CN116307287B

CN116307287B - Prediction method, system and prediction terminal for effective period of photovoltaic power generation

Info

Publication number: CN116307287B
Application number: CN202310566222.9A
Authority: CN
Inventors: 刘柱; 崔明涛; 董石磊; 李温静; 苏建平; 高丽媛; 周超; 李春阳; 赵红磊; 刘玉民; 张沛尧; 冯坤; 马红月; 张玲璐; 明萌
Original assignee: Sichuan Zhongdian Aostar Information Technologies Co ltd; State Grid Information and Telecommunication Co Ltd
Current assignee: Sichuan Zhongdian Aostar Information Technologies Co ltd; State Grid Information and Telecommunication Co Ltd
Priority date: 2023-05-19
Filing date: 2023-05-19
Publication date: 2023-08-01
Anticipated expiration: 2043-05-19
Also published as: CN116307287A

Abstract

The invention provides a prediction method, a prediction system and a prediction terminal for a photovoltaic power generation effective period, which relate to the technical field of distributed new energy, and are used for collecting and acquiring data information of a distributed photovoltaic power station in a prediction area and carrying out normalization processing; establishing an undirected graph with weights to consider spatial correlation among the distributed photovoltaic power stations; determining a neighbor power station set of each power station by adopting a graph convolution neural network; adopting a back propagation neural network to process missing values and abnormal values in the system; based on the data information of the distributed photovoltaic power station in the system, performing four-rule operation to obtain the operated characteristics of the data information; and establishing a prediction model by utilizing an integrated learning strategy, and executing a prediction process on the power generation effective period based on the prediction model. The method can help energy management personnel to better predict the available time period of new energy power generation, thereby optimizing energy supply and use plans and improving energy utilization efficiency.

Description

Prediction method, system and prediction terminal for effective period of photovoltaic power generation

Technical Field

The invention relates to the technical field of distributed new energy, in particular to a method and a system for predicting the effective period of photovoltaic power generation and a prediction terminal.

Background

At present, the distributed new energy power generation technology is widely developed, and the distributed new energy power generation can be used for arranging small-scale distributed power generation on a user side or a power grid side, and the complementation and the optimal utilization of electric energy are realized through the connection with the power grid. Compared with the traditional centralized power generation mode, the distributed new energy power generation technology has the following advantages: the energy consumption can be reduced, the power supply reliability can be improved, the energy cost can be reduced, the economic development and the environmental protection can be promoted, and the like.

The centralized photovoltaic system generally refers to a photovoltaic power station with a larger scale, and the generated energy data of the photovoltaic power station is collected and managed by a centralized monitoring system. Compared with a distributed photovoltaic system, the monitoring and maintenance of the centralized photovoltaic system are relatively easier, and are more stable and reliable. As a result, the problem of data loss in the centralized photovoltaic system is relatively small, but there are still some cases where data loss may occur, such as equipment failure, network problems, human error, and the like.

However, due to special properties of the distributed new energy power generation equipment, such as uneven equipment distribution, unstable power generation and the like, the power generation has uncertainty and fluctuation, and the future power generation capacity is difficult to accurately predict. This presents a large uncertainty and risk to grid operation and power market transactions, as well as challenges to the utilization and planning of distributed new energy power generation equipment.

Due to the large number of distributed photovoltaic systems, monitoring and maintenance work becomes more difficult. Therefore, in the distributed photovoltaic system, a situation of data missing often occurs, that is, the generated energy data of some photovoltaic power stations are not collected or transmitted in time. These data loss may be due to various reasons, such as equipment failure, network problems, transmission failure, etc. For practical distributed photovoltaics, especially in rural areas, small weather monitoring devices are not generally installed, and no small-scale real-time weather information exists. In this case, only local and wide-range fuzzy weather information can be obtained, and the scale is large. The data loss problem can affect aspects such as performance evaluation, operation management, power market transaction and the like of the distributed photovoltaic system. And if the generated energy data of some photovoltaic power stations are lost, the calculation of the total generated energy of the system is inaccurate, so that the power generation capacity assessment, the power generation income calculation and the like are influenced. The data loss also affects the fault diagnosis and maintenance of the system, and reduces the reliability and safety of the photovoltaic system.

For missing data sets, if special data enhancement is not performed, but a simple outlier missing processing method, such as average filling, is adopted, large errors can occur in the processed data. Especially when there is a large area of data missing, this simple processing method may not accurately fill the missing values, thereby affecting the subsequent data analysis and model modeling results. When processing missing data, the photovoltaic power stations are classified by simply relying on Euclidean distance and meshing, the mutual relation between the photovoltaic power stations and the origin is found by relying on the distance relation between the photovoltaic power stations and the origin, the mutual relation is too complicated, and after conversion, the implicit relation between weather data and historical power data between the photovoltaic power stations is difficult to find due to the fact that the resolution of NWP is too low.

Disclosure of Invention

The invention provides a prediction method for the effective period of photovoltaic power generation, which can accurately predict the power generation capacity of distributed new energy and provide powerful support for the fields of power market, power system scheduling and the like.

The method comprises the following steps: step 1: collecting and acquiring data information of a distributed photovoltaic power station in a preset area, and carrying out normalization treatment;

step 2, establishing an undirected graph with weights to consider the spatial correlation between the distributed photovoltaic power stations;

step 3, determining a neighbor power station set of each power station by adopting a graph convolution neural network;

step 4, adopting a back propagation neural network to process missing values and abnormal values in the system;

step 5, based on the data information of the distributed photovoltaic power station in the system, performing four arithmetic operations to obtain the operated characteristics of the data information, and simultaneously, performing two-by-two intersection of the environmental characteristics to expand new characteristics;

and 6, establishing a prediction model by utilizing an integrated learning strategy, and executing a prediction process for the effective period of power generation based on the prediction model.

It should be further noted that, the data information of the distributed photovoltaic power station in step 1 includes: position information, historical power data, illuminance information, power generation voltage information, power generation current information, battery SOC information, and climate information.

In step 2, the correlation of the historical power between the distributed photovoltaic power stations is used as the weight of the edge, and the weight of the node is determined according to the area and the missing degree of the data set.

It should be further noted that, step 3 further includes: constructing a local subgraph for each distributed photovoltaic power station and neighbor power stations thereof;

and splicing the characteristic vector of each power station with the characteristic vector of the neighboring node of each power station to serve as the node characteristic of the local subgraph.

It should be further noted that, performing convolution operation on the local subgraph to extract spatial features;

and taking the node characteristics obtained by convolution operation as new characteristics of the nodes, learning to obtain spatial characteristics of the position, the area and the perimeter of the power station, taking the spatial characteristics of the position, the area and the perimeter of the power station as the input of a back propagation neural network, and training to obtain a model filled with abnormal values and missing values.

It should be further noted that, step 5 further includes: using the obtained spatial characteristics as input, and training the historical characteristic data of a part of known distributed photovoltaic power stations and a centralized photovoltaic power station with a relatively close distance as output to obtain a model for processing missing values and abnormal values; and filling and processing the missing data of the distributed photovoltaic power station by using a data filling model prediction model.

It should be further noted that step 6 further includes: using a first base model XGBoost as input, using the generated power of the distributed photovoltaic power station as output, training to generate output characteristics, and adding the output characteristics into the original characteristics;

after the output characteristics are expanded, the output characteristics are combined and input into a first LightGBM algorithm, and the generated power of the distributed photovoltaic power station is taken as output to train and predict the generated power of the photovoltaic power station;

then the second base model LightGBM algorithm is used for expanding, the expanded characteristics are used as input, the photovoltaic power generation power is used as output, new characteristics are generated through training and added into the original characteristics, then the new characteristics and the expanded characteristics are combined to be used as input through the second base model XGBoost, the photovoltaic power generation power is used as output, and the predicted photovoltaic power generation power is trained and predicted;

based on the LSTM neural network, the expanded characteristics are used as input, the generated power is used as output, the attention mechanism is introduced after the LSTM neural network, and the characteristic relation is learned in the time sequence;

and finally, adding the results according to the weight ratio of 3:3:4 to obtain a final prediction model.

It should be further noted that, after step 6, the method further includes:

step 7, error calculation is carried out on the photovoltaic power generation power and the actual photovoltaic power generation power so as to evaluate the prediction performance of the prediction model, and whether the algorithm model needs to be carried out again or not is determined to be trained according to the error;

and if the error is greater than a preset error threshold, adjusting the input data information, and training the prediction model again based on the first base model XGBoost, the second base model XGBoost, the first LightGBM algorithm, the second base model LightGBM algorithm and the LSTM neural network in the step 6 to enable the prediction error of the prediction model to be within the preset error threshold.

The invention also provides a prediction system of the photovoltaic power generation effective period, which comprises: the system comprises a data collection module, a data configuration module, a data determination module, a reverse processing module, a characteristic processing module and a prediction model establishment module;

the data collection module is used for collecting and acquiring data information of the distributed photovoltaic power station in the preset area and carrying out normalization processing;

the data configuration module is used for establishing an undirected graph with weights to consider the spatial correlation among the distributed photovoltaic power stations;

the data determining module is used for determining a neighbor power station set of each power station by adopting a graph convolution neural network;

the reverse processing module is used for processing missing values and abnormal values in the system by adopting a back propagation neural network;

the feature processing module performs four arithmetic operations based on the data information of the distributed photovoltaic power station in the system to obtain the operated features of the data information, and simultaneously performs the environment feature pairwise intersection to expand new features;

the prediction model building module is used for building a prediction model by utilizing an integrated learning strategy, and executing a prediction process on the power generation effective period based on the prediction model.

The invention also provides a prediction terminal which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to realize the steps of the prediction method of the photovoltaic power generation effective period.

From the above technical scheme, the invention has the following advantages:

according to the prediction method for the photovoltaic power generation effective period, the historical data of the distributed new energy power generation is collected and analyzed, a deep learning model is constructed to process the photovoltaic missing value and the abnormal value by using technologies such as data mining and machine learning, and then a prediction model is constructed to evaluate the function prediction effective period. The predictive model may take into account a number of factors, such as weather, time, geographic location, etc., to improve the accuracy and reliability of the predictions. The method can help energy management personnel to better predict the available time period of new energy power generation, thereby optimizing energy supply and use plans and improving energy utilization efficiency.

The invention effectively solves the problems that the data loss can also affect the fault diagnosis and maintenance of the system and reduce the reliability and safety of the photovoltaic system. The defects that the photovoltaic power stations are classified by simply relying on Euclidean distance and meshing, and the interrelationship between the photovoltaic power stations is found by relying on the distance relation between the photovoltaic power stations and the origin point, which is too complicated, are overcome.

Drawings

In order to more clearly illustrate the technical solutions of the present invention, the drawings that are needed in the description will be briefly introduced below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a method for predicting the effective period of photovoltaic power generation;

fig. 2 is a schematic diagram of a prediction system for the effective period of photovoltaic power generation.

Detailed Description

The prediction method of the photovoltaic power generation effective period mainly solves the problem of data missing in a distributed photovoltaic system, and can effectively realize data completion so as to improve the performance and reliability of the system.

The invention can acquire and process the associated data based on the artificial intelligence technology. The prediction method utilizes the machine simulation of a digital computer, extends and expands the intelligence of people, senses the environment, acquires knowledge and uses the knowledge to acquire the theory, the method, the technology and the application device of the optimal result.

The prediction method has a hardware level technology and a software level technology. The basic technology of the intelligent diagnosis method of the numerical control machine tool generally comprises technologies such as a sensor, a special artificial intelligent chip, cloud computing, distributed storage, big data processing technology, an operation/interaction system, electromechanical integration and the like. The intelligent diagnosis method software technology of the numerical control machine mainly comprises a computer visual angle technology, machine learning/deep learning and computer program codes. The computer program code includes, but is not limited to, object oriented programming languages such as Java, smalltalk, C ++, and conventional procedural programming languages, such as the "C" language or similar programming languages.

Of course, the prediction method of the present invention also has a machine learning function, wherein the machine learning and the deep learning in the method of the present invention generally comprise artificial neural network, confidence network, reinforcement learning, migration learning, induction learning, teaching learning and other technologies.

The prediction method of the photovoltaic power generation effective period is applied to one or more prediction terminals, wherein the prediction terminals are equipment capable of automatically carrying out numerical calculation and/or information processing according to preset or stored instructions, and hardware of the prediction terminals comprises, but is not limited to, a microprocessor, an Application-specific integrated circuit (SpecificIntegratedCircuit, ASIC), a programmable gate array (Field-ProgrammableGate Array, FPGA), a digital processor (DigitalSignalProcessor, DSP), an embedded equipment and the like.

The predictive terminal may be any electronic product that can interact with a user, such as a personal computer, tablet, smart phone, personal digital assistant (PersonalDigitalAssistant, PDA), interactive web television (InternetProtocolTelevision, IPTV), smart wearable device, etc.

The network in which the terminal is predicted to be located includes, but is not limited to, the internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VirtualPrivateNetwork, VPN), and the like.

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. 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.

Referring to fig. 1, the method of the present invention includes:

s1: collecting and acquiring data information of a distributed photovoltaic power station in a preset area, and carrying out normalization treatment;

in particular, the data information that the present invention may collect includes: the position information, the historical power data, the illuminance information, the power generation voltage information, the power generation current information, the battery SOC information, and the climate information are not limited to the above information, and may be collected and acquired based on actual needs.

The collected position can be historical data stored in a database, or can be currently acquired operation data of the distributed photovoltaic power station. Wherein the power station, the photovoltaic power station and the distributed photovoltaic power station group referred to in the invention all represent distributed photovoltaic power stations.

The invention performs normalization processing on the acquired data information, wherein the normalization processing is performed on historical power data, weather data and the like by using a statistical method, so that generalization and robustness of the model are enhanced, and the training convergence speed is improved.

The climate information or the weather information of the invention refers to a method for predicting the atmospheric motion state and the weather phenomenon in a certain period of time in the future according to the actual condition of the atmosphere by carrying out numerical calculation through a large-scale computer under certain initial value and side value conditions and solving a hydrodynamic and thermodynamic equation set describing the weather evolution process.

The present invention also relates to meteorological factors, here various factors that have an impact on the meteorological process or event, including temperature, humidity, barometric pressure, wind speed, precipitation, etc.

S2: establishing undirected graphs with weights to account for spatial correlation between distributed photovoltaic power plants

The weighted undirected graph in this embodiment is a graph composed of points (vertices) and lines (edges) connecting the points, where the edges have no direction, that is, the edges connecting the two points can start from any one point. On the basis of the weighted undirected graph, each edge is given a weight value for representing the importance or cost of the edge.

For the distributed photovoltaic power stations, an undirected graph with weights can be established to consider the spatial correlation among the power stations, the correlation of historical power among the photovoltaic power stations is utilized as the weight of the side, and the weight of the node is determined according to the area and the missing degree of the data set.

S3: determining a neighbor power station set of each power station by adopting a graph convolution neural network;

by employing a graph convolution neural network, a set of neighbor power stations for each power station is determined. The invention relates to a graph convolution neural network: the deep learning model is characterized in that nodes with similar structures or adjacent relations are closer in a feature space through learning the feature representation of the nodes, so that the nodes can be effectively classified and predicted.

A local subgraph is built for each power station and its neighbors. And splicing the characteristic vector of each power station with the characteristic vector of the neighboring node to serve as the node characteristic of the local subgraph. And carrying out convolution operation on the local subgraph to extract spatial features. The node characteristics obtained by convolution operation are used as new characteristics of the node, so that spatial characteristics such as the position, the area, the perimeter and the like of a power station can be learned, and the characteristics are used as the input of a back propagation neural network so as to better extract the characteristics.

S4, adopting a back propagation neural network to process missing values and abnormal values in the system;

the missing value and the abnormal value are processed by adopting a back propagation neural network. Using the obtained spatial characteristics as input, historical characteristic data of part of distributed photovoltaic power stations and centralized photovoltaic power stations which are close to each other, such as power, temperature, humidity, solar irradiance and the like, are known as output, and training is performed to obtain a model for processing missing values and abnormal values. And then, filling and processing missing data of all the distributed photovoltaic power stations by using the model.

Illustratively, the back propagation neural network is based on a multi-layer neural network learning algorithm, and the back propagation neural network is to transmit a loss function in forward propagation into a back propagation process, and calculate partial derivatives of the loss function on weights and biases of various neurons layer by layer as gradients of the objective function on the weights and biases. Modifying the weights w and the bias b according to this calculated gradient, the learning of the network being done during the weight modification. When the error reaches the expected value, the network learning is ended.

S5, based on the data information of the distributed photovoltaic power station in the system, performing four arithmetic operations to obtain the operated characteristics of the data information, and simultaneously, performing two-by-two intersection of the environmental characteristics to expand new characteristics;

in the feature engineering related by the invention, the existing features are subjected to four-rule operation, such as temperature, humidity, solar irradiance and the like, according to the actual physical meaning, so that new features, such as temperature difference, dew point and the like, can be obtained. Meanwhile, the environmental features can be crossed two by two to develop new features, for example, the temperature and humidity environmental features can be crossed to obtain new features such as relative humidity. For the two environmental characteristics of solar irradiance and temperature, new characteristics of illumination intensity and the like can be obtained by crossing.

And S6, establishing a prediction model by utilizing an integrated learning strategy, and executing a prediction process for the effective period of power generation based on the prediction model.

Specifically, the characteristics after expansion are used as input, the photovoltaic power generation power is used as output, new characteristics are trained and generated, the new characteristics are added into the original characteristics, then the new characteristics and the characteristics after expansion are combined to be used as input through the LightGBM, the photovoltaic power generation power is used as output, and the photovoltaic power generation power is trained and predicted.

The second base model is the LightGBM, the expanded characteristics are used as input, the photovoltaic power generation power is used as output, new characteristics are generated through training, the new characteristics are added into the original characteristics, then the new characteristics and the expanded characteristics are combined to be used as input through the second base model XGBoost, the photovoltaic power generation power is used as output, and the predicted photovoltaic power generation power is trained and predicted.

and finally, adding the results according to the weight ratio of 3:3:4 to obtain a final prediction result.

In embodiments of the invention, the RNN round robin neural network involved is a neural network for processing sequence data. Compared to a general neural network, he can process data of a sequence variation.

For LSTM neural networks, it is desirable to solve the problems of gradient extinction and gradient explosion during long sequence training. In short, LSTM is able to perform better in longer sequences than normal RNNs.

The base model XGBoost is an integrated learning algorithm based on a decision tree and is used for solving the regression and classification problems.

The base model LightGBM is an efficient open source gradient lifting framework, can be used for classification, regression and sequencing tasks, and is a machine learning model based on a decision tree algorithm.

Attention mechanism (transducer): the idea of simulating human attention, which is commonly used in deep learning to process sequence data, can effectively strengthen and optimize the model. Specifically, the attention mechanism calculates the contribution degree of each input to the current output according to the current input and the previous history information, and then adjusts the weight of each input, so that the input with high contribution degree is more important in calculating the output. In this way, the model can automatically learn the relationships between the different inputs, thereby making predictions and classifications better.

The integrated learning strategy of the present invention is a technique that improves prediction accuracy by combining a plurality of different learning algorithms. Ensemble learning may be used for various machine learning tasks such as classification, regression, clustering, and the like. The basic idea is to reduce model errors by combining the predictions of multiple models, thereby improving overall performance.

The invention may also relate to feature crossings, euclidean distance, the Catboost model, etc. if desired.

Among these, euclidean distance is the most common distance measure, which is the absolute distance between two points in a multidimensional space.

The main purpose of Catboost is to predict the values of certain target variables using a decision tree model. Compared with the traditional gradient-lifted decision tree, catboost introduces two important characteristics in training the model: adaptive feature scaling and processing of class-type features. These characteristics can help Catboost better address the problems of missing values, discrete values, non-linear relationships, etc. in the data and perform well in reducing overfitting.

Feature interleaving is the creation of new features by combining multiple original features to achieve better model performance. For example, in a house price prediction model, features may include area, age, geographic location, etc. of a house, while by cross-combining these features, higher-dimensional features may be obtained, such as the product of area and geographic location, the ratio of area to age, etc. These new features may better describe the relationship between house price and other factors, thereby improving the predictive power and accuracy of the model. Feature interleaving is the creation of new features by combining multiple original features to achieve better model performance. For example, in a house price prediction model, features may include area, age, geographic location, etc. of a house, while by cross-combining these features, higher-dimensional features may be obtained, such as the product of area and geographic location, the ratio of area to age, etc. These new features may better describe the relationship between house price and other factors, thereby improving the predictive power and accuracy of the model.

And 7, in order to improve the prediction accuracy, the invention calculates the error between the predicted photovoltaic power generation power and the actual photovoltaic power generation power so as to evaluate the prediction performance of the model and determine whether to re-perform the algorithm model to train according to the error. If the error is large, then there is a need to examine the possible problems in the model, such as data quality, feature selection, model structure, etc., for further optimization and improvement. If the error is smaller, the model prediction performance is better, and the method can be continuously applied to actual production.

Based on the method, by collecting and analyzing historical data of distributed new energy power generation, a deep learning model is constructed by using technologies such as data mining and machine learning to process photovoltaic missing values and abnormal values, and then a prediction model is constructed to evaluate the functional prediction effective period. The predictive model may take into account a number of factors, such as weather, time, geographic location, etc., to improve the accuracy and reliability of the predictions. The method can help energy management personnel to better predict the available time period of new energy power generation, thereby optimizing energy supply and use plans and improving energy utilization efficiency.

The following is an embodiment of a photovoltaic power generation effective period prediction system provided by an embodiment of the present disclosure, which belongs to the same inventive concept as the photovoltaic power generation effective period prediction method of each embodiment, and details of which are not described in detail in the embodiment of the photovoltaic power generation effective period prediction system may refer to the embodiment of the photovoltaic power generation effective period prediction method.

As shown in fig. 2, the system includes: the system comprises a data collection module, a data configuration module, a data determination module, a reverse processing module, a characteristic processing module and a prediction model establishment module;

The units and algorithm steps of each example described in the embodiments disclosed in the prediction system for the photovoltaic power generation effective period provided by the invention can be implemented in electronic hardware, computer software or a combination of both, and in order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functions in the above description. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the effective period of photovoltaic power generation, the method comprising:

step 1: collecting and acquiring data information of a distributed photovoltaic power station in a preset area, and carrying out normalization treatment;

the correlation of historical power among distributed photovoltaic power stations is used as the weight of the edge, and the weight of the node is determined according to the area and the missing degree of the data set;

constructing a local subgraph for each distributed photovoltaic power station and neighbor power stations thereof;

splicing the characteristic vector of each power station with the characteristic vector of the neighbor node of each power station to serve as the node characteristic of the local subgraph;

carrying out convolution operation on the local subgraph, and extracting spatial features;

taking the node characteristics obtained by convolution operation as new characteristics of the nodes, learning to obtain spatial characteristics of the position, the area and the perimeter of the power station, taking the spatial characteristics of the position, the area and the perimeter of the power station as the input of a back propagation neural network, and training to obtain a model filled with abnormal values and missing values;

using the obtained spatial characteristics as input, and training the historical characteristic data of a part of known distributed photovoltaic power stations and a centralized photovoltaic power station with a relatively close distance as output to obtain a model for processing missing values and abnormal values; filling and processing missing data of the distributed photovoltaic power station by using a data filling model prediction model;

2. The method for predicting a photovoltaic power generation effective period according to claim 1, wherein the data information of the distributed photovoltaic power station in step 1 includes: position information, historical power data, illuminance information, power generation voltage information, power generation current information, battery SOC information, and climate information.

3. The method for predicting a photovoltaic power generation effective period according to claim 1, wherein step 6 further comprises: using a first base model XGBoost as input, using the generated power of the distributed photovoltaic power station as output, training to generate output characteristics, and adding the output characteristics into the original characteristics;

4. The method for predicting the effective period of photovoltaic power generation according to claim 1, further comprising, after step 6:

5. A photovoltaic power generation effective period prediction system, characterized in that the system adopts the photovoltaic power generation effective period prediction method according to any one of claims 1 to 4;

the system comprises: the system comprises a data collection module, a data configuration module, a data determination module, a reverse processing module, a characteristic processing module and a prediction model establishment module;

6. A prediction terminal comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor, when executing the program, implements the steps of the method for predicting the photovoltaic power generation validity period of any one of claims 1 to 4.