Photovoltaic string fault positioning and type diagnosis method based on dispersion rate

Abstract

The invention discloses a photovoltaic string fault positioning and type diagnosis method based on a dispersion rate, and belongs to the field of photovoltaic power station equipment abnormity diagnosis operation and maintenance. Comprising the steps of obtaining historical data of a photovoltaic string, including spatial configuration parameters, environmental monitoring data and electrical parameters, preprocessing the data, and performing normalization processing; performing data labeling on the processed data, dividing the processed data into normal data and abnormal data, and labeling an abnormal type of the abnormal data; and training an anomaly diagnosis classification model through the data so as to diagnose the photovoltaic string. According to the method, the discretization rate is used as important feature input of the diagnosis model, a discretization rate threshold value does not need to be set through manual experience, the abnormal position can be obtained, the abnormal type can be judged, and follow-up maintenance is facilitated; according to the method, the model is continuously updated through model automatic learning and incremental learning to adapt to the change of the state of the photovoltaic string, the positioning and maintenance efficiency is improved, the cost is reduced, and the generating capacity is increased.

Description

Technical Field

[0001] The present invention relates to the field of abnormal diagnosis and operation and maintenance of photovoltaic power station equipment, and particularly relates to a method for fault location and type diagnosis of photovoltaic strings based on the discrete rate. Background Art

[0002] The discrete rate is a statistical concept used to describe the degree of dispersion of data distribution. Generally, it refers to the degree to which data deviates from the normal level within a certain time or space range. By calculating the discrete rate, abnormal behaviors in the system can be effectively identified, and potential fault points can be located. Currently, in the fault location of photovoltaic equipment, the discrete rate can be used to evaluate the operation level of inverters or battery strings. By calculating the discrete rate of the inverter AC power, the difference degree of the AC output power of all inverters in the whole station can be measured, so as to locate the problematic inverter. In addition, the concept of the string current discrete rate has also been introduced to evaluate the overall operation of all branches of the inverter and help quickly locate the faults of the battery string branches. For example, "Detection Method, Device, Electronic Equipment and Storage Medium of Photovoltaic Power Generation Equipment", "Detection Method of Common Faults of a Photovoltaic Module", and "Abnormal Photovoltaic String Branch Identification Method, Device, Electronic Equipment and Storage Medium" all propose to compare the discrete rate with a preset threshold. If it is greater than the threshold, it is determined as an abnormality, and thus the target device is determined to be in an abnormal state.

[0003] "A Method and System for Identifying Inefficient Strings" proposes to perform clustering on the basis of comparing the discrete rate threshold to obtain the state of unknown strings, thereby realizing the location of inefficient strings.

[0004] "Abnormal Diagnosis Method, Device, Electronic Equipment and Storage Medium of Photovoltaic Strings" considers the influence of diagnostic conditions such as weather when calculating the discrete rate.

[0005] "Distributed Photovoltaic Fault Detection Method and Device Based on Discrete Rate" first determines abnormal inverters according to a preset discrete rate threshold, obtains the true discrete rate change amount through multi-layer screening of features such as weather and shadow, and classifies the true discrete rate change amount by using a support vector machine to obtain abnormal classified components.

[0006] It can be seen that the existing above methods have the following several problems:

[0007] 1. Setting the discrete rate threshold based on manual experience to obtain abnormal alternative components or directly determining them as abnormal components has a greater impact on the positioning accuracy;

[0008] 2. It can only obtain the approximate location of the abnormality and cannot determine or describe the type of the abnormality;

[0009] 3. Although conditions such as weather and shadows are considered, they are all achieved through manual screening and filtering, which affects the determination efficiency and increases the cost.

[0010] 4. Abnormalities such as data loss have a greater impact on the calculation of the discrete rate, and none of the above methods mention how to handle data loss, etc. to reduce the impact of the discrete rate on positioning. Summary of the Invention

[0011] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a method for photovoltaic string fault location and type diagnosis based on the discrete rate.

[0012] The purpose of the present invention is achieved through the following technical solutions: A method for photovoltaic string fault location and type diagnosis based on the discrete rate, including:

[0013] Obtain the historical data of the photovoltaic string, and the data includes spatial configuration parameters, environmental monitoring data, and electrical parameters; perform preprocessing on the data, and the preprocessing includes missing value completion and data standardization; perform normalization processing on the preprocessed data;

[0014] According to the data after normalization processing, extract string features; among them, the data includes normal data and abnormal data; calculate the discrete rate of the normal data, obtain its features, and jointly divide them into the normal data set; calculate the discrete rate of the abnormal data, obtain its features, and at the same time label the abnormal type of the abnormal data, and jointly divide them into the abnormal data set; the normal data set and the abnormal data set jointly constitute a fault diagnosis data set based on the discrete rate;

[0015] According to the fault diagnosis data set, use the discrete rate as a feature to input into a machine learning model, and train to obtain an abnormal diagnosis classification model;

[0016] Deploy the abnormal diagnosis classification model to the photovoltaic system, access the real-time string acquisition data, construct the features corresponding to the abnormal diagnosis classification model, and perform reasoning on the real-time data to output the diagnosis label and abnormal type corresponding to each string.

[0017] Further, the spatial configuration parameters include azimuth angle, tilt angle, and installation gap, the environmental monitoring data includes temperature, humidity, wind direction, and irradiance, and the electrical parameters include voltage and current value.

[0018] Further, the missing value completion adopts the KNN interpolation method.

[0019] Further, the machine learning model is a GBDT integrated tree model.

[0020] Further, the output of the diagnostic label and abnormal type corresponding to each string includes: the output diagnostic label and abnormal type are used as new data labels and input into the abnormal diagnosis classification model for incremental learning.

[0021] The beneficial effects of the present invention are as follows: Based on the discrete rate, the present invention does not take the discrete rate as the only string abnormality diagnosis index, but uses it as an important feature input of the diagnosis model. Through model learning, it can automatically identify the position of abnormal strings in a photovoltaic power station without the need for manual experience to set the discrete rate threshold, thereby reducing the impact on the positioning accuracy of abnormal strings.

[0022] Based on historical data, the present invention uses a model to identify the abnormal types of abnormal strings. It can not only obtain the position of the abnormality but also determine or describe the abnormal type, which is beneficial for providing more information for subsequent maintenance. For example, directly inputting the abnormal type into operation and maintenance assistants such as GPT can immediately obtain suggestions for maintenance and adjustment, thereby improving the maintenance efficiency.

[0023] The present invention comprehensively considers conditions such as weather and shadow, and realizes it through automatic learning of patterns by the model without manual screening and filtering, improving the positioning efficiency and reducing costs.

[0024] At the same time, the present invention continuously updates the model through incremental learning to adapt to the changes in the state of photovoltaic strings, providing more effective information for subsequent maintenance guidance, improving the maintenance efficiency, and increasing power generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0026] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art shall fall within the protection scope of the embodiments of the present application.

[0028] Such as Figure 1As shown in the figure, an embodiment of the present invention provides a photovoltaic string fault location and type diagnosis method based on the discrete rate, including the following steps:

[0029] Step 1: Historical data collection and processing. Build a digital model of the photovoltaic power station to obtain the historical collected data of the string, including spatial configuration parameters such as azimuth angle, tilt angle, installation gap, environmental monitoring data such as temperature, humidity, wind direction, irradiance, and electrical parameters such as voltage and current values. Perform corresponding preprocessing on the above data, including missing value filling, data standardization, etc. To eliminate the influence of missing values on the discrete rate calculation, imputation can be based on the KNN imputation method. First, it is necessary to calculate the distance between each data point in the dataset and the target missing value point. The distance calculation can be based on various distance metrics, such as Euclidean distance or Manhattan distance. Select the K nearest data points as neighbors, and the feature values of these neighbors will be used to estimate the missing value. Once the K nearest neighbors are determined, the average or majority value of these neighbors on the missing feature can be calculated, and this estimated value is used to fill the missing point in the original dataset. For continuous variables, the average of the K nearest neighbors is usually calculated; for discrete variables, the most frequent category is selected by voting as the imputation value. Finally, the data is normalized.

[0030] Step 2: Construction of a fault diagnosis dataset based on the discrete rate. Extract string features and construct a fault diagnosis dataset based on the discrete rate. The construction of the dataset includes a normal dataset and an abnormal dataset. In the historical collected data, it is known whether the data is normal or abnormal when obtained (such as through patrol inspection, etc.). If it is normal data, calculate the discrete rate of the data, and at the same time obtain other features such as temperature, humidity, and irradiance, and label it as normal data; if it is abnormal data, calculate the discrete rate of the data, and at the same time obtain other features such as temperature, humidity, and irradiance, and label its abnormal category as the abnormal type of the string, including fan failure, DC overvoltage protection, PV polarity reverse connection protection, island protection, module overheating protection, leakage current protection, etc.; the normal dataset and the abnormal dataset constitute a fault diagnosis dataset based on the discrete rate.

[0031] Among them, different from the original discrete rate threshold comparison method, the present invention takes the discrete rate as an important feature and inputs it into the subsequent model, allowing the algorithm to automatically learn the patterns therein without the need for manual threshold setting. Exemplarily, the current / voltage data of each photovoltaic string within a specific time period (15 minutes) is collected, and the average value and standard deviation of the current / voltage of each string are calculated. According to the discrete rate calculation formula: discrete rate = standard variance of string data / average value of string data * 100%. Thus, the current discrete rate and voltage discrete rate of the corresponding string can be obtained. On this basis, the average value data of the azimuth angle, tilt angle, installation gap, etc. of the string within this time period, as well as the average values of the corresponding weather data such as temperature, humidity, irradiance, and the coding of the string, are used as the features of the string.

[0032] Step 3: Construction of a fault diagnosis model based on the discrete rate. Using the discrete rate, weather conditions, string features, etc. as the basic features of the dataset, input them into a machine learning model to train the corresponding anomaly diagnosis classification model, such as the GBDT ensemble tree model. GBDT (Gradient Boosting Decision Tree), also known as MART (Multiple Additive Regression Tree) in Chinese, is an iterative decision tree algorithm. It constructs a series of weak learners (usually decision trees) and accumulates the prediction results of these weak learners to form a powerful prediction model. The core idea of GBDT is to iteratively construct multiple decision trees in a gradient descent manner. In each round of iteration, it calculates the residuals of the current model (i.e., the gradient of the loss function with respect to the model prediction values) and uses these residuals as the target variables to train a new decision tree. The newly trained decision tree aims to fit these residuals and then add its output to the current model with a certain learning rate (shrinkage) coefficient to form an updated model. This process is repeated multiple times until the preset number of iterations is reached or the stopping condition is met.

[0033] Step 4: Inference and incremental learning of the fault diagnosis model based on the discrete rate. Deploy the trained anomaly diagnosis classification model to the photovoltaic system, access the real-time string acquisition data, construct the corresponding features of the model, and perform real-time data inference to output the diagnostic label corresponding to each string, thereby simultaneously realizing the diagnosis and location of abnormal strings in the photovoltaic power station (the numbers of the faulty strings) and the output of the abnormal types. The output abnormal data can also be used as new data labels to perform incremental learning on the model and update the model parameters. The updated model continues to be deployed for inference, continuously improving the effect of the diagnosis model, providing more effective information for subsequent maintenance guidance, improving the maintenance efficiency, and increasing the power generation.

[0034] The above embodiments are only used to illustrate the design concept and characteristics of the present invention, and the purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design concepts disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A photovoltaic string fault location and type diagnosis method based on the discrete rate, characterized in that, It includes the following steps: Obtain historical data of the photovoltaic string, where the data includes spatial configuration parameters, environmental monitoring data, and electrical parameters; preprocess the data, and the preprocessing includes missing value filling and data standardization; Perform normalization processing on the preprocessed data; Extract string features based on the normalized data; among them, the data includes normal data and abnormal data; calculate the discrete rate of the normal data, obtain its features, and jointly divide them into the normal data set; calculate the discrete rate of the abnormal data, obtain its features, and at the same time label the abnormal types of the abnormal data, and jointly divide them into the abnormal data set; the normal data set and the abnormal data set together constitute a fault diagnosis data set based on the discrete rate; According to the fault diagnosis data set, use the discrete rate as a feature to input into a machine learning model, and train to obtain an abnormal diagnosis classification model; Deploy the abnormal diagnosis classification model to the photovoltaic system, access real-time string acquisition data, construct features corresponding to the abnormal diagnosis classification model, and perform inference on real-time data to output the diagnosis label and abnormal type corresponding to each string.

2. The method for photovoltaic string fault location and type diagnosis based on the discrete rate according to claim 1, wherein The spatial configuration parameters include azimuth angle, tilt angle, and installation clearance, the environmental monitoring data includes temperature, humidity, wind direction, and irradiance, and the electrical parameters include voltage and current value.

3. A method for photovoltaic string fault location and type diagnosis based on the discrete rate according to claim 1, characterized in that, The missing value filling adopts the KNN interpolation method.

4. A method for photovoltaic string fault location and type diagnosis based on the discrete rate according to claim 1, characterized in that, The machine learning model is a GBDT integrated tree model.

5. A method for photovoltaic string fault location and type diagnosis based on the discrete rate according to claim 1, characterized in that, The output of the diagnosis label and abnormal type corresponding to each string includes: the output diagnosis label and abnormal type are used as new data labels and input into the abnormal diagnosis classification model for incremental learning.

Images