​in the data, if the data of the adjacent time steps in the row where the missing value is located is normal, the missing value is filled with the average value of the adjacent data. If the data of multiple consecutive time steps are missing, the missing data is deleted. For data with constant values, the row is deleted. For outliers (outside the reasonable range of the data), they are deleted based on the 3σ criterion. Finally, the prepared data set is divided into training set and test set in a ratio of 8:2. The input feature data is normalized to avoid the dimensional differences of different features affecting the model training.

[0106] In this embodiment, historical data provides necessary information for model training and is the basis for building a fault prediction model. The collected long-term historical data can cover various operating conditions and scenarios, which helps the model learn the normal and abnormal behavior patterns of the inverter. The labeled data provides a clear learning goal for the model, that is, learning to distinguish between normal and potential fault states, which can more effectively train the model and identify fault precursors. Preprocessing can reduce noise in the data, improve data consistency and availability, and thus improve the training efficiency and prediction accuracy of the model.

[0107] In one embodiment, Figure 7 As shown, a photovoltaic inverter fault prediction method is provided, and the photovoltaic inverter fault prediction model determined in the above embodiment is applied. The method includes: steps S700-S720:

[0108] Step S700, obtaining current data of the photovoltaic inverter to be predicted.

[0109] Specifically, the current operating data can be collected from the photovoltaic inverter in real time or periodically to facilitate real-time monitoring of the inverter's operating status, and the collected data will be used as the prediction input of the model.

[0110] Step S710: input the current data of the photovoltaic inverter to be predicted into the fault prediction model of the photovoltaic inverter, determine the fault condition of the photovoltaic inverter and record it.

[0111] Specifically, after collecting the current data of the PV inverter, it is input into the previously trained fault prediction model. The model will analyze the data features and output a prediction result indicating whether the inverter is currently in a potential fault state. The prediction result will be recorded for subsequent analysis and decision-making reference.

[0112] Step S720: determining the failure probability of the photovoltaic inverter according to the failure conditions of the photovoltaic inverter recorded within a preset time period.

[0113] Specifically, within a certain period of time, such as a day, a week, or a month, the fault prediction results of multiple time points will be accumulated. By analyzing these results, the frequency or proportion of the inverter in a potential fault state can be calculated, thereby deriving a fault probability. The calculation of the fault probability can be based on simple statistical methods, such as frequency calculation, or more complex models, such as time series analysis, to consider the persistence and pattern of the fault state. Once the predicted fault probability exceeds the threshold, it means that the photovoltaic inverter has a potential fault, and an early warning message will be issued to remind the staff.

[0114] In this embodiment, by real-time monitoring of the operating status changes of the inverter, the response speed of the system can be improved. And the current data of the inverter also provides the latest information for the prediction model, improving the accuracy of fault prediction. After collecting the current data of the photovoltaic inverter, it is input into the previously trained fault prediction model, which can instantly identify the potential faults of the inverter and provide early warning signals for the operation and maintenance team. Data recording: Keeping records of the prediction results will help with subsequent analysis of fault trends and frequencies and optimize maintenance strategies. By determining the probability of failure, a quantitative indicator is provided to help the operation and maintenance team understand and evaluate the health status of the inverter. The probability of failure can be used as a basis for decision-making to determine whether maintenance or equipment replacement is needed in advance to reduce the impact of sudden failures on the system. This fault prediction and probability assessment method based on real-time data has significant advantages in improving the operation and maintenance efficiency of photovoltaic power stations.

[0115] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0116] In this embodiment, a fault prediction model determination device for a photovoltaic inverter is also provided, and the fault prediction model determination device for the photovoltaic inverter is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0117] Figure 8 is a structural block diagram of an optional photovoltaic inverter fault prediction model determination device according to an embodiment of the present application. Figure 8 As shown, including:

[0118] The sample set establishing module 801 is used to obtain historical data of multiple photovoltaic inverters and establish a data sample set based on the historical data.

[0119] The feature extraction module 802 is used to extract features from the data sample set using a multi-scale convolution method to determine the fault feature set at different time scales.

[0120] The training module 803 is used to concatenate the fault feature set and the data sample set to obtain training samples, and use the training samples to train multiple classifiers in sequence.

[0121] The weight determination module 804 is used to determine the weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively.

[0122] The combination module 805 is used to combine multiple classifiers according to the weights corresponding to the multiple classifiers to determine the fault prediction model of the photovoltaic inverter.

[0123] Through the above device, a multi-scale convolutional neural network is used to extract features from the data sample set, with the aim of capturing patterns and features at different time scales from time series data. By using convolution kernels of different sizes, various features can be extracted from short time scales to long time scales, generating more comprehensive and complex features, covering signals from small fluctuations to long-term trends. By capturing features at different time scales, the subsequently trained model can more accurately identify early signs of faults and improve the sensitivity and reliability of predictions. Then, multiple classifiers are trained in sequence using training samples. By adding features, the model can analyze the data more carefully and improve the accuracy and complexity of classification. The combination of multiple classifiers can improve the stability and accuracy of the model and have a stronger recognition ability for complex fault patterns. All trained classifiers are weighted and combined according to their weights to generate the final fault prediction model. In summary, through multi-scale convolution feature extraction and weighted classifier combination, the fault characteristics of photovoltaic inverters at different time scales can be effectively identified, and the accuracy and reliability of fault prediction can be improved. This method can not only process the time series features in the historical data of photovoltaic inverters, but also optimize the prediction performance of the model by adjusting the weights of the classifiers. In practical applications, the model can provide early warning of potential failures of photovoltaic inverters, reduce unplanned downtime, improve the operating efficiency and economy of photovoltaic systems, and provide a scientific basis for the maintenance and management of photovoltaic power stations.

[0124] In an exemplary embodiment, the feature extraction module 802 is further used to convert the data sample set into a vector form to obtain a data sample vector, wherein the data sample vector includes a time series of data of multiple different categories. Convolution kernels of multiple different scales are used to extract features from the data sample vector respectively to obtain multiple groups of fault features of different scales. Multiple groups of fault features of different scales are spliced ​​to obtain fault feature sets at different time scales.

[0125] In an exemplary embodiment, the training module 803 is also used to define initial weights for multiple sample features in the training sample, wherein the initial weight of each sample feature is equal. Obtain multiple classifiers to be trained. Use the training sample to train a first classifier, wherein the first classifier is one of the multiple classifiers. Adjust the initial weights of the multiple sample features according to the classification results output by the trained first classifier. Use one of the multiple classifiers that has not been trained as the next first classifier, and return to the step of training the first classifier using the training sample until the multiple classifiers are trained.

[0126] In an exemplary embodiment, the weight determination module 804 is further configured to determine the classification error rates of the multiple classifiers according to the classification results outputted by the multiple classifiers, and to determine the weights corresponding to the multiple classifiers according to the classification error rates of the multiple classifiers.

[0127] In an exemplary embodiment, the sample set establishment module 801 is also used to obtain historical data of multiple photovoltaic inverters. According to the maintenance records of the multiple photovoltaic inverters, labels are added to the historical data of the multiple photovoltaic inverters for classification; the classified historical data of the multiple photovoltaic inverters are preprocessed to obtain a data sample set.

[0128] An embodiment of the present application further provides a storage medium, which includes a stored program, wherein the program executes any of the above methods when it is run.

[0129] Optionally, in this embodiment, the storage medium may be configured to store program codes for executing the following steps:

[0130] S1, obtaining historical data of multiple photovoltaic inverters, and establishing a data sample set based on the historical data.

[0131] S2, uses multi-scale convolution method to extract features from the data sample set and determine the fault feature set at different time scales.

[0132] S3, concatenating the fault feature set and the data sample set to obtain training samples, and using the training samples to train multiple classifiers in sequence.

[0133] S4, determining weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively.

[0134] S5, combining the multiple classifiers according to the weights corresponding to the multiple classifiers, and determining a fault prediction model for the photovoltaic inverter.

[0135] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0136] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0137] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0138] S1, obtaining historical data of multiple photovoltaic inverters, and establishing a data sample set based on the historical data.

[0139] S2, uses multi-scale convolution method to extract features from the data sample set and determine the fault feature set at different time scales.

[0140] S3, concatenating the fault feature set and the data sample set to obtain training samples, and using the training samples to train multiple classifiers in sequence.

[0141] S4, determining weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively.

[0142] S5, combining the multiple classifiers according to the weights corresponding to the multiple classifiers, and determining a fault prediction model for the photovoltaic inverter.

[0143] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0144] An embodiment of the present application further provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores the computer program product, and when the computer program is executed by a processor, the steps of the method in each embodiment of the present application are implemented.

[0145] Optionally, in this embodiment, the above computer program may be configured to implement the following steps when executed by a processor:

[0146] S1, obtaining historical data of multiple photovoltaic inverters, and establishing a data sample set based on the historical data.

[0147] S2, uses multi-scale convolution method to extract features from the data sample set and determine the fault feature set at different time scales.

[0148] S3, concatenating the fault feature set and the data sample set to obtain training samples, and using the training samples to train multiple classifiers in sequence.

[0149] S4, determining weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively.

[0150] S5, combining the multiple classifiers according to the weights corresponding to the multiple classifiers, and determining a fault prediction model for the photovoltaic inverter.

[0151] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0152] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0153] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for determining a fault prediction model for a photovoltaic inverter, characterized in that: The method comprises: Acquire historical data of a plurality of photovoltaic inverters, and establish a data sample set based on the historical data; A multi-scale convolution method is used to extract features from the data sample set to determine fault feature sets at different time scales; splicing the fault feature set and the data sample set to obtain training samples, and using the training samples to train multiple classifiers in sequence; Determine the weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively; The multiple classifiers are combined according to the weights respectively corresponding to the multiple classifiers to determine a fault prediction model for the photovoltaic inverter.

2. The method for determining a fault prediction model for a photovoltaic inverter according to claim 1, characterized in that: The method of extracting features from the data sample set by a multi-scale convolution method to determine fault feature sets at different time scales includes: Converting the data sample set into a vector form to obtain a data sample vector, wherein the data sample vector includes a time series of data of multiple different categories; Using convolution kernels of multiple different scales to extract features from the data sample vectors respectively, to obtain multiple groups of fault features of different scales; The multiple groups of fault features at different scales are concatenated to obtain the fault feature sets at different time scales.

3. The method for determining a fault prediction model for a photovoltaic inverter according to claim 1, characterized in that: The step of combining the fault feature set with the data sample set to obtain training samples, and using the training samples to train multiple classifiers in sequence, includes: Defining initial weights for a plurality of sample features in the training sample, wherein the initial weight of each sample feature is equal; Obtain multiple classifiers to be trained; Using the training sample to train a first classifier, wherein the first classifier is one of the multiple classifiers; Adjusting the initial weights of the plurality of sample features according to the classification results output by the trained first classifier; A classifier among the multiple classifiers that has not been trained is used as the next first classifier, and the step of training the first classifier using the training sample is returned to be executed until the multiple classifiers are all trained.

4. The method for determining a fault prediction model for a photovoltaic inverter according to claim 3, characterized in that: The determining the weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively includes: Determining classification error rates of the multiple classifiers respectively according to the classification results respectively output by the multiple classifiers; The weights corresponding to the multiple classifiers are determined according to the classification error rates of the multiple classifiers.

5. The method for determining a fault prediction model for a photovoltaic inverter according to any one of claims 1 to 4, characterized in that: The acquiring of historical data of a plurality of photovoltaic inverters and establishing a data sample set based on the historical data includes: Get historical data of multiple PV inverters; According to the maintenance records of the plurality of photovoltaic inverters, tags are added to the historical data of the plurality of photovoltaic inverters for classification; The classified historical data of the plurality of photovoltaic inverters are preprocessed to obtain the data sample set.

6. A photovoltaic inverter fault prediction method, characterized in that: Applying the fault prediction model of the photovoltaic inverter determined according to any one of claims 1 to 5, the method comprises: Obtaining current data of the photovoltaic inverter to be predicted; Inputting the current data of the photovoltaic inverter to be predicted into the fault prediction model of the photovoltaic inverter, determining the fault condition of the photovoltaic inverter and recording it; The failure probability of the photovoltaic inverter is determined according to the failure conditions of the photovoltaic inverter recorded within a preset time period.

7. A fault prediction model determination device for a photovoltaic inverter, characterized in that: The device comprises: A sample set establishment module, used to obtain historical data of multiple photovoltaic inverters and establish a data sample set based on the historical data; A feature extraction module, used to extract features from the data sample set using a multi-scale convolution method to determine fault feature sets at different time scales; A training module, used for concatenating the fault feature set with the data sample set to obtain training samples, and using the training samples to train multiple classifiers in sequence; A weight determination module, used to determine the weights corresponding to the multiple classifiers respectively based on the classification error rates of the multiple classifiers respectively; The combination module is used to combine the multiple classifiers according to the weights respectively corresponding to the multiple classifiers to determine the fault prediction model of the photovoltaic inverter.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method according to any one of claims 1 to 6 when executed.

9. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 6 through the computer program.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images