CN117713688B

CN117713688B - Low-efficiency identification and electric quantity lifting method of photovoltaic module under multi-orientation and inclination angles

Info

Publication number: CN117713688B
Application number: CN202410168758.XA
Authority: CN
Inventors: 孙翰墨; 陈正安; 李梓维; 单泽宇; 韩金阳; 郑雄; 包洁; 王垚
Original assignee: Solway Online Beijing New Energy Technology Co ltd
Current assignee: Solway Online Beijing New Energy Technology Co ltd
Priority date: 2024-02-06
Filing date: 2024-02-06
Publication date: 2024-04-09
Anticipated expiration: 2044-02-06
Also published as: CN117713688A

Abstract

The invention provides a low-efficiency identification and electric quantity lifting method of a photovoltaic module under multi-orientation and inclination angles. Belongs to the technical field of photovoltaics; the method comprises the following steps: the method comprises the steps of collecting historical operation data of a photovoltaic power station through data collection equipment, preprocessing the collected historical operation data, storing the preprocessed data into a cloud space for further processing, and constructing an MLP-Mixer model by adopting a TensorFlow deep learning framework; integrating the output result of the MLP-Mixer model with priori knowledge of a photovoltaic power station to form a second data set, and establishing an XGBoost model by adopting an XGBoost framework; and acquiring real-time operation data of the photovoltaic power station in real time through data acquisition equipment, and transmitting the real-time operation data into an MLP-Mixer model. The low-efficiency group strings are identified through the MLP-Mixer model, the maintenance advice is generated through the XGBoost model, and the power generation efficiency is improved and the cost is saved through the mutual cooperation of the two models.

Description

Low-efficiency identification and electric quantity lifting method of photovoltaic module under multi-orientation and inclination angles

Technical Field

The invention provides a low-efficiency identification and electric quantity lifting method of a photovoltaic module under multi-orientation and inclination angles, and belongs to the technical field of photovoltaic.

Background

In photovoltaic power plant operation, components of different orientations and inclinations may cause the current of a string to be significantly lower than other components, thereby affecting the power generation efficiency of the entire string. Although some measures may be taken during the plant design phase to minimize the problem of series inefficiency, this problem may still occur frequently due to specific reasons such as construction units or field conditions. Photovoltaic power plants typically occupy a wide area and are numerous in components, making it difficult for operation and maintenance personnel to efficiently manually verify and screen each string, which results in a significant increase in labor costs. Therefore, it becomes necessary to develop an artificial intelligence method that automatically identifies component inefficiency.

At present, a method for solving the problem of low efficiency of photovoltaic power station string can only be adopted for offline inspection, and the method has the following defects: (1) The labor cost is high, because the photovoltaic power station occupies a large area, the components are numerous, the attribution problem of the components and the strings is difficult to see at a glance, and a great deal of time and effort are required to check whether the components of each string are in the same direction and inclination angle manually. (2) The system is low in efficiency, easy to make mistakes, low in manual inspection efficiency, easy to make mistakes, and many low-efficiency problems can not be found, so that the power generation efficiency of the power station is affected. (3) Lacking real-time, traditional manual inspection methods are generally periodic, rather than real-time. This results in a large delay between the discovery of the problem and taking corrective action, reducing the ability to quickly respond to the potential problem.

Disclosure of Invention

The invention provides a low-efficiency identification and electric quantity lifting method of a photovoltaic module under multi-orientation and inclination angles, which is used for solving the problems of high cost, low efficiency and poor instantaneity caused by manual inspection in the prior art; high hardware cost and difficult maintenance caused by adding the sensor, and the like:

the invention provides a low-efficiency identification and electric quantity lifting method of a photovoltaic module under multi-orientation and inclination angles, which comprises the following steps:

the method comprises the steps of collecting historical operation data of a photovoltaic power station through data collection equipment, preprocessing the collected historical operation data, storing the preprocessed data into a cloud space for further processing, and constructing an MLP-Mixer model by adopting a TensorFlow deep learning framework;

integrating the output result of the MLP-Mixer model with priori knowledge of a photovoltaic power station to form a second data set, and establishing an XGBoost model by adopting an XGBoost framework;

real-time operation data of the photovoltaic power station are acquired in real time through data acquisition equipment, the real-time operation data are transmitted into an MLP-Mixer model, and an output result and priori knowledge of the MLP-Mixer are input into an XGBoost model to obtain a prediction result;

generating a maintenance suggestion according to the prediction result, and pushing the prediction result and the maintenance suggestion to related personnel through a lifting system; and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting the MLP-Mixer and the XGBoost model according to the real-time data.

Further, the data acquisition device acquires historical operation data of the photovoltaic power station, performs preprocessing on the acquired historical operation data, stores the preprocessed data into a cloud space for further processing, and constructs an MLP-Mixer model by adopting a TensorFlow deep learning framework, wherein the method comprises the following steps:

the method comprises the steps that historical operation data of a photovoltaic power station are collected through data collection equipment, the historical operation data are transmitted to edge equipment, the edge equipment is used for preprocessing the historical operation data, and the preprocessed historical operation data are transmitted to a cloud space;

the cloud space receives the historical operation data, stores the data into different storage spaces in a classified mode according to data types, divides the data in the storage spaces into a plurality of data blocks, and processes the data blocks through a parallel processing algorithm to obtain processed first data;

combining the first data through a combination algorithm to form a first data set; dividing the first data set into a first training set and a first testing set;

and constructing an MLP-Mixer model by using a TensorFlow deep learning framework, training the MLP-Mixer model by using a first training set, and verifying and optimizing the model by using a first verification set.

Further, the integrating the output result of the MLP-Mixer model with the priori knowledge of the photovoltaic power station to form a second data set, and building the XGBoost model by adopting the XGBoost framework includes:

processing the photovoltaic power station data by using an MLP-Mixer model, and obtaining an output result of the model; integrating the output result of the MLP-Mixer model with priori knowledge of the photovoltaic power station to form a second data set;

converting the second dataset into a feature matrix, wherein each row represents a sample and each column represents a feature; and preparing a target variable;

dividing the second data set into a second training set and a second validation set;

an XGBoost framework is adopted, an XGBoost model is established, the XGBoost model is trained through a second training set, and model parameters are adjusted according to training results;

and evaluating the trained and optimized XGBoost model through the second verification set, and judging the evaluation result of the XGBoost model according to the evaluation result.

Further, the step of acquiring real-time operation data of the photovoltaic power station in real time through the data acquisition device, transmitting the real-time operation parameter data to the MLP-Mixer model, and inputting an output result of the MLP-Mixer and priori knowledge into the XGBoost model to obtain a prediction result, includes:

Real-time operation data of the photovoltaic power station are collected in real time through data collection equipment, and the real-time operation data are preprocessed through edge equipment;

taking the preprocessed real-time data as input, transmitting the input data into an MLP-Mixer model, and predicting through the MLP-Mixer model; obtaining an output result of the MLP-Mixer model;

integrating the output result of the MLP-Mixer model with priori knowledge of the photovoltaic power station to form a third data set; transmitting the third data set into the XGBoost model for prediction; and obtaining a prediction result of the XGBoost model.

Further, generating a maintenance suggestion according to the prediction result, and pushing the maintenance result and the maintenance suggestion to related personnel through a lifting system; and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting MLP-Mixer and XGBoost models according to the real-time data, wherein the method comprises the following steps:

generating maintenance suggestions according to the prediction results of the MLP-Mixer and the XGBoost model;

pushing the maintenance result and the maintenance suggestion to related personnel through a lifting system;

monitoring operation parameters of the photovoltaic power station in real time, and adjusting MLP-Mixer and XGBoost models according to real-time data;

the invention provides a system for realizing a low-efficiency identification and electric quantity lifting method of a photovoltaic module under multi-orientation and inclination angles, which comprises the following components:

And a data acquisition module: the method comprises the steps of collecting historical operation data of a photovoltaic power station through data collection equipment, preprocessing the collected historical operation data, storing the preprocessed data into a cloud space for further processing, and constructing an MLP-Mixer model by adopting a TensorFlow deep learning framework;

model construction module: integrating the output result of the MLP-Mixer model with priori knowledge of a photovoltaic power station to form a second data set, and establishing an XGBoost model by adopting an XGBoost framework;

and a data input module: real-time operation data of the photovoltaic power station are acquired in real time through data acquisition equipment, the real-time operation data are transmitted into an MLP-Mixer model, and an output result and priori knowledge of the MLP-Mixer are input into an XGBoost model to obtain a prediction result;

a maintenance suggestion module: generating a maintenance suggestion according to the prediction result, and pushing the prediction result and the maintenance suggestion to related personnel through a lifting system; and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting the MLP-Mixer and the XGBoost model according to the real-time data.

Further, the data acquisition module includes:

and a data preprocessing module: the method comprises the steps that historical operation data of a photovoltaic power station are collected through data collection equipment, the historical operation data are transmitted to edge equipment, the edge equipment is used for preprocessing the historical operation data, and the preprocessed historical operation data are transmitted to a cloud space;

And a data post-dividing module: the cloud space receives the historical operation data, stores the data into different storage spaces in a classified mode according to data types, divides the data in the storage spaces into a plurality of data blocks, and processes the data blocks through a parallel processing algorithm to obtain processed first data;

and a data merging module: combining the first data through a combination algorithm to form a first data set; dividing the first data set into a first training set and a first testing set;

and (5) verifying and optimizing a module: and constructing an MLP-Mixer model by using a TensorFlow deep learning framework, training the MLP-Mixer model by using a first training set, and verifying and optimizing the model by using a first verification set.

Further, the model building module includes:

and a result output module: processing the photovoltaic power station data by using an MLP-Mixer model, and obtaining an output result of the model; integrating the output result of the MLP-Mixer model with priori knowledge of the photovoltaic power station to form a second data set;

and a data conversion module: converting the second dataset into a feature matrix, wherein each row represents a sample and each column represents a feature; and preparing a target variable;

A data set dividing module: dividing the second data set into a second training set and a second validation set;

parameter adjustment module: an XGBoost framework is adopted, an XGBoost model is established, the XGBoost model is trained through a second training set, and model parameters are adjusted according to training results;

model evaluation module: and evaluating the trained and optimized XGBoost model through the second verification set, and judging the evaluation result of the XGBoost model according to the evaluation result.

Further, the data input module includes:

and the real-time acquisition module is used for: real-time operation data of the photovoltaic power station are collected in real time through data collection equipment, and the real-time operation data are preprocessed through edge equipment;

and the real-time input module is used for: taking the preprocessed real-time data as input, transmitting the input data into an MLP-Mixer model, and predicting through the MLP-Mixer model; obtaining an output result of the MLP-Mixer model;

and a data integration module: integrating the output result of the MLP-Mixer model with priori knowledge of the photovoltaic power station to form a third data set; transmitting the third data set into the XGBoost model for prediction; and obtaining a prediction result of the XGBoost model.

Further, the maintenance suggestion module includes:

The suggestion generation module: generating maintenance suggestions according to the prediction results of the MLP-Mixer and the XGBoost model;

and a result pushing module: pushing the maintenance result and the maintenance suggestion to related personnel through a lifting system;

and the real-time adjustment module is used for: and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting the MLP-Mixer and the XGBoost model according to the real-time data.

The invention has the beneficial effects that: according to the invention, the low-efficiency string is identified through the MLP-Mixer model, the maintenance suggestion is generated through the XGBoost model, and the power generation efficiency is improved and the cost is saved through the mutual cooperation of the two models. According to the invention, the MLP-Mixer model based on historical operation data is adopted, and a sensor is not required to be additionally installed, so that the hardware cost is reduced. Meanwhile, the construction and maintenance cost of the photovoltaic power station is greatly reduced, and the economy of the project is improved. The MLP-Mixer model has the advantages of global receptive field and time sequence modeling, and improves the accurate identification capability of the inefficient components. The series connection inefficiency problem existing in the photovoltaic power station can be found in time, and the loss of potential power generation capacity is reduced. Through the MLP-Mixer model, real-time monitoring of historical operation data is achieved, and intelligent power generation capacity lifting suggestions are provided based on association rules. The real-time performance of problem identification is improved, the overall performance of the photovoltaic power station is timely optimized, and the power generation efficiency is improved. The intervention requirement of manpower on the low-efficiency group strings is reduced through intelligent generating capacity lifting suggestion. The operation and maintenance cost is reduced, the automation degree of the photovoltaic power station is improved, and the operation and maintenance are more economical and efficient.

Drawings

FIG. 1 is a diagram of the steps of the method of the present invention;

FIG. 2 is a block diagram of a system according to the present invention;

FIG. 3 is a diagram of the MLP-Mixer model according to the present invention;

fig. 4 is a flow chart of the method of the present invention.

Detailed Description

In order that the above-recited objects, features and advantages of the present invention will be more clearly understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. It should be noted that, in the case of no conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, and the described embodiments are merely some, rather than all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

In one embodiment of the present invention, as shown in fig. 1 and 4, a method for inefficient identification and power boost of a photovoltaic module under multi-orientation and tilt angles, the method includes:

s1: the method comprises the steps of collecting historical operation data of a photovoltaic power station through data collection equipment, preprocessing the collected historical operation data, storing the preprocessed data into a cloud space for further processing, and obtaining a processed first data set; adopting a TensorFlow deep learning framework to construct an MLP-Mixer model; dividing the first data set into a first training set and a first verification set;

s2: integrating the output result of the MLP-Mixer model with priori knowledge of a photovoltaic power station to form a second data set, and establishing an XGBoost model by adopting an XGBoost framework; and dividing the second data set into a second training set and a second validation set;

s3: real-time operation data of the photovoltaic power station are acquired in real time through data acquisition equipment, the real-time operation parameter data are transmitted to an MLP-Mixer model, and an output result and priori knowledge of the MLP-Mixer are input into an XGBoost model to obtain a prediction result; the prediction result includes the number of inefficient components, the cost of rectifying, and the time of returning.

S4: generating a maintenance suggestion according to the prediction result, and pushing the prediction result and the maintenance suggestion to related personnel through a lifting system; and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting the MLP-Mixer and the XGBoost model according to the real-time data.

The working principle of the technical scheme is as follows: historical operation data of the photovoltaic power station are collected through data acquisition equipment, and the data are preprocessed. Preprocessing may include operations such as data cleansing, outlier handling, and missing value padding to ensure the quality and integrity of the data. The preprocessed data is stored in cloud space for further processing; the MLP-Mixer model was constructed using the TensorFlow deep learning framework. The MLP-Mixer model is a mixed model based on a multi-layer perceptron, combines a plurality of full-connection layers with the mixed layer, and can extract characteristic information in input data; the historical operating data is partitioned, with a portion being used as a training set and a portion being used as a validation set. The training set is used for training the MLP-Mixer model, and the verification set is used for evaluating the performance and generalization capability of the model; and integrating the output result of the MLP-Mixer model with priori knowledge of the photovoltaic power station to form a second data set. For example, information about the power rating, temperature range, etc. of the component can be taken as a priori knowledge and combined with the output of the MLP-Mixer. An XGBoost model is built using an XGBoost framework. XGBoost is an integrated learning algorithm, and is based on an integrated method of a decision tree, and can be used for solving nonlinear problems and improving generalization capability of a model. The second data set is divided into a training set and a validation set for training and evaluating the performance of the XGBoost model. And monitoring the operation parameters of the photovoltaic power station in real time, and transmitting real-time data into an MLP-Mixer model for prediction. The MLP-Mixer model transmits the output result and priori knowledge into the XGBoost model together to obtain a final prediction result. And generating maintenance suggestions according to the prediction results, and pushing the maintenance results and the suggestions to related personnel through a lifting system. The maintenance recommendations may include information such as the number of inefficient components, costs of modification, and time to return to the home to assist relevant personnel in developing a corresponding maintenance schedule. Meanwhile, according to the data acquired in real time, the MLP-Mixer and the XGBoost model are dynamically adjusted and optimized to adapt to the actual running condition of the photovoltaic power station.

The technical scheme has the effects that: according to the technical scheme, historical operation data of the photovoltaic power station are collected, and are preprocessed and analyzed to obtain information of the low-efficiency component. And then, transmitting the information to an MLP-Mixer model for processing and integration, combining priori knowledge to form a second data set, and establishing an XGBoost model by using an XGBoost framework for prediction and identification. In actual operation, the low-efficiency components are timely modified or replaced according to the prediction result, so that the power generation efficiency of the photovoltaic power station is improved; according to the technical scheme, the maintenance plan can be formulated in a targeted manner by predicting the number of the low-efficiency components and the correction cost, so that unnecessary maintenance investment is reduced, and the maintenance cost is reduced. Meanwhile, the technical scheme realizes intelligent maintenance, avoids the complicated and dangerous manual inspection, and improves the maintenance efficiency and safety; according to the technical scheme, through monitoring the operation parameters of the photovoltaic power station in real time, problems are found, prediction and maintenance are carried out, the reliability of the photovoltaic power station can be improved, and the downtime and loss are reduced. The real-time monitoring provides more accurate and timely data for operators, so that the operators can respond to problems rapidly, and corresponding measures are taken to ensure the normal operation of the photovoltaic power station; the cost-return time information in the predicted result of the technical scheme can provide decision support for related personnel, and help the related personnel to reasonably arrange maintenance work and investment plan. Meanwhile, the technical scheme can generate maintenance suggestions and push the maintenance suggestions to related personnel through a lifting system, so that the maintenance suggestions can be helped to quickly respond to problems and take measures; according to the technical scheme, the MLP-Mixer and the XGBoost model are dynamically adjusted and optimized according to the data acquired in real time, so that the method is suitable for the actual running condition of a photovoltaic power station, and the prediction accuracy and the generalization capability of the model are improved. By constantly learning and optimizing, the model can better predict the number of inefficient components and the cost of rectifying and modifying, and more accurate maintenance suggestions are provided for operators.

According to one embodiment of the invention, the data acquisition device acquires historical operation data of the photovoltaic power station, performs preprocessing on the acquired historical operation data, stores the preprocessed data into a cloud space for further processing, and constructs an MLP-Mixer model by adopting a TensorFlow deep learning framework, wherein the method comprises the following steps of:

the method comprises the steps that historical operation data of a photovoltaic power station are collected through data collection equipment, the historical operation data are transmitted to edge equipment, the edge equipment is used for preprocessing the historical operation data, and the preprocessed historical operation data are transmitted to a cloud space; the historical operating data comprises string current, inverter input power, inverter output power, solar radiation intensity and ambient temperature;

combining the first data through a combination algorithm to form a first data set; dividing the first data set into a first training set and a first testing set; the first data set is a 36 x 1440 data matrix (minute data).

And constructing an MLP-Mixer model by using a TensorFlow deep learning framework, training the MLP-Mixer model by using a first training set, and verifying and optimizing the model by using a first verification set. The MLP-Mixer model outputs 32 types of data, and judges whether the group strings 1-32 are inefficient. Considering that the input data is a matrix, modeling is performed by adopting a convolution kernel MLP_Miser combined method, the type and description of each layer of the model are shown in table 1, and the structure diagram of the MLP-Mixer model is shown in FIG. 3:

the core structure of the MLP-Mixer model is composed of a plurality of MLP blocks, and each MLP block comprises two full connection layers and one residual connection. The following are key MLP-Mixer model formulas and descriptions:

1) Input data (time series of 36 x 1440)

The input data is represented as a matrix X, which has a shape (36, 1440).

2) Convolutional layer

The convolution operation is carried out on the input data, the dimension is transformed, and the specific operation is as follows:

Conv1D(X) = Conv1D (X, filters = 64, kernel_size = 1,padding = ‘same')

the output shape is (36, 64).

3) MLP block

First full connection layer:

FC1 (X) = GELU (Dense (X, units = 64))

first Dropout layer:

Dropout1 (FC1 (X) ) = Dropout (FC1 (X) ,rate = 0.1)

the second full connection layer:

FC2 (Dropoutl (FC1 (X) ) ) = Dense (Dropout1 (FC1 (X) ), units = 64)

second Dropout layer:

Dropout2 (FC2 (Dropout1 (FC1 (X) ) ) ) - Dropout (FC2 (Dropout1 (FC1 (X) ) ), rate = 0.1)

residual connection:

Residual (X) = X + Dropout2 (FC2 (Dropout1 (FC1 (X) ) ) )

4) Global averaging pooling layer

Global average pooling of the output of the last MLP block:

The output shape is (64)

5) Full connection layer

And performing fully-connected operation on the output of the global average pooling layer:

Dense (GlobalAveragePooling (Residual (X) ), units = 32 , activation = ' softmax' )

the output shape is (32)

In the above formula, GELU represents a GELU activation function, dense represents a fully connected layer, and Dropout represents a dropped layer. The whole model is to extract the features of the sequence data by stacking multiple MLP blocks and reduce the dimension of the time sequence to a fixed size by a global averaging pooling layer. Finally, the fully connected layer maps the extracted features onto 32 categories.

The working principle of the technical scheme is as follows: the historical operation data of the photovoltaic power station is collected by using the data collection equipment, and the historical operation data comprise parameters such as string current, inverter input power, inverter output power, solar radiation intensity, ambient temperature and the like. Transmitting the collected historical operation data to edge equipment, and performing preprocessing operations on the edge equipment, such as data cleaning, abnormal value removal, data normalization and the like, so as to ensure the quality and consistency of the data; and transmitting the preprocessed historical operation data to a cloud space for further processing. In the cloud space, data is classified and stored to different storage spaces according to data types, for example, group string current data is stored in one storage space, and inverter input power data is stored in another storage space. Then, dividing the data in the storage space into a plurality of data blocks, and processing the data blocks through a parallel processing algorithm to obtain processed first data; and combining the processed first data by utilizing a combining algorithm to form a first data set. The data set is a 36 x 1440 data matrix, where 36 represents 36 historical parameters and 1440 represents the number of data points per day. Then, dividing the first data set into a first training set and a first testing set for training and verifying the model; an MLP-Mixer model is constructed using a TensorFlow deep learning framework, which can process sequence data and has deep learning capabilities. And training the MLP-Mixer model by using a first training set, and continuously adjusting weights and parameters through a back propagation algorithm to minimize prediction errors and improve the performance and generalization capability of the model. And meanwhile, the first verification set is used for verifying and optimizing the model, so that the stability and reliability of the model are ensured.

The technical scheme has the effects that: historical operation data of the photovoltaic power station is collected and preprocessed through the data collection equipment, and accuracy and usability of the data are guaranteed. The pretreatment process can comprise the steps of data cleaning, denoising, missing value filling and the like, so that the quality and the integrity of data are improved; and storing the historical operation data in a cloud space, classifying and storing according to the data types, and processing the data blocks by using a parallel processing algorithm. Therefore, the efficiency and the speed of data processing can be improved, and the calculation time and the resource consumption are reduced; the processed data set is divided into a training set and a testing set for constructing and training an MLP-Mixer model. Training the model through a training set, and verifying and optimizing the model by using a verification set, so that the accuracy and generalization capability of the model are improved; and constructing an MLP-Mixer model by using a TensorFlow deep learning framework, wherein the model has strong nonlinear modeling capability and adaptability. Through model training and optimization, characteristics and rules in historical operation data can be captured better, and performance prediction and optimization control of the photovoltaic power station are achieved.

According to one embodiment of the invention, the integrating the output result of the MLP-Mixer model with the priori knowledge of the photovoltaic power station to form a second data set, and building the XGBoost model by adopting the XGBoost framework comprises the following steps:

converting the second dataset into a feature matrix, wherein each row represents a sample and each column represents a feature; and preparing a target variable; the target variable is the photovoltaic power station related index to be predicted;

dividing the second data set into a second training set and a second validation set; 70% of data is used as a training set, and 30% of data is used as a verification set; and ensures that the number and order of samples of the training set and the validation set are consistent with the second data set.

An XGBoost framework is adopted, an XGBoost model is established, the XGBoost model is trained through a second training set, and model parameters are adjusted according to training results; the model parameters comprise learning rate, tree depth and regularization parameters;

The XGBoost model is established as follows:

based on XGBoost building model, predicting the group strings with the low-efficiency components, identifying the number of the low-efficiency components, predicting the cost of correction and the time of returning the correction, the specific formulas are as follows:

(1) Inefficient component count model:

(2) modifying the cost model:

(3) returning to the time model:

total model:

in the method, in the process of the invention,is a weight for adjusting the contribution of each sub-model.

Model parameter adjustment

The method based on Bayesian optimization adjusts the parameters of the model, and comprises the following specific steps:

(1) defining a hyper-parameter space:

let the super parameters of each model beThe superparameter space is->Wherein->The value range of (2) is +.>。

(2) Establishing an initial model:

randomly selecting a set of superparameters in a superparameter spaceEstablishing an initial model by using training data;

(3) evaluation of performance:

evaluating initial model performance using a cross-validation method;

(4) and (3) establishing a proxy model:

using the previous super-parameters and performance data, establishing a proxy model based on a Gaussian process method to estimate the mapping relation between the super-parameters and the performance;

(5) selecting the following set of super parameters:

selecting a next set of most likely performance enhancing hyper-parameters in the hyper-parameter space using the proxy model;

(6) and (3) establishing a model:

using new combinations of super parametersBuilding a new model by using training data;

(7) evaluation of performance:

evaluating the performance of the model using a cross-validation method;

(8) updating the proxy model:

New hyper-parameters and performance dataAdding the mapping relation into the proxy model, and updating the mapping relation;

(9) repeating:

repeating steps (5) to (8) until a predetermined number of times or convergence conditions are reached.

A final model:

selecting the super-parameter combination with the best performance on the verification set, and reestablishing a model by using the whole training set;

the working principle of the technical scheme is as follows: and processing the photovoltaic power station data by using an MLP-Mixer model, and obtaining an output result of the model. The MLP-Mixer model takes historical operation data of a photovoltaic power station as input, extracts characteristic information in the data through a series of multi-layer perceptron and mixed layer operation, and generates an output result of the model. And integrating the output result of the MLP-Mixer model with priori knowledge of the photovoltaic power station to form a second data set. The a priori knowledge here may include information on the structure of the photovoltaic power plant, plant parameters, meteorological data, etc. By integrating priori knowledge, the model output result can be better utilized to be combined with the actual situation, and the interpretability and accuracy of the data are enhanced; the second dataset is converted into a feature matrix, wherein each row represents one sample, each column represents one feature, and the target variable is prepared. And processing the integrated second data set, and converting the second data set into a feature matrix form which can be processed by a machine learning algorithm. Meanwhile, determining a photovoltaic power station related index to be predicted as a target variable; the second data set is divided into a second training set and a second validation set, in a ratio of 70% to 30%. Ensuring that the number and sequence of samples of the training set and the verification set are consistent with those of the second data set so as to ensure the reliability of the model in the training and verification process; and establishing an XGBoost model by adopting an XGBoost framework. And training the XGBoost model by using the second training set, and adjusting model parameters according to the training result. XGBoost is a machine learning algorithm based on a gradient lifting tree (Gradient Boosting Tree) that progressively improves the performance of the model by iteratively training multiple weak classifiers. Adjusting model parameters includes learning rate, tree depth, regularization parameters, and the like. The learning rate controls the contribution degree of each iteration to the model, the complexity of the model is determined by the depth of the tree, and the regularization parameters are used for controlling the complexity of the model and preventing overfitting; and evaluating the trained and optimized XGBoost model through the second verification set, and judging the prediction result of the XGBoost model according to the evaluation result. And comparing the accuracy and generalization capability of the model with the true value of the verification set, and judging whether the model achieves the expected effect. It is assumed that the evaluation is performed using a second validation set comprising a set of sample data of known true values. These sample data may be input into a trained and optimized XGBoost model and then compared to the true values. For example, the XGBoost model may be used to predict the generated power of a photovoltaic power plant. The photovoltaic power plant data of the second verification set may be used as input, including characteristic information such as weather conditions, component temperatures, irradiance, etc., and the corresponding actual generated power as a true value. For each sample, the model will give a predicted value representing the predicted generated power of the photovoltaic power plant under the given conditions. We can compare these predictions to the true values and use some evaluation criteria to measure the performance of the model, such as Root Mean Square Error (RMSE), mean Absolute Error (MAE), etc. If the evaluation result shows that the prediction error of the XGBoost model is smaller, namely the values of the RMSE and the MAE are lower, the model can be considered to have better generalization capability, and the power generation power of the photovoltaic power station can be accurately predicted. Conversely, if the evaluation result shows that the prediction error of the XGBoost model is large, it may be necessary to further optimize the model, adjust model parameters or retrain the model to improve its accuracy and performance.

The technical scheme has the effects that: the output result of the MLP-Mixer model is integrated with priori knowledge of the photovoltaic power station, so that the factors such as the structure, the equipment parameters and the meteorological data of the photovoltaic power station can be more comprehensively considered, and the interpretability and the accuracy of the data are improved; the MLP-Mixer model can effectively extract characteristic information in photovoltaic power station data through multi-layer perceptron and mixed layer operation. The characteristics can be used as the input of the XGBoost model, so that the model is helped to better understand and predict the related indexes of the photovoltaic power station; XGBoost is a machine learning algorithm based on gradient lifting trees, and has strong generalization capability and accuracy. Through training and optimizing the XGBoost model, the method can be better adapted to the characteristics of the photovoltaic power station data, and the accuracy of prediction is improved; the second data set is divided into a training set and a verification set, the training set can be used for training and parameter tuning of the model, and the verification set can be used for evaluating and selecting the model. Thus, the problem of overfitting of the model in the training process can be avoided, and the generalization capability of the model is improved; through the prediction and evaluation of the trained and optimized XGBoost model on the verification set, whether the model achieves the expected effect can be judged. If the predicted result of the model accords with the true value, the model has higher accuracy and reliability.

In one embodiment of the present invention, the method for acquiring real-time operation data of a photovoltaic power station in real time by a data acquisition device, transmitting the real-time operation data into an MLP-Mixer model, and inputting an output result and a priori knowledge of the MLP-Mixer into an XGBoost model to obtain a prediction result, includes:

real-time operation data of the photovoltaic power station are collected in real time through data collection equipment, and the real-time operation data are preprocessed through edge equipment; the preprocessing comprises missing value processing, abnormal value detection, denoising and the like. And according to the prior knowledge, carrying out normalization or standardization treatment on the real-time data so as to keep consistent with the input data of the MLP-Mixer model;

taking the preprocessed real-time data as input, transmitting the input data into an MLP-Mixer model, and predicting through the MLP-Mixer model; obtaining an output result of the MLP-Mixer model; and the output result is used for predicting the performance index of the photovoltaic power station.

The working principle of the technical scheme is as follows: and acquiring real-time operation data of the photovoltaic power station in real time by using data acquisition equipment, and preprocessing the data by using edge equipment. The preprocessing comprises missing value processing, abnormal value detection, denoising and the like, so that the quality and the accuracy of data are ensured. Meanwhile, according to priori knowledge, carrying out normalization or standardization treatment on the real-time data to enable the real-time data to be consistent with the input data of the MLP-Mixer model; and taking the preprocessed real-time data as input, and transmitting the input data into an MLP-Mixer model for prediction. The MLP-Mixer model is a model based on a multi-layer perceptron and mixed layer operation, and can effectively extract characteristic information in photovoltaic power station data. The preliminary prediction result of the performance index of the photovoltaic power station can be obtained through the prediction of the MLP-Mixer model; and integrating the output result of the MLP-Mixer model with priori knowledge of the photovoltaic power station to form a third data set. The prior knowledge comprises related information such as the structure, equipment parameters, meteorological data and the like of the photovoltaic power station. By integrating priori knowledge, the characteristics and environmental factors of the photovoltaic power station can be considered more comprehensively, and the accuracy and reliability of prediction are improved; and transmitting the integrated third data set into the XGBoost model for further prediction. XGBoost is a machine learning algorithm based on gradient lifting trees, and can adapt to the characteristics of photovoltaic power station data through training and tuning, so that the prediction accuracy is improved. Through the prediction of the XGBoost model, a more accurate prediction result of the performance index of the photovoltaic power station can be obtained.

The technical scheme has the effects that: the latest data information can be timely obtained by collecting real-time operation data of the photovoltaic power station in real time and preprocessing the data by combining with the edge equipment, and the real-time performance index prediction is realized, so that the problems can be timely found and solved; the MLP-Mixer model and the XGBoost model are both machine learning models which are trained and optimized, and accurate prediction can be performed according to real-time data and priori knowledge of the photovoltaic power station. By integrating priori knowledge, the characteristics and environmental factors of the photovoltaic power station can be considered more comprehensively, and the accuracy and reliability of prediction are improved; according to the technical scheme, by integrating real-time data, priori knowledge and output results of a machine learning model, a plurality of factors such as the running state, equipment parameters, meteorological data and the like of the photovoltaic power station are comprehensively considered, so that the performance index of the photovoltaic power station is predicted more comprehensively; the accurate performance index prediction can help the photovoltaic power generation industry to monitor and optimize operation. Through real-time prediction, abnormal conditions can be found in time and corresponding measures can be taken, so that the efficiency and reliability of the photovoltaic power station are improved, and the operation and maintenance cost is reduced.

According to one embodiment of the invention, the maintenance advice is generated according to the prediction result, and the prediction result and the maintenance advice are pushed to related personnel through a lifting system; and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting MLP-Mixer and XGBoost models according to the real-time data, wherein the method comprises the following steps:

pushing the maintenance result and the maintenance suggestion to related personnel through a lifting system; the maintenance result and the advice can be sent to maintenance personnel or management personnel by means of mail, short message or APP, etc. so as to take corresponding measures in time.

the adjustment of the MLP-Mixer model comprises the adjustment of parameters such as the layer number, the neuron number, the activation function and the like of the model so as to improve the expression capacity and the fitting capacity of the model.

And adjusting parameters such as depth, learning rate and the like of the decision tree to improve generalization capability and prediction accuracy of the model.

Training and verifying again on the third data set according to the adjusted model; and repeating the steps to continuously optimize and perfect the model.

The working principle of the technical scheme is as follows: the operation parameters of the photovoltaic power station are monitored in real time, real-time data are input into the trained MLP-Mixer and XGBoost model, performance indexes of the photovoltaic power station are predicted, and corresponding maintenance suggestions are generated. For example, if the predicted results indicate that some components are inefficient, it may be recommended to replace the failed component or perform cleaning operations; and pushing the maintenance result and the maintenance suggestion to related personnel through the lifting system. The maintenance result and the advice can be sent to maintenance personnel or management personnel in a mail, short message or APP mode, so that corresponding measures can be taken in time; and continuously monitoring the operation parameters of the photovoltaic power station in real time, and adjusting the MLP-Mixer and the XGBoost model according to the real-time data. Adjusting parameters and super parameters of the model, such as the number of layers, the number of neurons, an activation function, the depth of a decision tree, the learning rate and the like, so as to improve the expression capacity, generalization capacity and prediction accuracy of the model; and (3) training and verifying again on the third data set according to the adjusted model to verify the performance and effect of the model, and continuously optimizing and perfecting the model.

The technical scheme has the effects that: by predicting the performance index of the photovoltaic power station and generating maintenance suggestions, maintenance personnel can be helped to know the running state of the power station and the existing problems in time, corresponding maintenance measures are adopted in advance, and the occurrence of faults and the maintenance cost are reduced; the maintenance result and the maintenance suggestion are pushed to related personnel through the lifting system, so that timely information transmission and communication can be realized, the maintenance response speed is improved, the influence of faults on the power station productivity is reduced, and the stability and reliability of the photovoltaic power station are improved; the accuracy and the adaptability of the model can be kept by monitoring the operation parameters of the photovoltaic power station in real time and adjusting the MLP-Mixer and the XGBoost model according to the real-time data so as to adapt to the change of the operation state of the photovoltaic power station and improve the accuracy and the reliability of the prediction model; the model after adjustment can be continuously optimized and improved by retraining and verifying the model on the third data set, and the generalization capability and the prediction accuracy of the model can be improved, so that the model can be better adapted to the running conditions of different photovoltaic power stations.

In one embodiment of the present invention, as shown in fig. 2, a system for implementing a method for inefficient identification and power boost of a photovoltaic module under multi-orientation and tilt angles, the system includes:

And a data acquisition module: the method comprises the steps of collecting historical operation data of a photovoltaic power station through data collection equipment, preprocessing the collected historical operation data, storing the preprocessed data into a cloud space for further processing, and obtaining a processed first data set; adopting a TensorFlow deep learning framework to construct an MLP-Mixer model; dividing the first data set into a first training set and a first verification set;

model construction module: integrating the output result of the MLP-Mixer model with priori knowledge of a photovoltaic power station to form a second data set, and establishing an XGBoost model by adopting an XGBoost framework; and dividing the second data set into a second training set and a second validation set;

and a data input module: real-time operation data of the photovoltaic power station are acquired in real time through data acquisition equipment, the real-time operation data are transmitted into an MLP-Mixer model, and an output result and priori knowledge of the MLP-Mixer are input into an XGBoost model to obtain a prediction result; the prediction result includes the number of inefficient components, the cost of rectifying, and the time of returning.

In one embodiment of the present invention, the data acquisition module includes:

and a data preprocessing module: the method comprises the steps that historical operation data of a photovoltaic power station are collected through data collection equipment, the historical operation data are transmitted to edge equipment, the edge equipment is used for preprocessing the historical operation data, and the preprocessed historical operation data are transmitted to a cloud space; the historical operating data comprises string current, inverter input power, inverter output power, solar radiation intensity and ambient temperature;

and a data merging module: combining the first data through a combination algorithm to form a first data set; dividing the first data set into a first training set and a first testing set; the first data set is a 36 x 1440 data matrix.

In one embodiment of the present invention, the model building module includes:

and a data conversion module: converting the second dataset into a feature matrix, wherein each row represents a sample and each column represents a feature; and preparing a target variable; the target variable is the photovoltaic power station related index to be predicted;

a data set dividing module: dividing the second data set into a second training set and a second validation set; 70% of data is used as a training set, and 30% of data is used as a verification set; and ensures that the number and order of samples of the training set and the validation set are consistent with the second data set.

Parameter adjustment module: an XGBoost framework is adopted, an XGBoost model is established, the XGBoost model is trained through a second training set, and model parameters are adjusted according to training results; the model parameters comprise learning rate, tree depth and regularization parameters;

In one embodiment of the present invention, the data input module includes:

and the real-time acquisition module is used for: real-time operation data of the photovoltaic power station are collected in real time through data collection equipment, and the real-time operation data are preprocessed through edge equipment; the preprocessing comprises missing value processing, abnormal value detection, denoising and the like. And according to the prior knowledge, carrying out normalization or standardization treatment on the real-time data so as to keep consistent with the input data of the MLP-Mixer model;

and the real-time input module is used for: taking the preprocessed real-time data as input, transmitting the input data into an MLP-Mixer model, and predicting through the MLP-Mixer model; obtaining an output result of the MLP-Mixer model; and the output result is used for predicting the performance index of the photovoltaic power station.

In one embodiment of the present invention, the maintenance recommendation module includes:

and a result pushing module: pushing the maintenance result and the maintenance suggestion to related personnel through a lifting system; the maintenance result and the advice can be sent to maintenance personnel or management personnel by means of mail, short message or APP, etc. so as to take corresponding measures in time.

It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.

Claims

1. The method for low-efficiency identification and electric quantity lifting of the photovoltaic module under multi-orientation and inclination angles is characterized by comprising the following steps:

generating maintenance suggestions according to the prediction results, and pushing the prediction results and the maintenance suggestions to related personnel through a lifting system; the operation parameters of the photovoltaic power station are monitored in real time, and the MLP-Mixer and the XGBoost model are adjusted according to the real-time data;

the method comprises the steps of collecting historical operation data of a photovoltaic power station through data collection equipment, preprocessing the collected historical operation data, storing the preprocessed data into a cloud space for further processing, and constructing an MLP-Mixer model by adopting a TensorFlow deep learning framework, wherein the method comprises the following steps of:

2. The method for low-efficiency recognition and electric quantity lifting of a photovoltaic module under multi-orientation and inclination angle according to claim 1, wherein integrating the output result of the MLP-Mixer model with the priori knowledge of the photovoltaic power station to form a second data set, and establishing an XGBoost model by adopting an XGBoost framework comprises the following steps:

3. The method for low-efficiency recognition and electric quantity lifting of a photovoltaic module under multi-orientation and inclination angle according to claim 1, wherein the steps of collecting real-time operation data of a photovoltaic power station in real time through a data collection device, transmitting the real-time operation data into an MLP-Mixer model, inputting an output result and priori knowledge of the MLP-Mixer into an XGBoost model, and obtaining a prediction result comprise the following steps:

4. The method for low-efficiency recognition and electric quantity lifting of the photovoltaic module under multi-orientation and inclination angle according to claim 1, wherein the maintenance advice is generated according to the prediction result and is pushed to related personnel through a lifting system; and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting MLP-Mixer and XGBoost models according to the real-time data, wherein the method comprises the following steps:

and monitoring the operation parameters of the photovoltaic power station in real time, and adjusting the MLP-Mixer and the XGBoost model according to the real-time data.

5. A system for implementing a method for inefficient identification and power boost of photovoltaic modules at multi-orientations and inclinations, the system comprising:

a maintenance suggestion module: generating maintenance suggestions according to the prediction results, and pushing the prediction results and the maintenance suggestions to related personnel through a lifting system; the operation parameters of the photovoltaic power station are monitored in real time, and the MLP-Mixer and the XGBoost model are adjusted according to the real-time data;

the data acquisition module comprises:

6. The system for implementing the method for inefficient identification and power boost of photovoltaic modules at multi-orientations and inclinations of claim 5, wherein the model building module comprises:

7. The system for implementing the method for inefficient identification and power boost of photovoltaic modules at multi-orientations and inclinations of claim 5, wherein the data input module comprises:

8. The system for implementing the method for inefficient identification and power boost of photovoltaic modules at multi-orientations and inclinations of claim 5, wherein the maintenance recommendation module comprises: