A method for assessing dust accumulation in photovoltaic power plants from particle size to string-level hourly level

Abstract

The present invention discloses a method for evaluating the dust accumulation of photovoltaic power stations with a granularity of hourly string level, and relates to the technical field of photovoltaic power station dust accumulation evaluation. The method comprises the following steps: obtaining the actual operation information of the photovoltaic power station and the construction drawings to clarify the affiliation between the string and the inverter to construct the string-level topological structure; collecting the historical operation data of the photovoltaic power station to implement the labeling operation, and selecting the characteristic variables to judge the string current, and setting the dust accumulation degree state classification basis according to the judgment result; building a training string-level hourly dust accumulation evaluation model based on the convolutional neural network algorithm, characteristic variables and state classification basis; inputting the real-time collected data into the dust accumulation evaluation model to predict the string current, and comparing the predicted value with the actual value to evaluate the dust accumulation state. The present invention realizes real-time monitoring and prediction of dust accumulation, improves the operation efficiency of the photovoltaic power station, and realizes effective management of dust accumulation problems while reducing costs.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station dust accumulation assessment, and in particular to a photovoltaic power station dust accumulation assessment method with a particle size down to a string level and hour level. Background Art

[0002] Photovoltaic power stations, also known as solar power stations, are power generation facilities that use photovoltaic modules to directly convert sunlight into electrical energy. They are an important part of renewable energy technology and are of great significance for promoting energy structure transformation and achieving low-carbon economic development. The main components of photovoltaic power stations include photovoltaic modules (solar panels), support systems, DC combiner boxes, inverters, AC combiner boxes, monitoring systems and grid connection facilities. Photovoltaic modules are the core part, responsible for converting sunlight into direct current; inverters convert direct current into alternating current for use in the power grid or directly in powering consumer devices.

[0003] During the use of photovoltaic power stations, the dust accumulation problem will seriously affect the power generation performance and economic benefits of photovoltaic modules. At present, the dust accumulation problem of photovoltaic power stations is usually solved by regular manual cleaning. This includes regular cleaning of the surface of photovoltaic modules to remove dust and ensure that the modules can fully absorb sunlight and maintain high power generation efficiency. In addition, there are also some automatic cleaning systems, such as robots or water flushing systems, which are used in some large photovoltaic power stations.

[0004] However, the above dust cleaning operation has the following defects:

[0005] 1) Dependence on manual cleaning: Manual cleaning requires regular manpower input and is subject to factors such as weather conditions and labor costs. It may lead to untimely cleaning and affect the power generation efficiency of the photovoltaic power station.

[0006] 2) Cost of automatic cleaning system: The investment and operation and maintenance costs of the automatic cleaning system are high, making it difficult for small photovoltaic power stations or those with limited economic conditions to adopt automatic cleaning technology.

[0007] 3) Lack of real-time monitoring and prediction: Traditional cleaning methods lack real-time monitoring and prediction mechanisms, making it difficult to detect dust accumulation problems in advance, which may lead to power generation losses.

[0008] 4) Unable to accurately assess the degree of dust accumulation: Traditional cleaning methods make it difficult to accurately assess the degree of dust accumulation on photovoltaic modules, resulting in the inability to make precise adjustments to the timing and intensity of cleaning. Summary of the invention

[0009] In order to solve the above problems, the present invention proposes a photovoltaic power station dust accumulation assessment method with a granularity down to the string level and hour level, thereby achieving real-time monitoring and prediction of the dust accumulation situation.

[0010] In order to achieve the above object, the present invention adopts the following technical solution:

[0011] The present invention provides a method for evaluating dust accumulation in a photovoltaic power station with a granularity down to the hourly level of a string group. The method for evaluating dust accumulation in a photovoltaic power station with a granularity down to the hourly level of a string group comprises the following steps:

[0012] S1. Obtain the actual operation information and construction drawings of the photovoltaic power station to clarify the relationship between the strings and inverters and build a string-level topology structure;

[0013] S2. Collect historical operation data of the photovoltaic power station to perform labeling operations, select characteristic variables to determine the string current, and set the basis for classifying the dust accumulation degree according to the determination results;

[0014] S3. Build a training group cascade hourly dust accumulation assessment model based on convolutional neural network algorithm, characteristic variables and state classification;

[0015] S4. Input the real-time collected data into the dust accumulation assessment model to predict the string current, and compare the predicted value with the actual value to evaluate the dust accumulation status.

[0016] Preferably, obtaining the actual operation information of the photovoltaic power station and the construction drawings to clarify the affiliation between the strings and the inverters and constructing the string-level topology structure includes the following steps:

[0017] S11. Build a power station database according to the construction drawings, and obtain the actual operation information on site through the power station database, the operation information including the type of photovoltaic power station components, the number of strings, the string connection method, the inverter type, the number of inverters and the inverter connection method;

[0018] S12. Implement graph data processing according to the operation information, design a dynamic topology structure including strings and inverters, and adjust the dynamic topology structure using a depth-first traversal method combined with an optimization algorithm;

[0019] S13. Use the dynamic topology structure to clarify the relationship between the string and the inverter, and analyze the actual access status of the string branch.

[0020] Preferably, implementing graph data processing according to the operation information, designing a dynamic topology structure including strings and inverters, and adjusting the dynamic topology structure using a depth-first traversal method combined with an optimization algorithm includes the following steps:

[0021] S121, using the power station database to convert the operation information, classify the operation information into two types: string and inverter, and design an intelligent recognition layer inside the power station database to establish a dynamic topology structure;

[0022] S122, using an optimization algorithm to set the strings as group points, each group of string points as a single node, and the access relationship between the strings and the inverter, and optimizing the determination speed of the dynamic topology structure according to the setting results;

[0023] S123, using a depth-first traversal method to regard the beginning of the group string point as the initial point, and traversing and marking each group string point one by one starting from the initial point;

[0024] S124. Determine whether the group string point is visited multiple times or not visited during the traversal process according to the marking result. If yes, return to step S121 to re-establish the dynamic topology structure. If no, complete the establishment of the dynamic topology structure.

[0025] Preferably, collecting historical operation data of the photovoltaic power station to implement labeling operation, selecting characteristic variables to judge the string current, and setting the dust accumulation degree state classification basis according to the judgment result include the following steps:

[0026] S21, obtaining the historical operation data of the photovoltaic power station to implement abnormal value processing operations, and performing label adding operations on the historical data of the strings in the clean time period;

[0027] S22. Use data management theory to establish a data conversion model, and make the implicit knowledge of multi-label data explicit. Use feature selection algorithm to extract input feature variables, including solar radiation intensity, ambient temperature, wind speed and humidity.

[0028] S23, using the maximum information coefficient and mutual information to calculate the correlation between the output current of each string and the input characteristic variable;

[0029] S24, comparing the string current judgment result with the current of a standard battery panel in a clean state to set a dust accumulation degree state classification, and setting corresponding operation measures.

[0030] Preferably, the data transformation model is established by using data management theory, and the implicit knowledge of multi-label data is made explicit. The input feature variables are extracted by using feature selection algorithm, which includes the following steps:

[0031] S221. Use data management theory combined with multi-label data to generate four groups of independent individuals as the model basis for dynamic knowledge conversion. The four groups of independent individuals include label exchange individuals, data clarification individuals, data explicit individuals, and data conversion individuals.

[0032] S222, using the label communication individuals to perform data consistency detection on multiple data labels, and the data clarification individuals to integrate the multi-label data to promote label clarification;

[0033] S223, using data explicit individuals to obtain explicit knowledge of multiple data labels to enhance implicit knowledge, and using data conversion individuals to complete data conversion;

[0034] S224. Randomly select samples from the multi-label data, find neighboring samples of the samples, calculate the feature weights of multiple sample points, and set a threshold value according to the feature weight sorting result to extract the input feature variable.

[0035] Preferably, calculating the correlation between each string output current and the input characteristic variable using the maximum information coefficient and the mutual information comprises the following steps:

[0036] S231, using mutual information to evaluate the mutual dependence between the input feature vector and the string output current, and determining the mutual information value according to the evaluation result;

[0037] S232. Measure the strength of the nonlinear relationship between the output currents of each string according to the mutual information value combined with the maximum information coefficient, and calculate the characteristic variable weight coefficient of the degree of relationship between the input characteristic variable and the output current by using a grid division method.

[0038] Preferably, the calculation formula for the relationship between the input characteristic variable and the output current is:

[0039]

[0040] In the formula, MIC(X, Y) represents the relationship between the input characteristic variable and the output current; X represents the input characteristic variable; Y represents the output current of each string; G represents the grid division set; g represents the number of grid divisions; I(X g , Y g ) represents the mutual information between X and Y; X g With Y g Represent the discretized versions of X and Y on G respectively.

[0041] Preferably, the dust accumulation degree state includes a first level dust accumulation degree, a second level dust accumulation degree and a third level dust accumulation degree;

[0042] The first level of dust accumulation is when the difference between the string output current and the current setting value of the standard solar panel is between 1 and 1.2 times, indicating that dust accumulation exists;

[0043] The second level of dust accumulation is when the difference between the string output current and the current setting value of the standard solar panel is between 1.2 and 1.5 times, indicating that dust accumulation is more obvious;

[0044] The third level of dust accumulation is when the difference between the string output current and the current setting value of the standard solar panel is more than 1.5 times, indicating serious dust accumulation.

[0045] Preferably, building a training group cascade hourly dust accumulation assessment model based on a convolutional neural network algorithm, characteristic variables and state classification comprises the following steps:

[0046] S31, constructing a matrix as an input matrix according to the input characteristic variables of each string group at each hour, and using the output current of each string group at each hour as a label to form a data set;

[0047] S32, randomly shuffling the order of the data set, standardizing it and dividing it into a training set and a test set, and setting the characteristic variables as model inputs, and the state classification basis and the string current as model outputs;

[0048] S33. Use the convolutional neural network structure to build a network model, and use the mean square error as the loss function to train the network model to complete the construction of the group-level hourly dust accumulation assessment model.

[0049] Preferably, a convolutional neural network structure is used to build a network model, and the mean square error is used as a loss function to train the network model to complete the construction of the group string hourly dust accumulation assessment model, which includes the following steps:

[0050] S331, using the input layer in the convolutional neural network structure to receive the input matrix, and using the convolutional layer, the activation layer and the pooling layer to output the feature map and connect it with the fully connected layer;

[0051] S332, using a fully connected layer to flatten the feature map into a one-dimensional vector and combine it with the maximum information coefficient, and using the output layer to set the predicted value of the neuron output string current to build a network model;

[0052] S333. Taking the mean square error as the loss function, the Adam algorithm is used to train the network model and adjust the network parameters to complete the construction of the group string-level hourly dust accumulation assessment model.

[0053] The beneficial effects of the present invention are:

[0054] 1. The present invention proposes a dust accumulation assessment method based on hourly data at the group string level. By constructing a model and utilizing technologies such as the maximum information coefficient, real-time monitoring and prediction of dust accumulation are achieved. Specifically, advanced technologies such as convolutional neural networks are utilized to realize intelligent analysis of hourly data at the group string level, thereby improving the accuracy of dust accumulation status. Moreover, through processing real-time data, dust accumulation problems of photovoltaic modules can be discovered in a timely manner, which helps to take cleaning measures early and reduce power generation losses. Compared with the automatic cleaning system, dust accumulation can be monitored without adding significant additional costs, thereby hopefully solving the shortcomings of traditional cleaning methods, improving the operating efficiency of photovoltaic power stations, and achieving effective management of dust accumulation problems while reducing costs.

[0055] 2. The present invention proposes a dust accumulation assessment model based on hourly string-level data. By collecting historical data and using the maximum information coefficient technology to construct a convolutional neural network model, the model is used to predict the output current of photovoltaic strings, so that the dust accumulation situation can be evaluated at a time granularity of minutes, thereby realizing fine-grained monitoring of the performance of photovoltaic power stations. The minute-level data of each hour and each string, solar radiation intensity, ambient temperature, wind speed, and humidity are constructed into a matrix, and analyzed through a deep learning model, so that the assessment of the dust accumulation situation is more accurate and real-time.

[0056] 3. When establishing the model, the present invention comprehensively considers the influence of meteorological factors such as solar radiation intensity, ambient temperature, wind speed, humidity, etc. on the string output current, so that the model is more adaptable and can better adapt to the operation of photovoltaic power stations under different meteorological conditions, and proposes classification standards for dust accumulation status, including slight dust accumulation, moderate dust accumulation and severe dust accumulation, and recommends optimizing and adjusting the classification standards through real-time feedback from on-site personnel through a real-time feedback system.

[0057] 4. The classification standard and feedback system proposed in the present invention improve the accuracy of dust accumulation status and the intelligence of the system, make up for the shortcomings of traditional methods, improve the accuracy and real-time performance of dust accumulation status, and thus provide an innovative solution for the operation and maintenance management of photovoltaic power stations, which can monitor and predict in real time. Through the processing of real-time data, the dust accumulation problem of photovoltaic modules can be discovered in time, which helps to take timely cleaning measures and minimize power generation losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0059] Figure 1 The present invention is a flowchart of a method for evaluating dust accumulation in a photovoltaic power station with a granularity down to the hourly level of a string level. DETAILED DESCRIPTION

[0060] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0061] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0062] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device 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 devices.

[0063] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0064] See also Figure 1 The present invention provides a photovoltaic power station dust accumulation assessment method with a granularity of group string level and hour level, and the photovoltaic power station dust accumulation assessment method with a granularity of group string level and hour level comprises the following steps:

[0065] S1. Obtain the actual operation information of the photovoltaic power station and the construction drawings to clarify the relationship between the strings and inverters and build the string-level topology structure.

[0066] In this embodiment, obtaining the actual operation information of the photovoltaic power station and the construction drawings to clarify the affiliation between the strings and the inverters and constructing the string-level topology structure includes the following steps:

[0067] S11. Build a power station database according to the construction drawings, and obtain the actual operation information on site through the power station database. The operation information includes the type of photovoltaic power station components, the number of strings, the string connection method, the inverter type, the number of inverters and the inverter connection method.

[0068] Specifically, building a power plant database according to the construction drawings and obtaining on-site actual operation information through the power plant database includes the following steps:

[0069] According to the construction drawings, obtain the PV power station component type, string quantity, string connection method, inverter type, inverter quantity and inverter connection method, and clarify the type of operation information that needs to be collected and monitored.

[0070] Based on the structure and monitoring requirements of the photovoltaic power station, the database architecture is designed, the table structure, fields and data types of the database are determined, and a program for automatically collecting data from the inverter is developed to implement data preprocessing functions.

[0071] Implement the database using an appropriate database management system (including MySQL, MongoDB) based on the designed architecture, and import the static information in the construction drawings into the database as basic data.

[0072] S12. Implement graph data processing according to the operation information, design a dynamic topology structure including strings and inverters, and adjust the dynamic topology structure using a depth-first traversal method combined with an optimization algorithm.

[0073] Specifically, the graph data processing is implemented according to the operation information, the dynamic topology structure including the strings and inverters is designed, and the dynamic topology structure is adjusted by using the depth-first traversal method combined with the optimization algorithm, including the following steps:

[0074] S121, using the power station database to convert the operation information, classify the operation information into two types: string and inverter, and design an intelligent recognition layer inside the power station database to establish a dynamic topology structure;

[0075] S122, using an optimization algorithm to set the strings as group points, each group of string points as a single node, and the access relationship between the strings and the inverter, and optimizing the determination speed of the dynamic topology structure according to the setting results;

[0076] S123, using a depth-first traversal method to regard the beginning of the group string point as the initial point, and traversing and marking each group string point one by one starting from the initial point;

[0077] S124. Determine whether the group string point is visited multiple times or not visited during the traversal process according to the marking result. If yes, return to step S121 to re-establish the dynamic topology structure. If no, complete the establishment of the dynamic topology structure.

[0078] S13. Use the dynamic topology structure to clarify the relationship between the string and the inverter, and analyze the actual access status of the string branch.

[0079] Specifically, according to the optimized topology, the affiliation between each string and its corresponding inverter is clarified, the actual access status of the string branches is analyzed, and any potential configuration errors or optimization opportunities are identified.

[0080] Among them, the topology diagram of the actual on-site access of the inverter and the string is shown in Table 1:

[0081] Table 1: Actual on-site connection of inverters and strings

[0082]

[0083] In the table, a value of 1 indicates that the inverter string is connected, and a value of 0 indicates that it is not connected.

[0084] The configuration file information of each inverter is as follows:

[0085] Table 2: Configuration file information of the inverter

[0086] Serial number Inverter name Capacity(kW) Number of strings Number of components Axis Type 1 NB01 500 16 24 Flat single shaft 2 NB02 200 8 16 Dual Axis

[0087] Specifically, Tables 1 and 2 scientifically locate the information of each string of the photovoltaic power station, providing a data basis for the subsequent model establishment.

[0088] S2. Collect historical operation data of the photovoltaic power station to implement labeling operations, select characteristic variables to determine the string current, and set the basis for dust accumulation degree status classification based on the determination results.

[0089] In this embodiment, collecting historical operation data of the photovoltaic power station to implement labeling operation, selecting characteristic variables to judge the string current, and setting the dust accumulation degree state classification basis according to the judgment result include the following steps:

[0090] S21, obtaining historical operation data of the photovoltaic power station to perform abnormal value processing operations, and performing label adding operations on historical data of the strings in the clean time period.

[0091] Specifically, obtaining the historical operation data of the photovoltaic power station to implement the abnormal value processing operation, and implementing the label adding operation for the historical data of the strings in the clean time period includes the following steps:

[0092] The historical operation data is obtained from the data management system of the photovoltaic power station, and the collected data is preliminarily processed, including format conversion, scaling, and using statistical methods to identify outliers.

[0093] Analyze the identified outliers to determine whether they are real anomalies, decide on the treatment method based on the nature of the outliers, and add labels based on the data in the identified clean time period to ensure the consistency and accuracy of the labels.

[0094] S22. Use data management theory to establish a data conversion model, and make the implicit knowledge of multi-label data explicit. Use feature selection algorithm to extract input feature variables, which include solar radiation intensity, ambient temperature, wind speed and humidity.

[0095] Specifically, the data transformation model is established by using data management theory, and the implicit knowledge of multi-label data is made explicit. The input feature variables are extracted by using feature selection algorithm, which includes the following steps:

[0096] S221. Use data management theory combined with multi-label data to generate four groups of independent individuals as the model basis for dynamic knowledge conversion. The four groups of independent individuals include label exchange individuals, data clarification individuals, data explicit individuals, and data conversion individuals.

[0097] S222, using the label communication individuals to perform data consistency detection on multiple data labels, and the data clarification individuals to integrate the multi-label data to promote label clarification;

[0098] S223, using data explicit individuals to obtain explicit knowledge of multiple data labels to enhance implicit knowledge, and using data conversion individuals to complete data conversion;

[0099] S224. Randomly select samples from the multi-label data, find neighboring samples of the samples, calculate the feature weights of multiple sample points, and set a threshold value according to the feature weight sorting result to extract the input feature variable.

[0100] S23. Calculate the correlation between the output current of each string and the input characteristic variable using the maximum information coefficient and the mutual information.

[0101] Specifically, using the maximum information coefficient and mutual information to calculate the correlation between each string output current and the input characteristic variable includes the following steps:

[0102] S231, using mutual information to evaluate the mutual dependence between the input feature vector and the string output current, and determining the mutual information value according to the evaluation result;

[0103] S232. Measure the strength of the nonlinear relationship between the output currents of each string according to the mutual information value combined with the maximum information coefficient, and calculate the characteristic variable weight coefficient of the degree of relationship between the input characteristic variable and the output current by using a grid division method.

[0104] Among them, the calculation formula for the relationship between the input characteristic variable and the output current is:

[0105]

[0106] In the formula, MIC(X, Y) represents the relationship between the input characteristic variable and the output current; X represents the input characteristic variable; Y represents the output current of each string; G represents the grid division set; g represents the number of grid divisions; I(X g ; Y g ) represents the mutual information between X and Y; X g With Y g Represent the discretized versions of X and Y on G respectively.

[0107] Specifically, the calculation formula of the mutual information of X and Y is as follows:

[0108]

[0109] Where p(x, y) represents the joint probability of X and Y, and p(x) and p(y) represent the marginal probability distributions of X and Y, respectively.

[0110] S24, comparing the string current judgment result with the current of a standard battery panel in a clean state to set a dust accumulation degree state classification, and setting corresponding operation measures.

[0111] Specifically, multiple standard panels are maintained at the photovoltaic power station site. The standard panels are cleaned to keep them in a clean state. The current of the clean standard panels is compared with the current of the actual components. The dust accumulation degree of the components is evaluated in three categories, which are described as follows:

[0112] 1) Slight dust accumulation (level 1 dust accumulation)

[0113] The slight dust accumulation state means that the standard panel measured value of the PV string output current is between 1-1.2 times the actual value, but still within the acceptable range. This may indicate that slight dust accumulation affects the performance of the string, but has not yet reached the level that requires immediate cleaning measures.

[0114] 2) Moderate dust accumulation (secondary dust accumulation level)

[0115] Moderate dust accumulation means that the standard panel measurement value of the PV string is between 1.2-1.5 times the actual value, indicating that the dust accumulation of the string is more obvious. In this case, appropriate cleaning actions may be required to reduce the negative impact of dust accumulation on power generation performance and ensure the efficient operation of the PV power station.

[0116] 3) Severe dust accumulation (level 3 dust accumulation)

[0117] Severe dust accumulation means that the standard panel measurement value of the PV string is more than 1.5 times the actual value, indicating that the dust accumulation problem is very serious. At this time, it is crucial to take cleaning measures in time to restore the normal power generation performance of the string and prevent potential equipment damage.

[0118] Therefore, by establishing a data conversion model and making the implicit knowledge of multi-label data explicit, the intrinsic connections and patterns between the data can be revealed. At the same time, the maximum information coefficient and mutual information are used to calculate the correlation between the input characteristic variables and the string output current. The main factors affecting the current change can be accurately identified. By comparing the judgment result of the string current with the standard state, the degree of dust accumulation can be accurately judged, thus providing a basis for maintenance and cleaning work.

[0119] S3. Build a training group cascade hourly dust accumulation assessment model based on the convolutional neural network algorithm, characteristic variables and state classification.

[0120] In this embodiment, building a training group cascade hourly dust accumulation assessment model based on a convolutional neural network algorithm, characteristic variables, and state classification includes the following steps:

[0121] S31. A matrix is ​​constructed according to the input characteristic variables of each string group every hour as an input matrix, and the output current of each string group every hour is used as a label to form a data set.

[0122] Specifically, constructing a matrix as an input matrix according to the input feature variables of each string per hour, and using the output current of each string per hour as a label to form a data set includes the following steps:

[0123] Set the minute-level data of each string every hour, solar radiation intensity, ambient temperature, wind speed, and humidity to construct a 4*60 matrix as the input matrix.

[0124] The output current of each string per hour is set as the label, and the input matrix is ​​merged with the label to generate a data set.

[0125] S32, randomly shuffle the order of the data set, standardize it and divide it into a training set and a test set, and set the characteristic variables as model inputs, and the state classification basis and the string current as model outputs.

[0126] S33. Use the convolutional neural network structure to build a network model, and use the mean square error as the loss function to train the network model to complete the construction of the group-level hourly dust accumulation assessment model.

[0127] Specifically, the network model is built using a convolutional neural network structure, and the mean square error is used as a loss function to train the network model to complete the construction of the group string hourly dust accumulation assessment model, which includes the following steps:

[0128] S331. Use the input layer in the convolutional neural network structure to receive the input matrix, and use the convolutional layer, activation layer and pooling layer to output the feature map and connect it with the fully connected layer.

[0129] Specifically, using the input layer in the convolutional neural network structure to receive the input matrix, and using the convolutional layer, activation layer and pooling layer to output the feature map and connect it with the fully connected layer includes the following steps:

[0130] The input layer is used to receive the 4*60 input matrix, and the convolution layer 1 is used to use 16 3*3 convolution kernels, the step size is 1, the padding is the same, and the output is 16 4*60 feature maps. Then the activation layer 1 is used, and the ReLU function is used to output 16 4*60 feature maps.

[0131] Use the maximum pooling in pooling layer 1 with a pooling kernel size of 2 and a step size of 2 to output 16 2*30 feature maps. Use convolution layer 2 with 32 3*3 convolution kernels, a step size of 1, and padding of the same to output 32 2*30 feature maps. Use activation layer 2 and ReLU function to output 32 2*30 feature maps. Use the maximum pooling in pooling layer 2 with a pooling kernel size of 2 and a step size of 2 to output 32 1*15 feature maps.

[0132] S332. Use the fully connected layer to flatten the feature map into a one-dimensional vector and combine it with the maximum information coefficient, and use the output layer to set the predicted value of the neuron output string current to build a network model.

[0133] The fully connected layer is used to flatten the feature map into a one-dimensional vector and combine it with the maximum information coefficient, and the output layer is used to set the predicted value of the neuron output string current. The network model is built in the following steps:

[0134] The 32 1*15 feature maps are flattened into a one-dimensional vector with a length of 480 using a fully connected layer, and combined with the maximum information coefficient result, and the length is 484. The output layer uses a neuron to output the predicted value of the string current to set the predicted value of the neuron output string current to build a network model.

[0135] S333. Take the mean square error as the loss function, use the Adam algorithm to train the network model, adjust the network parameters, and complete the construction of the group string-level hourly dust accumulation assessment model.

[0136] Specifically, the mean square error is used as the loss function, and the Adam algorithm is used to train the network model and adjust the network parameters to complete the construction of the group string hourly dust accumulation assessment model, including:

[0137] The mean square error (MSE) is used as the loss function, the Adam algorithm is used to train the model, and the network parameters are adjusted so that the model can accurately predict the string current.

[0138] The formula of MSE corresponds to the following:

[0139]

[0140] In the formula, y i is the ith observation, represents the corresponding predicted value, and n represents the number of observations.

[0141] The Adam algorithm formula corresponds to the following:

[0142]

[0143] m t =β1m t-1 +(1-β1)g t ;

[0144]

[0145]

[0146]

[0147]

[0148] In the formula, θ t represents the parameters of step t, f t represents the objective function of step t, g t represents the gradient of the tth step, m t and v t denotes the first-order and second-order moment estimates at step t, and represents the bias-corrected moment estimate, a represents the learning rate, β1 and β2 represent the decay rate of the moment estimate, and ε represents a small constant used to prevent division by zero errors.

[0149] Therefore, by constructing an input feature variable matrix and using the output current as a label to form a data set, this structured data representation method can effectively capture the spatial characteristics of time series data and provide a suitable input format for the convolutional neural network. The network model built using the convolutional neural network structure can effectively process time series data and improve the accuracy and reliability of dust accumulation assessment.

[0150] S4. Input the real-time collected data into the dust accumulation assessment model to predict the string current, and compare the predicted value with the actual value to evaluate the dust accumulation status.

[0151] In this embodiment, inputting the real-time collected data into the dust accumulation assessment model to predict the string current, and comparing the predicted value with the actual value to evaluate the dust accumulation state includes the following steps:

[0152] The required characteristic variable data, including solar radiation intensity, ambient temperature, wind speed and humidity, are collected in real time from the monitoring system of the photovoltaic power station, and the collected real-time data are preprocessed to ensure that the data format and range are consistent with the input requirements of the dust accumulation assessment model.

[0153] The preprocessed data is used as input into the trained dust accumulation assessment model. The model will predict the string current based on the input real-time data, collect the model's prediction results, and properly process the predicted string current values ​​for comparison with the actual current values. The actual string current values ​​at the corresponding time point are obtained from the monitoring system as a reference for the prediction results.

[0154] The predicted string current value is compared with the actual current value to evaluate the difference between the two. The size and characteristics of the difference can be used to evaluate the dust accumulation state of the string. According to the difference between the predicted value and the actual value, the dust accumulation state of the string is determined using a preset judgment rule or threshold.

[0155] Based on the evaluation results of the dust accumulation status, corresponding cleaning or maintenance suggestions are put forward, and the evaluation results and implemented operations are fed back to the monitoring system and operation and maintenance team for future decision-making and model optimization.

[0156] In order to ensure the stability and accuracy of the dust accumulation status classification system, continuous optimization and improvement are required. Through real-time feedback from on-site personnel, the classification standards can be adjusted according to the actual cleaning effect. By accumulating lessons learned, the system can be made more adaptable and intelligent, and the accuracy of identifying various dust accumulation conditions can be improved.

[0157] The specific process of the present invention is embodied as follows:

[0158] 1. Data collection and preprocessing:

[0159] Collect historical data of the photovoltaic power station, including minute-level data of the strings, solar radiation intensity, ambient temperature, wind speed, and humidity, and organize the data into a matrix form for each hour and each string as input data.

[0160] 2. Model feature variable identification:

[0161] The maximum information coefficient (MIC) is used to calculate the correlation between the output current of each string and the solar radiation intensity, ambient temperature, wind speed, and humidity in each hour, and obtain the characteristic variable weight coefficient.

[0162] 3. Network construction and training:

[0163] A convolutional neural network (CNN) model is built to receive a 4*60 input matrix.

[0164] Set the convolution layer, activation layer, pooling layer, fully connected layer and output layer, adjust the layer parameters, use mean square error (MSE) as the loss function, use the Adam algorithm to train the model, and adjust the network parameters.

[0165] 4. Hourly dust accumulation status assessment of the group:

[0166] Based on the relationship between the predicted value and the actual value, the dust accumulation status is judged as mild, moderate or severe.

[0167] 5. On-site feedback and model optimization and adjustment:

[0168] A real-time feedback system is established to enable on-site personnel to provide real-time feedback on cleaning results and actual conditions. Based on the real-time feedback, the classification standards are adjusted and the model is optimized.

[0169] To sum up, with the help of the above-mentioned technical scheme of the present invention, the present invention proposes a dust accumulation assessment method based on group string hourly data, and realizes real-time monitoring and prediction of dust accumulation by constructing a model and utilizing technologies such as maximum information coefficient. Specifically, advanced technologies such as convolutional neural networks are utilized to realize intelligent analysis of group string hourly data, thereby improving the accuracy of dust accumulation status. Moreover, through the processing of real-time data, dust accumulation problems of photovoltaic modules can be discovered in time, which helps to take cleaning measures as early as possible and reduce power generation losses. Compared with the automatic cleaning system, dust accumulation can be monitored without adding significant additional costs, thereby solving the shortcomings of traditional cleaning methods, improving the operating efficiency of photovoltaic power stations, and realizing effective management of dust accumulation problems while reducing costs. The present invention proposes a dust accumulation assessment model based on hourly string-level data. By collecting historical data and using the maximum information coefficient technology to construct a convolutional neural network model for predicting the output current of photovoltaic strings, the dust accumulation situation can be evaluated at a time granularity of minutes, thereby achieving fine-grained monitoring of the performance of photovoltaic power stations. The minute-level data of each hour and each string, solar radiation intensity, ambient temperature, wind speed, and humidity are constructed into a matrix, and analyzed through a deep learning model, making the assessment of the dust accumulation situation more accurate and real-time.

[0170] When establishing the model, the present invention comprehensively considers the influence of meteorological factors such as solar radiation intensity, ambient temperature, wind speed, humidity, etc. on the string output current, so that the model is more adaptable and can better adapt to the operation of photovoltaic power stations under different meteorological conditions, and proposes classification standards for dust accumulation status, including slight dust accumulation, moderate dust accumulation and severe dust accumulation, and recommends optimizing and adjusting the classification standards through a real-time feedback system and real-time feedback from on-site personnel. The classification standards and feedback system proposed by the present invention improve the accuracy of the dust accumulation status and the intelligence of the system, make up for the shortcomings of traditional methods, and improve the accuracy and real-time performance of the dust accumulation status, thereby providing an innovative solution for the operation and maintenance management of photovoltaic power stations, which can be monitored and predicted in real time. Through the processing of real-time data, the dust accumulation problem of photovoltaic modules can be discovered in a timely manner, which helps to take timely cleaning measures and minimize power generation losses.

[0171] Those skilled in the art will appreciate that the units, i.e., algorithm steps, of the various examples described in conjunction with this embodiment can be implemented in electronic hardware or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0172] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for evaluating dust accumulation in photovoltaic power plants with a particle size down to the hourly level of the string level, characterized in that: The photovoltaic power station dust accumulation assessment method with granularity down to the string level and hour level includes the following steps: S1. Obtain the actual operation information and construction drawings of the photovoltaic power station to clarify the relationship between the strings and inverters and build a string-level topology structure; S2. Collect historical operation data of the photovoltaic power station to perform labeling operations, select characteristic variables to determine the string current, and set the basis for classifying the dust accumulation degree according to the determination results; S3. Build a training group cascade hourly dust accumulation assessment model based on convolutional neural network algorithm, characteristic variables and state classification; S4. Input the real-time collected data into the dust accumulation assessment model to predict the string current, and compare the predicted value with the actual value to evaluate the dust accumulation status; The step of obtaining the actual operation information of the photovoltaic power station and the construction drawings to clarify the affiliation between the strings and the inverters and constructing the string-level topology structure includes the following steps: S11. Build a power station database according to the construction drawings, and obtain actual on-site operation information through the power station database, wherein the operation information includes the type of photovoltaic power station components, the number of strings, the string connection method, the inverter type, the number of inverters, and the inverter connection method; S12. Implement graph data processing according to the operation information, design a dynamic topology structure including strings and inverters, and adjust the dynamic topology structure using a depth-first traversal method combined with an optimization algorithm; S13. Use the dynamic topology structure to clarify the relationship between the string and the inverter, and analyze the actual access status of the string branch; The collecting of historical operation data of the photovoltaic power station to implement labeling operation, selecting characteristic variables to judge the string current, and setting the dust accumulation degree state classification basis according to the judgment result include the following steps: S21, obtaining the historical operation data of the photovoltaic power station to implement abnormal value processing operations, and performing label adding operations on the historical data of the strings in the clean time period; S22. Use data management theory to establish a data conversion model, and implement explicit processing on the implicit knowledge of multi-label data, and use feature selection algorithm to extract input feature variables, wherein the input feature variables include solar radiation intensity, ambient temperature, wind speed and humidity; S23, using the maximum information coefficient and mutual information to calculate the correlation between the output current of each string and the input characteristic variable; S24, comparing the string current judgment result with the current of a standard battery panel in a clean state to set a dust accumulation degree state classification, and setting corresponding operation measures; The method of building a training group cascade hourly dust accumulation assessment model based on a convolutional neural network algorithm, characteristic variables and state classification includes the following steps: S31, constructing a matrix according to the input characteristic variables of each string group at each hour as an input matrix, and using the output current of each string group at each hour as a label to form a data set; Specifically, constructing a matrix as an input matrix according to the input feature variables of each string per hour, and using the output current of each string per hour as a label to form a data set includes the following steps: Set the minute-level data of each string every hour, solar radiation intensity, ambient temperature, wind speed, and humidity to form a 4*60 matrix as the input matrix; Set the output current of each string per hour as a label, and merge the input matrix with the label to generate a data set; S32, randomly shuffling the order of the data set, standardizing it and dividing it into a training set and a test set, and setting the characteristic variables as model inputs, and the state classification basis and the string current as model outputs; S33. Use the convolutional neural network structure to build a network model, and use the mean square error as the loss function to train the network model to complete the construction of the group-level hourly dust accumulation assessment model.

2. A photovoltaic power station dust accumulation assessment method with particle size down to the hourly level of the string level according to claim 1, characterized in that: The method of implementing graph data processing according to the operation information, designing a dynamic topology structure including strings and inverters, and adjusting the dynamic topology structure by using a depth-first traversal method combined with an optimization algorithm includes the following steps: S121, using the power station database to convert the operation information, classify the operation information into two types: string and inverter, and design an intelligent recognition layer inside the power station database to establish a dynamic topology structure; S122, using an optimization algorithm to set the strings as group points, each group of string points as a single node, and the access relationship between the strings and the inverter, and optimizing the determination speed of the dynamic topology structure according to the setting results; S123, using a depth-first traversal method to regard the beginning of the group string point as the initial point, and traversing and marking each group string point one by one starting from the initial point; S124. Determine whether the group string point is visited multiple times or not visited during the traversal process according to the marking result. If yes, return to step S121 to re-establish the dynamic topology structure. If no, complete the establishment of the dynamic topology structure.

3. A photovoltaic power station dust accumulation assessment method with particle size down to the hourly level of the string level according to claim 1, characterized in that: The method of using data management theory to establish a data conversion model, making the implicit knowledge of multi-label data explicit, and extracting input feature variables using a feature selection algorithm includes the following steps: S221. Using data management theory combined with multi-label data to generate four groups of independent individuals as the model basis for dynamic knowledge conversion, the four groups of independent individuals include label communication individuals, data clarification individuals, data explicit individuals and data conversion individuals; S222, using the label communication individuals to perform data consistency detection on multiple data labels, and the data clarification individuals to integrate the multi-label data to promote label clarification; S223, using data explicit individuals to obtain explicit knowledge of multiple data labels to enhance implicit knowledge, and using data conversion individuals to complete data conversion; S224. Randomly select samples from the multi-label data, find neighboring samples of the samples, calculate the feature weights of multiple sample points, and set a threshold value according to the feature weight sorting result to extract the input feature variable.

4. A method for evaluating dust accumulation in photovoltaic power plants with a particle size down to the hourly level of the string level according to claim 3, characterized in that: The method of calculating the correlation degree between each string output current and the input characteristic variable by using the maximum information coefficient and the mutual information comprises the following steps: S231, using mutual information to evaluate the mutual dependence between the input feature vector and the string output current, and determining the mutual information value according to the evaluation result; S232. Measure the strength of the nonlinear relationship between the output currents of each string according to the mutual information value combined with the maximum information coefficient, and calculate the characteristic variable weight coefficient of the degree of relationship between the input characteristic variable and the output current by using a grid division method.

5. A photovoltaic power station dust accumulation assessment method with particle size down to the hourly level of the string level according to claim 4, characterized in that: The calculation formula for the relationship between the input characteristic variable and the output current is: In the formula, MIC(X, Y) represents the relationship between the input characteristic variable and the output current; X represents the input feature variable; Y represents the output current of each string; G represents the grid partition collection; g represents the number of grid divisions; I(X g ; Y g ) represents the mutual information between X and Y; X g With Y g Represent the discretized versions of X and Y on G respectively.

6. A method for evaluating dust accumulation in photovoltaic power plants with a particle size down to the hourly level of the string level according to claim 1, characterized in that: The dust accumulation degree states include a first level dust accumulation degree, a second level dust accumulation degree and a third level dust accumulation degree; The first level dust accumulation degree is when the difference between the string output current and the current setting value of the standard solar panel is between 1 and 1.2 times, indicating the presence of dust accumulation; The second level dust accumulation degree is when the difference between the string output current and the current setting value of the standard solar panel is between 1.2 and 1.5 times, indicating that the dust accumulation is more obvious; The third level of dust accumulation is when the difference between the string output current and the current setting value of the standard solar panel is more than 1.5 times, indicating that the dust accumulation is serious.

7. A method for evaluating dust accumulation in photovoltaic power plants with a particle size down to the hourly level of the string level according to claim 1, characterized in that: The method of using a convolutional neural network structure to build a network model and using the mean square error as a loss function to train the network model to complete the construction of a group string hourly dust accumulation assessment model includes the following steps: S331, using the input layer in the convolutional neural network structure to receive the input matrix, and using the convolutional layer, the activation layer and the pooling layer to output the feature map and connect it with the fully connected layer; S332, using a fully connected layer to flatten the feature map into a one-dimensional vector and combine it with the maximum information coefficient, and using the output layer to set the predicted value of the neuron output string current to build a network model; S333. Take the mean square error as the loss function, use the Adam algorithm to train the network model, adjust the network parameters, and complete the construction of the group string-level hourly dust accumulation assessment model.

Images