Distributed photovoltaic horizontal plane irradiance prediction method and system

Abstract

The invention relates to the technical field of intelligent operation and maintenance of photovoltaic power stations, and discloses a distributed photovoltaic horizontal irradiance prediction method and system, and the method comprises the steps: selecting a plurality of distributed photovoltaic power stations at different geographic positions and under different weather conditions as sample photovoltaic power stations; acquiring monitoring data, weather data, seasonal data and actually measured horizontal irradiance of all power stations in a preset area range of each sample power station, extracting corresponding features, and forming a data set based on the features and the actually measured horizontal irradiance; carrying out model training by utilizing integrated learning based on the data set to obtain a horizontal plane irradiance prediction model; and inputting monitoring data, weather data and season data of a photovoltaic power station to be predicted into the model to obtain a prediction result. The method is simple in data acquisition mode and low in cost, the problems of sequence length and lag effect in a traditional time sequence method can be avoided, the model can process features of different time points more flexibly, and prediction accuracy and model adaptability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of photovoltaic power stations, and particularly relates to a method and system for predicting horizontal irradiance of distributed photovoltaics. Background Art

[0002] The horizontal irradiance, ambient temperature, and backplane temperature measured by a pyranometer are widely used in the photovoltaic industry, mainly for calculating the system energy efficiency of a photovoltaic power station and evaluating the current power generation status of the photovoltaic power station. Due to the difficulty in obtaining pyranometer data in distributed photovoltaic scenarios, existing technical solutions are mainly obtained through training and prediction by an artificial intelligence model based on image data of satellite cloud maps and time-series data of irradiance, and are mainly applied to centralized photovoltaic scenarios. There is no technology involved in distributed photovoltaic scenarios. Currently, the technical solutions for predicting the horizontal irradiance of distributed photovoltaics are mainly divided into two categories:

[0003] I. Multimodal models based on cloud physical parameters and solar irradiance time-series data. Such models aim to more comprehensively capture the variation laws and influencing factors of solar irradiance by combining information from different data sources. In this method, cloud physical parameters (such as cloud thickness, cloud type, cloud movement speed, etc.) are used as a key input and are incorporated into the model together with the historical time-series data of solar irradiance to form multimodal inputs. This multi-source data fusion can help the model more accurately understand the influence of different meteorological conditions on irradiance, especially in cases where cloud cover and solar irradiance changes are relatively complex. In addition, the introduction of time-series data enables the model to capture time dependence and seasonal fluctuations, thereby improving the prediction accuracy.

[0004] Although this method can provide relatively accurate prediction results, it usually requires high-quality real-time meteorological data (especially satellite cloud map data). Therefore, certain challenges and cost pressures may be faced in data acquisition and processing, resulting in difficulties in implementing such solutions in some application scenarios or being unable to meet the requirements.

[0005] Second, relying on historical solar irradiance data and other meteorological data, prediction is achieved by training time-series data. Such models mainly establish prediction models by using historical solar irradiance data and related meteorological data (such as temperature, humidity, wind speed, air pressure, etc.). By inputting these multi-dimensional time-series data into the model, the model can learn the correlations between different meteorological factors and solar irradiance, and then predict the change trend of irradiance at a future moment. This method is suitable for scenarios with long-term data accumulation, where historical data can provide sufficient information to help the model capture the regular fluctuations of irradiance. During the training process, the time dependence of the time-series data is fully considered, and the model can predict future irradiance based on past irradiance values and meteorological data. These models usually use regression analysis, time-series analysis, or deep learning methods (such as LSTM, GRU, etc.) to train and predict time-series data.

[0006] Compared with multi-modal models that rely on real-time cloud physical properties parameters, the advantage of this type of model is that it has relatively low requirements for data sources and can use historical data for training. However, it also has a dependence on the quality and quantity of meteorological data. Through this method, short-term and medium-term solar irradiance can be effectively predicted, especially in cases where it is difficult to obtain real-time meteorological data. However, this complexity poses relatively high computational resource requirements for real-time inference. Especially in prediction scenarios that require high frequency and low latency, the computational burden will increase significantly, which may cause greater pressure on the enterprise's hardware and infrastructure, further limiting the popularity and application value of these solutions. Summary of the Invention

[0007] In view of this, the present invention provides a method and system for predicting the horizontal plane irradiance of distributed photovoltaics, which solves the problem of how to improve the accuracy of predicting the horizontal plane irradiance of distributed photovoltaics efficiently and at low cost.

[0008] In the first aspect, the present invention provides a method for predicting the horizontal plane irradiance of distributed photovoltaics, and the method includes:

[0009] Select multiple distributed photovoltaic power stations under different geographical locations and climate conditions as sample photovoltaic power stations;

[0010] Obtain the monitoring data, weather data, season data, and measured horizontal plane irradiance of all photovoltaic power stations within the preset area range of each sample photovoltaic power station;

[0011] Extract the characteristic data corresponding to the monitoring data, weather data, and season data of the photovoltaic power station respectively, and form a data set based on the characteristic data and the measured horizontal plane irradiance;

[0012] Based on the said data set, use ensemble learning to train the model, and obtain the trained ensemble model as the horizontal irradiance prediction model;

[0013] Input the monitoring data, weather data and season data of the photovoltaic power station to be predicted into the said horizontal irradiance prediction model, and obtain the horizontal irradiance prediction result.

[0014] In the horizontal irradiance prediction method for distributed photovoltaic provided by the embodiment of the present invention, selecting distributed photovoltaic power stations under different geographical locations and climate conditions as samples can cover rich and diverse data, making the obtained data more representative, which can reflect the operation characteristics and irradiance change rules of photovoltaic power stations in different environments, avoiding the strong dependence on meteorological data (especially real-time satellite cloud images) in traditional methods, reducing the complexity and cost of data acquisition, and at the same time reducing the risk of data loss compared with the time series data of the environmental monitor; multi-dimensional data fusion can more comprehensively describe the environment and operation state of the photovoltaic power station. Training an ensemble model based on the data set constructed from multi-source data, and ensemble learning synthesizes the results of multiple learners, reducing the bias and variance of a single model, enabling the model to better adapt to the irradiance prediction of photovoltaic power stations under different geographical and climate conditions; it is more efficient in terms of computing resource consumption, and can quickly perform real-time prediction under limited computing resources to meet the requirements of high frequency and low latency; directly modeling and fitting the features of each independent time point, abandoning the requirement of relying on complex time series, not only simplifies the structure of the model, but also can avoid problems such as sequence length and lag effect that may be encountered in traditional time series methods, enabling the model to more flexibly process the features of different time points, further improving the prediction accuracy and the adaptability of the model.

[0015] In an optional implementation manner, the selecting multiple photovoltaic power stations under different geographical locations and climate conditions as sample photovoltaic power stations includes:

[0016] Taking a preset administrative region as a unit, obtain the relevant data of distributed photovoltaic power stations within each administrative region;

[0017] Screen out the photovoltaic power stations that have environmental monitors and meet the preset stable operation conditions among each distributed photovoltaic power station as sample photovoltaic power stations.

[0018] The sample photovoltaic power stations screened by the embodiment of the present invention from each administrative region include power stations in different geographical regions and take into account different climate conditions, making the samples cover a variety of geographical and climate characteristics. When constructing a prediction model later, it has stronger representativeness and can better adapt to the horizontal irradiance prediction of photovoltaic power stations in different environments. At the same time, by screening stable operating power stations, the reliability of the sample power station data can be ensured, the model training quality can be improved, and the model can better reflect the operation rules of photovoltaic power stations under normal working conditions.

[0019] In an alternative embodiment, the monitoring data of the photovoltaic power station includes: photovoltaic power and its installed capacity; feature data obtained by performing feature extraction on the acquired monitoring data includes:

[0020] Obtain the historical photovoltaic power data of each sample photovoltaic power station and all photovoltaic power stations within its preset area at a preset time granularity;

[0021] Calculate the ratio of the historical photovoltaic power data of each photovoltaic power station to its photovoltaic installed capacity to obtain the utilization rate of the corresponding photovoltaic system;

[0022] Preprocess the utilization rates of each photovoltaic system, including removing outliers and noise from the data;

[0023] Extract multiple statistical features related to the utilization rate based on the preprocessed utilization rate data, and use at least two statistical features to obtain a statistical relationship feature characterizing the relationship between the statistical features. Use the statistical features related to the utilization rate and the statistical relationship feature as the regional power feature data.

[0024] The embodiment of the present invention extracts multiple statistical features based on the preprocessed utilization rate data, which reflects the distribution and change of the utilization rate of the photovoltaic system from different perspectives. Use at least two statistical features to obtain a statistical relationship feature characterizing the relationship between the statistical features, further exploring the potential connections between the data, and providing a good data basis for subsequent model training.

[0025] In an alternative embodiment, the monitoring data of the photovoltaic power station includes the string current and string voltage data of the inverter; feature data obtained by performing feature extraction on the acquired monitoring data includes:

[0026] Obtain the string current and string voltage of the inverter at each moment of each sample photovoltaic power station and all photovoltaic power stations within its preset area at a preset time granularity;

[0027] Calculate the ratio of the string current to the optimal operating current under preset standard conditions and the ratio of the string voltage of the inverter at each moment to the preset standard optimal operating voltage respectively;

[0028] Preprocess the two calculated ratio data, including removing outliers and noise from the data;

[0029] Extract multiple statistical features based on the preprocessed ratio data, and use the statistical features related to the ratio data as the inverter feature data.

[0030] In an embodiment of the present invention, by calculating the ratio of the string current to the optimal operating current under preset standard conditions and the ratio of the string voltage to the preset standard optimal operating voltage, the degree of proximity of the inverter to the ideal operating state during actual operation can be intuitively reflected. Based on the preprocessed ratio data, multiple statistical features are extracted, which can comprehensively describe the operating characteristics of the inverter from multiple perspectives, enabling a deeper understanding of the operating rules and characteristics of the inverter and providing a good data basis for subsequent model training.

[0031] In an alternative embodiment, the weather data includes historical weather forecast data; the feature data obtained by performing feature extraction on the acquired weather data includes:

[0032] Obtain the power station temperature data and weather type data within the preset regional range of each sample photovoltaic power station within the preset historical time period at a preset time granularity;

[0033] Encode the weather type data based on the model type in ensemble learning to obtain weather type codes;

[0034] Combine the power station temperature data and the corresponding weather type codes as weather feature data.

[0035] In an embodiment of the present invention, encoding the weather type data based on the model type in ensemble learning can convert the originally non-numerical weather type data into a numerical form that is easy for a computer to process, enabling the weather data to be uniformly analyzed and processed with other numerical data, facilitating subsequent input into various machine learning or data analysis models, improving the compatibility and generality of the models. By combining various weather factors into weather feature data, the model can learn the influence patterns of a wider range of weather conditions on the photovoltaic power station, thus having better generalization ability when facing different weather conditions and actual application scenarios.

[0036] In an alternative embodiment, the season data includes obtaining the month to which the monitoring data and weather data of the photovoltaic power station belong, and using the month as seasonal feature data.

[0037] Month is a simple and clear feature variable. In the embodiments of the present invention, it is used as seasonal feature data. Without complex calculations and conversions, it can provide intuitive time dimension information for data analysis and modeling. Compared with other seasonal indicators that may require a large amount of calculations and processing, month data is easier to obtain and understand, can reduce the difficulty and cost of data processing, and improve the analysis efficiency. When constructing data analysis models and prediction models related to photovoltaic power plants, adding this seasonal feature data of month can enable the model to better capture the seasonal trends and periodic changes in the data. The model can learn the influence patterns of different months on the monitoring data and weather data of the photovoltaic power plant, so as to perform data fitting and prediction more accurately.

[0038] In an alternative embodiment, the method of using ensemble learning to train a model based on the dataset to obtain a trained ensemble model as a horizontal irradiance prediction model includes:

[0039] Dividing the training set constructed by the regional power feature data, inverter feature data, weather feature data, seasonal feature data of the photovoltaic power plant, and the corresponding measured horizontal irradiance into multiple data subsets;

[0040] Randomly inputting the multiple data subsets into multiple learners, and training each learner separately, where the number of learners is equal to the number of data subsets;

[0041] Taking the ensemble weighted model obtained based on the trained learners as the horizontal irradiance prediction model, and the prediction result of the horizontal irradiance prediction model is the weighted average of the prediction results output by each trained learner.

[0042] In the embodiments of the present invention, dividing the training set including regional power feature data, inverter feature data, weather feature data, seasonal feature data, and the corresponding measured horizontal irradiance into multiple data subsets can fully explore different features and laws in the data. Different subsets may contain data under different times, different weather conditions, different equipment states, etc., enabling the model to learn more comprehensive information, avoiding relying on a single data feature for prediction, and improving the adaptability of the model to complex actual situations. The ensemble weighted model is the weighted average of the prediction results output by each trained learner. This method can effectively reduce the variance of the model. Different learners may show different advantages and disadvantages on different data subsets. By weighted averaging, the advantages of each learner can be integrated, smoothing out the fluctuations and errors of a single learner, making the final prediction result more stable and reliable, improving the generalization ability of the model, and reducing the uncertainty of the prediction result.

[0043] In a second aspect, the present invention provides a horizontal irradiance prediction system for distributed photovoltaics, including:

[0044] A sample photovoltaic power station selection module for selecting multiple distributed photovoltaic power stations under different geographical locations and climatic conditions as sample photovoltaic power stations;

[0045] A sample data acquisition module for acquiring the monitoring data, weather data, seasonal data, and measured horizontal irradiance of all photovoltaic power stations within the preset area of each sample photovoltaic power station;

[0046] A dataset construction module for respectively extracting the characteristic data corresponding to the monitoring data, weather data, and seasonal data of the photovoltaic power station, and forming a dataset based on the characteristic data and the measured horizontal irradiance;

[0047] A model training module for training a model using ensemble learning based on the dataset to obtain a trained ensemble model as a horizontal irradiance prediction model;

[0048] A prediction module for inputting the monitoring data, weather data, and seasonal data of the photovoltaic power station to be predicted into the horizontal irradiance prediction model to obtain a horizontal irradiance prediction result.

[0049] In a third aspect, the present invention provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the method for predicting the horizontal irradiance of distributed photovoltaics in the first aspect or any corresponding embodiment thereof.

[0050] In a fourth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the method for predicting the horizontal irradiance of distributed photovoltaics in the first aspect or any corresponding embodiment thereof.

[0051] In a fifth aspect, the present invention provides a computer program product, including computer instructions, and the computer instructions are used to cause a computer to execute the method for predicting the horizontal irradiance of distributed photovoltaics in the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0053] Figure 1 It is a flowchart showing the method for predicting the horizontal irradiance of distributed photovoltaics according to an embodiment of the present invention;

[0054] Figure 2 is a flowchart for extracting feature data based on photovoltaic power and its installed capacity according to an embodiment of the present invention;

[0055] Figure 3 is a flowchart for extracting feature data based on the series current and series voltage data of an inverter according to an embodiment of the present invention;

[0056] Figure 4 is a flowchart for extracting feature data based on historical weather forecast data according to an embodiment of the present invention;

[0057] Figure 5 is a flowchart for model training according to an embodiment of the present invention;

[0058] Figure 6 is a structural block diagram of a horizontal plane irradiance prediction system for distributed photovoltaics according to an embodiment of the present invention;

[0059] Figure 7 is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed implementation manners

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] The existing technical solutions have the following deficiencies:

[0062] For scenarios where satellite data cannot be obtained, the irradiance prediction process highly depends on physical property parameters of clouds, such as data like satellite cloud images. This makes the accuracy and real-time nature of cloud-related data crucial. However, obtaining high-precision real-time meteorological data (especially cloud image data) often poses huge economic costs and technical thresholds for enterprises, resulting in the difficulty of implementing such solutions in certain application scenarios or their inability to meet requirements. Secondly, the model designs of these solutions are generally complex, and the models used are relatively heavy. In practical applications, this complexity poses high computational resource requirements for real-time inference. Especially in prediction scenarios that require high frequency and low latency, the computational burden will increase significantly, which may impose great pressure on the enterprise's hardware and infrastructure, further limiting the popularity and application value of these solutions.

[0063] For the solutions relying on time series, there are the following technical problems:

[0064] (1) Noise and complexity of time - series data: Time - series data usually contains various noises or outliers, which may interfere with the training of the model and affect the prediction accuracy. In addition, over time, the relationship between solar irradiance and meteorological factors may change, resulting in historical data being unable to fully represent future change trends.

[0065] (2) Risk of overfitting of the model: If the training dataset overly relies on historical data, especially when training is carried out in a short period, the model may exhibit overfitting, resulting in good performance on the training set but poor prediction performance on new data or the test set. Especially when the model processes highly complex and variable meteorological data, the overfitting problem may be more prominent.

[0066] (3) Dependence on long - term data accumulation: This method performs well in scenarios of long - term data accumulation, but in cases of scarce data or insufficient data accumulation, the training effect and prediction ability of the model will be greatly reduced. If a power station or region lacks sufficient historical data, the model may not be effectively trained, thus affecting prediction accuracy.

[0067] (4) Adaptability to seasonal changes and extreme events: The training of this model relies on historical data, so its adaptability to extreme weather events or seasonal fluctuations is poor. If the model does not fully consider these changes, it may not be able to accurately predict the performance of irradiance under extreme conditions.

[0068] To overcome the deficiencies of the prior art, a method for predicting the horizontal irradiance of distributed photovoltaics is provided in this embodiment. Figure 1 It is a flowchart of the method for predicting the horizontal irradiance of distributed photovoltaics according to an embodiment of the present invention, as Figure 1 shown. The process includes the following steps:

[0069] S11, Select multiple distributed photovoltaic power stations under different geographical locations and climate conditions as sample photovoltaic power stations.

[0070] Specifically, when selecting sample photovoltaic power stations, factors such as the geographical distribution of the power stations, the surrounding environment, the similarity of meteorological conditions, the performance of photovoltaic power generation equipment, and the operation status of the power stations need to be considered to ensure that the photovoltaic power generation situations under different geographical locations and climate conditions can be reflected. In the embodiment of the present invention, taking a preset administrative region as a unit, relevant data of distributed photovoltaic power stations within each administrative region are obtained; photovoltaic power stations with environmental monitors and meeting the preset stable operation conditions are selected from each distributed photovoltaic power station as sample photovoltaic power stations.

[0071] In one embodiment, taking the provincial administrative region as a unit, through the data platform of the power grid enterprise (such as State Grid) that masters the grid-connected data of a large number of distributed photovoltaic power stations, query the relevant data of the distributed photovoltaic power stations. During the construction and acceptance phases of the photovoltaic power stations, there will be detailed technical documents and equipment lists, from which it is possible to check whether an environmental monitor is equipped, as well as information such as the model, function, and monitoring parameters of the environmental monitor. By querying the relevant data of the photovoltaic power stations, check whether the power quality indicators such as the voltage, frequency, and harmonics output by the power stations are within the allowable range, whether the output power of the power stations is stable, whether there are frequent power fluctuations or mutations, whether the main equipment such as photovoltaic modules, inverters, and transformers are operating normally, and whether there are fault alarms or abnormal heating conditions, so as to screen out the photovoltaic power stations that meet the preset stable operation conditions. For example, the voltage deviation should be within ±10%, the frequency deviation should be within ±0.5 Hz, the equipment intact rate should not be less than 95%, and the standard deviation of power fluctuations should be less than 5%.

[0072] S12. Obtain the monitoring data, weather data, season data, and measured horizontal plane irradiance of all photovoltaic power stations within the preset area of each sample photovoltaic power station.

[0073] Specifically, each photovoltaic power station is usually equipped with a monitoring system that can record data such as the power generation power, voltage, current, inverter operating status, and photovoltaic module temperature. By accessing the background of the power station's monitoring system or using relevant data interfaces, these real-time and historical monitoring data can be obtained. The power grid enterprise will have metering and monitoring equipment at the connection point of the photovoltaic power station, which can collect data such as the electricity quantity and power factor transmitted by the power station to the power grid; the weather data can be obtained from the data in the weather forecast published by professional platforms such as the meteorological data network, or the meteorological parameters of the location of the monitored power station can be obtained from the photovoltaic meteorological station set within or near the photovoltaic power station area; the season data is the month in which the data is located; the horizontal plane irradiance is measured by an environmental monitor.

[0074] The data within the preset area of the sample photovoltaic power station (such as the power stations within a radius of 20 kilometers centered on the sample photovoltaic power station) generally has similar weather conditions to the sample photovoltaic power station. In the embodiment of the present invention, the data of all photovoltaic power stations within the preset area of each sample photovoltaic power station is obtained. By analyzing these data, the overall power generation and operation conditions of the photovoltaic power stations within the preset area can be comprehensively understood, avoiding data deviation caused by the data of a single sample power station.

[0075] S13. Respectively extract the characteristic data corresponding to the monitoring data, weather data, and season data of the photovoltaic power station, and form a data set based on the characteristic data and the measured horizontal plane irradiance.

[0076] Specifically, the monitoring data of the photovoltaic power station in the embodiments of the present invention includes the photovoltaic power data of the power station and its corresponding installed capacity, the string current and string voltage data of the inverter, the weather data includes temperature data and weather type data, and the season data includes the month to which the monitoring data and weather data belong. On this basis, the regional power characteristics are extracted based on the photovoltaic power data of the power station and its corresponding installed capacity, the inverter characteristics are extracted based on the string current and string voltage data of the inverter, the weather characteristics are extracted based on the weather data, and the seasonal characteristics are extracted based on the season data.

[0077] On the basis of feature extraction, taking a unified time interval (such as every 15 minutes, every hour) as the benchmark, each feature is time-matched to obtain a feature dataset, ensuring that each data record corresponds to each other in time and accurately reflecting the status of different types of data at the same moment or time period. Then, the measured horizontal plane irradiance collected by the environmental monitor corresponding to the same moment is used as the prediction target value, and a dataset is jointly formed with the above feature data.

[0078] S14. Based on the dataset, use ensemble learning to train the model to obtain a trained ensemble model as the horizontal plane irradiance prediction model.

[0079] Specifically, from the constructed dataset, it is divided according to a certain ratio (such as 80% training set, 20% test set). Then, multiple basic learners are selected to be trained respectively using the training set, and the trained ensemble model is evaluated using the test set, and the above evaluation indicators are calculated. The prediction result of the model is compared with the measured horizontal plane irradiance in the test set, the performance of the model under different indicators is analyzed, and the trained and qualified ensemble model is deployed to the actual application environment, such as the monitoring and management system of the photovoltaic power station.

[0080] S15. Input the monitoring data, weather data, and season data of the photovoltaic power station to be predicted into the horizontal plane irradiance prediction model to obtain the horizontal plane irradiance prediction result.

[0081] Specifically, input the actual power and installed capacity of the photovoltaic power station itself and the power stations within the preset area range that need to predict the horizontal plane irradiance, the weather data in the weather forecast, and the month at the time of prediction into the horizontal plane irradiance prediction model to obtain the prediction result, and write the prediction result into the database (such as MySQL).

[0082] The horizontal irradiance prediction method for distributed photovoltaic provided by the embodiments of the present invention selects distributed photovoltaic power stations under different geographical locations and climatic conditions as samples, which can cover rich and diverse data, reflect the operating characteristics of photovoltaic power stations and the irradiance change rules under different environments, avoid the strong dependence of traditional methods on meteorological data (especially real-time satellite cloud images), reduce the complexity and cost of data acquisition, and at the same time, compared with the time series data of environmental monitors, reduce the risk of data loss; multi-dimensional data fusion can more comprehensively reflect the environment and operating status of photovoltaic power stations. The integrated model is trained based on the data set constructed from multi-source data. Ensemble learning synthesizes the results of multiple learners, reduces the bias and variance of a single model, and enables the model to better adapt to the irradiance prediction of photovoltaic power stations under different geographical and climatic conditions; it is more efficient in terms of computing resource consumption, can quickly perform real-time prediction under limited computing resources, and meet the requirements of high frequency and low latency; the direct modeling and fitting of the features of each independent time point abandons the requirement of relying on complex time series, not only simplifies the structure of the model, but also can avoid problems such as sequence length and lag effect that may be encountered in traditional time series methods, enabling the model to more flexibly process the features of different time points and further improving the prediction accuracy and the adaptability of the model.

[0083] In one embodiment, when the monitoring data of the photovoltaic power station is the photovoltaic power and its installed capacity, the process of extracting feature data based on it is as Figure 2 shown, and it includes the following steps:

[0084] S21, with a preset time granularity, obtain the historical photovoltaic power data of each sample photovoltaic power station and all photovoltaic power stations within its preset area range.

[0085] Specifically, for example, according to the preset time granularity, such as every 15 minutes, every hour, etc., extract the corresponding historical photovoltaic power data from the database. The power of each power station forms a field, and the historical sequence of the photovoltaic power of each power station can be expressed as P = {P1,..., P k}, where k is the total number of the sample photovoltaic power station itself and all photovoltaic power stations within the preset area range.

[0086] S22, calculate the ratio of the historical photovoltaic power data of each photovoltaic power station to its photovoltaic installed capacity to obtain the utilization rate of the corresponding photovoltaic system.

[0087] Specifically, the installed capacity C = {C1,..., C k}, then the utilization rate A = {A1,..., A k} = {P1 / C1,..., P k / C k} In the embodiments of the present invention, the utilization rate is defined as the ratio of the actual power generation capacity to the installed capacity, which can effectively reflect the power generation efficiency of the photovoltaic system within a specific time period. The higher the utilization rate, the better the power generation efficiency of the photovoltaic system during this time period. This indicator is usually used to evaluate the performance of the photovoltaic system and analyze its actual power generation capacity under different weather conditions, seasons, and other environmental factors. It is equivalent to normalizing the power generation capacities of all power stations within the selected range, that is, using the utilization rate as the indicator to measure the power generation capacity.

[0088] S23. Preprocess the utilization rates of each photovoltaic system, including removing outliers and noise from the data.

[0089] Specifically, for example, remove data with a utilization rate of 0 for a long time (such as one day), which indicates that there may be a malfunction in the operation of the power station. For a small number of missing values within a day, linear interpolation is used for supplementation, while for a large number of missing values, the data for that day is deleted. At the same time, delete values less than 0 and greater than 1, because the power generation of a photovoltaic power station cannot be negative and cannot exceed the installed capacity. This situation generally corresponds to negative power generation and extremely large power generation. Preprocessing the utilization rate data of the photovoltaic system to remove outliers and noise can improve the accuracy and reliability of the data.

[0090] S24. Extract multiple statistical features related to the utilization rate from the preprocessed utilization rate data, and use at least two statistical features to obtain statistical relationship features representing the relationship between the statistical features. Take the statistical features related to the utilization rate and the statistical relationship features as the regional power feature data.

[0091] Specifically, for example, calculate statistical features such as the mean, quantile, standard deviation, maximum value, and minimum value of the utilization rate; in one embodiment, rank the power station set according to the average peak sunshine hours of the photovoltaic power station within 30 days. After performing anomaly processing, select 40 power stations with the top rankings for feature extraction (if there are less than 40, all are included in the calculation). The selected statistical features include the following:

[0092] 1. Mean feature of the utilization rate at a certain moment: A avg = avg(A1,…,A k )

[0093] 2. Quantile feature of the utilization rate at a certain moment:

[0094] 1) A Q1 : The first quartile (Q1), also known as the "lower quartile", is equal to the 25% value of all the numerical values in the sample arranged from small to large;

[0095] 2) A Q2: The second quartile (Q2), also known as the "median", is equal to the 50% value after arranging all the values in the sample from small to large;

[0096] 3)A Q3 : The third quartile (Q3), also known as the "upper quartile", is equal to the 75% value after arranging all the values in the sample from small to large.

[0097] 3. Maximum value feature of the utilization rate at a certain moment: A max = max(A1,…,A k ).

[0098] 4. Standard deviation feature of the utilization rate at a certain moment: It represents the degree of dispersion of the data, indicating the volatility of the data.

[0099] 5. Variance feature of the utilization rate at a certain moment: It is the square of the standard deviation, representing the degree of dispersion of the data.

[0100] 6. Aggregating the utilization rates of power stations into relationship features: The interaction relationship between the statistical features obtained by aggregating two or more utilization rates is used as the relationship feature. The selected relationship features are, for example:

[0101] 1) Feature ratios (Ratios): such as maximum value / mean value;

[0102] 2) Feature products (Products): such as the product of the mean value and the median;

[0103] 3) Feature differences (Differences): such as the difference between the maximum value and the minimum value.

[0104] Extracting multiple statistical features based on the preprocessed utilization rate data reflects the distribution and variation of the utilization rate of the photovoltaic system from different perspectives. Using at least two statistical features to obtain statistical relationship features that characterize the relationship between statistical features further explores the potential connections between data, providing a good data basis for subsequent model training.

[0105] In one embodiment, when the monitoring data of the photovoltaic power station is the string current and string voltage data of the inverter, the process of extracting feature data based on the string current and string voltage data of the inverter is as Figure 3 shown, including the following steps:

[0106] S31, at a preset time granularity, obtain the string current and string voltage of the inverter at each moment of all photovoltaic power stations within the preset area range of each sample photovoltaic power station.

[0107] The data acquisition unit in the local monitoring system of a photovoltaic power station can be directly connected to the inverter. Through a specific communication protocol, at preset time granularities, such as every 15 minutes, every hour, etc., it periodically reads the string current and string voltage data of the inverter and stores them in the local database. Data queries and exports can be performed through the interface of the local monitoring software, or through the remote monitoring platform, according to information such as the identification of the power station and the number of the inverter, and filter and download the string current and string voltage data of the inverters of each power station at the preset time granularity.

[0108] S32. Calculate the ratio of the string current to the optimal operating current under the preset standard conditions, and the ratio of the string voltage of the inverter at each moment to the preset standard optimal operating voltage respectively.

[0109] Specifically, calculate the ratio of all string currents at a certain moment to the optimal operating current under STC (i.e., standard conditions): i a ={I1,…,I k}, and the ratio of all string voltages at a certain moment to the optimal operating voltage under STC (i.e., standard conditions): v a ={V1,…,V k}. Calculating the ratio of the string current to the optimal operating current under the preset standard conditions and the ratio of the string voltage to the preset standard optimal operating voltage can intuitively reflect the degree of proximity of the inverter to the ideal operating state during actual operation. The closer the ratio is to 1, the higher the working efficiency of the inverter. Conversely, it indicates that there may be some problems affecting the performance of the inverter, such as insufficient light, component aging, line loss, etc. In this way, it can be quickly determined whether the inverter is in a good working state.

[0110] S33. Preprocess the two calculated ratio data, including removing outliers and noise from the data.

[0111] Specifically, values of 0, less than 1, and greater than the STC optimal working current / voltage are removed. A value of 0 may be an abnormal value caused by sensor failure, data acquisition error, etc. It cannot reflect the real physical situation and will interfere with subsequent analysis. Removing such values can avoid the impact of incorrect data on the overall data accuracy and make the data more truly reflect the operating status of the photovoltaic power station; values less than 1 may be noise data generated by measurement errors, environmental interference, etc., or abnormal values that do not conform to the normal operating rules under special circumstances. Removing these values helps to eliminate unreasonable fluctuations in the data and make the data more stable and reliable; values greater than the STC optimal working current / voltage usually exceed the normal operating range of the photovoltaic system and may be caused by equipment failure, extreme environmental conditions, or data recording errors. Removing these abnormal values can ensure that the data conforms to the actual physical characteristics and working limits of the photovoltaic system and improve the rationality and credibility of the data.

[0112] S34. Extract multiple statistical features from the preprocessed ratio data, and use the statistical features related to the ratio data as the inverter feature data.

[0113] Specifically, aggregation and selection of statistical features can be performed on two ratios; the selected statistical features can include the mean, median, maximum value, etc. Taking current as an example:

[0114] i avg : The mean of the ratios of the currents of all strings to the STC optimal working current;

[0115] i Q2 : The second quartile (Q2), also known as the "median", is equal to the 50% value after arranging all the values of the ratios of the currents of all strings to the STC optimal working current from small to large;

[0116] i Q3 : The third quartile (Q3), also known as the "upper quartile", is equal to the 75% value after arranging all the values of the ratios of the currents of all strings to the STC optimal working current from small to large;

[0117] i max : The maximum value of the ratios of the currents of all strings to the STC optimal working current.

[0118] The above statistical features are only examples and are not limited thereto. Through these multi-dimensional statistical features, the operating rules and characteristics of the inverter can be understood more deeply, providing a good data basis for subsequent model training.

[0119] In one embodiment, the weather data includes: historical weather forecast data; feature data obtained by extracting features from the historical weather forecast data, such as Figure 4As shown, it includes:

[0120] S41, obtaining power station temperature data and weather type data of each sample photovoltaic power station and within its preset area range in a preset historical time period at a preset time granularity.

[0121] For example, if one wants to obtain temperature data and weather type data of certain sample photovoltaic power stations and power stations within a certain area around them in the past five years at an hourly preset time granularity. First, determine the geographical location coordinates of each photovoltaic power station, then log in to a certain weather database platform, input the coordinate information of the photovoltaic power station, set the time range to the past five years, and the time granularity to 1 hour, then the temperature data of this area can be queried. For weather type data, the database will provide information on weather types such as sunny, cloudy, rainy, snowy, etc. according to its internal meteorological model and data sources, and can be obtained in batches according to the set time granularity.

[0122] S42, encoding the weather type data based on the model type in ensemble learning to obtain weather type encodings.

[0123] For example, when introducing weather features (such as overcast, sunny, etc.) in a regression model, it is usually necessary to convert these categorical variables into numerical variables so that the model can process them. Adopt label encoding to directly encode the weather type into digital labels. For example:

[0124] 'Sunny' → 0;

[0125] 'Overcast' → 1;

[0126] 'Light rain' → 2;

[0127] 'Heavy rain' → 3;

[0128] In comparison, although this encoding method is simpler, it may introduce unnecessary order relationships in the decision tree model (such as the model may mistakenly think that heavy rain has a greater impact than light rain). Therefore, other encoding forms can be adopted, such as one-hot encoding, which is the most commonly used method for processing categorical variables, and each weather type is converted into a binary feature. For example:

[0129] 'Sunny' → [1, 0, 0, 0];

[0130] 'Overcast' → [0, 1, 0, 0];

[0131] 'Light rain' → [0, 0, 1, 0];

[0132] 'Heavy rain' → [0, 0, 0, 1];

[0133] The above encoding method is only for example and not limited thereto.

[0134] S43. Combine the power station temperature data and the corresponding weather type code as weather feature data.

[0135] Specifically, if numerical encoding is adopted, the temperature data can be directly concatenated with the numerical value of the weather type code. Suppose the obtained power station temperature is 25°C and the weather type is sunny (encoded as 1), then the combined weather feature data can be expressed as [25, 1]. If one-hot encoding is adopted, for example, the temperature is 18°C and the weather type is cloudy (one-hot encoding is [0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0]), the combined weather feature data is [18, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0]. This method is simple and direct, and retains the characteristics of the original data.

[0136] Regard the temperature data and the weather type code as different dimensions of a complete feature vector. For example, in a machine learning model, use the combined weather feature data as the input feature vector. The dimension of the feature vector of each sample is fixed. If numerical encoding is adopted, the dimension of the feature vector is 2 (temperature dimension + weather type code dimension); if one-hot encoding is adopted, the dimension of the feature vector is 1 + n (1 temperature dimension + n weather type one-hot encoding dimensions). Such a feature vector can be directly used for subsequent data analysis and model training.

[0137] In one embodiment, the season data includes: obtaining the month M to which the monitoring data and weather data of the photovoltaic power station belong, and using the month as the seasonal feature data. Different months correspond to different seasonal feature data, and seasonal changes have a significant impact on the power generation of the photovoltaic power station. For example, in summer and autumn with rich light resources, the power generation of the photovoltaic power station is usually high, while in winter with shorter light hours and more complex weather conditions, the power generation may decrease.

[0138] In the embodiment of the present invention, by using the month as the seasonal feature data, this seasonal change rule can be intuitively reflected. The month is a simple and clear feature variable. Using it as the seasonal feature data can provide intuitive time dimension information for data analysis and modeling without complex calculations and conversions. Compared with other seasonal indicators that may require a large amount of calculation and processing, month data is easier to obtain and understand, can reduce the difficulty and cost of data processing, and improve the analysis efficiency. When constructing a data analysis model and a prediction model related to a photovoltaic power station, adding this seasonal feature data of the month can enable the model to better capture the seasonal trends and periodic changes in the data. The model can learn the influence patterns of different months on the monitoring data and weather data of the photovoltaic power station, so as to perform data fitting and prediction more accurately.

[0139] In one embodiment, in the process of training a model using ensemble learning based on a dataset to obtain a trained ensemble model as a horizontal plane irradiance prediction model, as Figure 5 shown, the following steps are included:

[0140] S51, dividing the training set constructed from the regional power characteristic data, inverter characteristic data, weather characteristic data, seasonal characteristic data, and the corresponding measured horizontal plane irradiance of a photovoltaic power station into multiple data subsets;

[0141] Specifically, for example, the data of a single power station at a certain moment can be expressed as:

[0142] data = {A avg , A Q1 , A Q2, A Q3 , i avg , i max , …, M}, that is, the set of all features of this power station at this time point. The dataset can be expressed as Dataset = {data 1,1 … data k,j}, where k is the total number of sample photovoltaic power stations and the power stations within their preset regional ranges, and j is the length of the time series.

[0143] The multi-source feature data such as regional power, inverter, weather, and season respectively reflect the factors affecting the horizontal plane irradiance from different aspects. By integrating this information from different sources through ensemble learning, the relationship between irradiance and each influencing factor can be described more comprehensively and accurately, thereby improving the accuracy of prediction.

[0144] The data is divided into a training set and a test set in chronological order. The first 80% of the data is used to train the model, and the last 20% of the data is used for model performance testing to ensure that the time-related characteristics are not disrupted and to improve the generalization ability of the prediction model.

[0145] S52, randomly inputting multiple data subsets into multiple learners, and training each learner separately. The number of learners is equal to the number of data subsets;

[0146] Specifically, common learners include decision trees, support vector machines, random forests, multi-layer perceptrons, etc. In practice, according to the data characteristics and task requirements, a suitable learner is selected for the dataset. In the embodiment of the present invention during ensemble learning training, the training set is divided into two subsets and input into LightGBM and a linear regression model respectively for training. Among them, the weather in the LightGBM dataset uses one-hot encoding, and the weather data in the dataset of the linear regression model uses label encoding, and the test set is divided into two parts to evaluate the performance of the model respectively.

[0147] LightGBM is an efficient distributed gradient boosting framework based on decision trees, suitable for large-scale data and complex feature modeling. Its advantages lie in significantly improving the training speed and memory usage efficiency through techniques such as histogram algorithm, Gradient-based One-Side Sampling (GOSS), and Exclusive Feature Bundling (EFB). To balance non-linear feature capture and linear trend fitting, a weighted combination model is used.

[0148] S53, the integrated weighted model obtained based on the trained learners, is used as the horizontal plane irradiance prediction model. The prediction result of the horizontal plane irradiance prediction model is the weighted average of the output prediction results of each trained learner.

[0149] In one embodiment, the final prediction result is the weighted sum of 60% of the LightGBM prediction value and 40% of the linear regression model prediction value:

[0150] y = 60% * y LightGBM + 40% * y 线性回归

[0151] Through this weighting scheme, the model can simultaneously utilize the non-linear modeling ability of LightGBM and the simple trend capture characteristics of linear regression, thereby improving the prediction accuracy of solar irradiance data and providing a robust performance while maintaining the model complexity.

[0152] In the embodiments of the present invention, by using the method of ensemble learning to weight the prediction results of multiple models, the advantages of each model can be integrated, reducing the possible biases or overfitting problems of a single model. The results of other models can also supplement and correct it, thereby enhancing the accuracy and robustness of the prediction results and being able to adapt to various practical application scenarios such as different seasons, different weather conditions, and different device states.

[0153] In this embodiment, a horizontal plane irradiance prediction system for distributed photovoltaic is also provided. This system is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0154] This embodiment provides a horizontal plane irradiance prediction system for distributed photovoltaic, as Figure 6 shown, including:

[0155] A sample photovoltaic power station selection module 601, configured to select multiple distributed photovoltaic power stations under different geographical locations and climatic conditions as sample photovoltaic power stations;

[0156] The sample data acquisition module 602 is configured to acquire the monitoring data, weather data, season data, and measured horizontal irradiance of all photovoltaic power stations within the preset area of each sample photovoltaic power station.

[0157] The dataset construction module 603 is configured to respectively extract the characteristic data corresponding to the monitoring data, weather data, and season data of the photovoltaic power station, and form a dataset based on the characteristic data and the measured horizontal irradiance.

[0158] The model training module 604 is configured to perform model training using ensemble learning based on the dataset to obtain a trained ensemble model as the horizontal irradiance prediction model.

[0159] The prediction module 605 is configured to input the monitoring data, weather data, and season data of the photovoltaic power station to be predicted into the horizontal irradiance prediction model to obtain the horizontal irradiance prediction result.

[0160] In some alternative embodiments, the sample photovoltaic power station selection module 601 includes:

[0161] The data acquisition unit is configured to acquire the relevant data of the distributed photovoltaic power stations within each administrative region in units of the preset administrative region.

[0162] The sample photovoltaic power station screening unit is configured to screen out the photovoltaic power stations with environmental monitors and meeting the preset stable operation conditions among the distributed photovoltaic power stations as the sample photovoltaic power stations.

[0163] In some alternative embodiments, the monitoring data of the photovoltaic power station includes: photovoltaic power and its installed capacity; the extraction of characteristic data based on the acquired data includes:

[0164] Acquire the historical photovoltaic power data of each sample photovoltaic power station and all photovoltaic power stations within its preset area at a preset time granularity.

[0165] Calculate the ratio of the historical photovoltaic power data of each photovoltaic power station to its photovoltaic installed capacity to obtain the utilization rate of the corresponding photovoltaic system.

[0166] Preprocess the utilization rate of each photovoltaic system, including removing outliers and noise from the data.

[0167] Extract multiple statistical features related to the utilization rate based on the preprocessed utilization rate data, and use at least two statistical features to obtain statistical relationship features representing the relationship between the statistical features. Use the statistical features related to the utilization rate and the statistical relationship features as the regional power characteristic data.

[0168] In some alternative embodiments, the monitoring data of the photovoltaic power station in the sample data acquisition module 602 includes the string current and string voltage data of the inverters; correspondingly, in the data set construction module 603, feature extraction is performed on the acquired data to obtain feature data, including:

[0169] At a preset time granularity, obtain the string current and string voltage of the inverters at each moment of all photovoltaic power stations within each sample photovoltaic power station and its preset area range;

[0170] Calculate the ratio of the string current to the optimal operating current under preset standard conditions, and the ratio of the string voltage of the inverter at each moment to the preset standard optimal operating voltage respectively;

[0171] Preprocess the two calculated ratio data, including removing outliers and noise from the data;

[0172] Extract multiple statistical features based on the preprocessed ratio data, and use the statistical features related to the ratio data as the inverter feature data.

[0173] In some alternative embodiments, the weather data in the sample data acquisition module 602 includes: historical weather forecast data; correspondingly, in the data set construction module 603, feature extraction is performed on the acquired data to obtain feature data, including:

[0174] At a preset time granularity, obtain the power station temperature data and weather type data of each sample photovoltaic power station and its preset area range within a preset historical time period;

[0175] Encode the weather type data based on the model type in the ensemble learning to obtain the weather type code;

[0176] Combine the power station temperature data and the corresponding weather type code as the weather feature data.

[0177] In some alternative embodiments, the season data in the sample data acquisition module 602 includes: obtaining the month to which the monitoring data and weather data of the photovoltaic power station belong, and correspondingly, in the data set construction module 603, using the month to which it belongs as the seasonal feature data.

[0178] In some alternative embodiments, the model training module 604 includes:

[0179] A training set construction unit, configured to divide the training set constructed from the regional power feature data, inverter feature data, weather feature data, seasonal feature data of the photovoltaic power station and the corresponding measured horizontal irradiance into multiple data subsets;

[0180] A training unit for randomly inputting multiple data subsets into multiple learners and training each learner separately, where the number of learners is equal to the number of data subsets;

[0181] A model integration unit for obtaining an integrated weighted model based on the trained learners as a horizontal plane irradiance prediction model, where the prediction result of the horizontal plane irradiance prediction model is the weighted average of the output prediction results of each trained learner.

[0182] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above and will not be elaborated here.

[0183] The horizontal plane irradiance prediction system for distributed photovoltaics in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0184] An embodiment of the present invention further provides a computer device having the above-mentioned Figure 6 horizontal plane irradiance prediction system for distributed photovoltaics shown.

[0185] Please refer to Figure 7 , Figure 7 which is a schematic structural diagram of a computer device provided by an alternative embodiment of the present invention. As shown in Figure 7 , the computer device includes: one or more processors 10, a memory 20, and an interface for connecting each component, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as an array of servers, a set of blade servers, or a multi-processor system). Figure 7 In

[0186] The processor 10 may be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 may further include a hardware chip. The above-mentioned hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device may be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0187] Among them, the memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.

[0188] The memory 20 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely provided with respect to the processor 10, and these remote memories may be connected to the computer device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0189] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 may also include a combination of the above types of memories.

[0190] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0191] The embodiments of the present invention also provide a computer-readable storage medium. The method according to the embodiments of the present invention may be implemented in hardware, firmware, or may be implemented as computer code recorded on a storage medium, or may be implemented as computer code originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and to be stored in a local storage medium, so that the method described herein may be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium may be a magnetic disk, an optical disc, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium may also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor central control system, or programmable hardware includes a storage component capable of storing or receiving software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0192] A part of the present invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the present invention through the operation of the computer. Those skilled in the art should understand that the forms of existence of computer program instructions in a computer-readable medium include but are not limited to source files, executable files, installation package files, etc. Correspondingly, the ways for computer program instructions to be executed by a computer include but are not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible by the computer.

[0193] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for predicting horizontal irradiance of distributed photovoltaics, characterized in that: include: Select multiple distributed photovoltaic power stations in different geographical locations and climatic conditions as sample photovoltaic power stations; Obtain monitoring data, weather data, seasonal data and measured horizontal irradiance of all photovoltaic power stations within the preset area of ​​each sample photovoltaic power station; Extract characteristic data corresponding to the monitoring data, weather data and seasonal data of the photovoltaic power station respectively, and form a data set based on the characteristic data and the measured horizontal plane irradiance; Based on the data set, model training is performed using ensemble learning to obtain a trained ensemble model as a horizontal surface irradiance prediction model; The monitoring data, weather data and season data of the photovoltaic power station to be predicted are input into the horizontal plane irradiance prediction model to obtain the horizontal plane irradiance prediction result.

2. The method according to claim 1, characterized in that The plurality of photovoltaic power stations in different geographical locations and climate conditions are selected as sample photovoltaic power stations, including: Taking the preset administrative area as the unit, obtain the relevant data of distributed photovoltaic power stations in each administrative area; The photovoltaic power stations with environmental monitoring instruments and meeting the preset stable operation conditions are selected from the distributed photovoltaic power stations as sample photovoltaic power stations.

3. The method according to claim 1, characterized in that: The monitoring data of the photovoltaic power station includes: photovoltaic power and its installed capacity; feature data is obtained by extracting features based on the acquired monitoring data, including: Obtain historical photovoltaic power data of each sample photovoltaic power station and all photovoltaic power stations within a preset area at a preset time granularity; Calculate the ratio of the historical photovoltaic power data of each photovoltaic power station to its photovoltaic installed capacity to obtain the utilization rate of the corresponding photovoltaic system; Preprocess the utilization of each photovoltaic system, including removing outliers and noise from the data; Based on the preprocessed utilization data, multiple statistical features related to utilization are extracted, and statistical relationship features representing the statistical feature relationship are obtained using at least two statistical features. The statistical features related to utilization and the statistical relationship features are used as regional power feature data.

4. The method according to claim 1, characterized in that The monitoring data of the photovoltaic power station includes the string current and string voltage data of the inverter; feature data is obtained by performing feature extraction based on the acquired monitoring data, including: Obtain the string current and string voltage of the inverters of all photovoltaic power stations at each time within each sample photovoltaic power station and its preset area at a preset time granularity; Calculate the ratio of the string current to the optimal working current under the preset standard conditions, and the ratio of the string voltage of the inverter to the preset standard optimal working voltage at each moment; Preprocessing the two calculated ratio data includes removing outliers and noise from the data; A plurality of statistical features are extracted based on the preprocessed ratio data, and the statistical features related to the ratio data are used as inverter feature data.

5. The method according to claim 1, characterized in that The weather data includes: weather forecast historical data; feature data obtained by feature extraction based on the acquired weather data, including: Obtain the power station temperature data and weather type data of each sample photovoltaic power station and its preset area within a preset historical time period at a preset time granularity; Encode the weather type data based on the model type in the ensemble learning to obtain the weather type code; The power station temperature data and the corresponding weather type code are combined as weather characteristic data.

6. The method according to claim 1, characterized in that The seasonal data includes: obtaining the month to which the monitoring data and weather data of the photovoltaic power station belong, and using the month as seasonal characteristic data.

7. The method according to any one of claims 1 to 6, characterized in that: The method of performing model training by using ensemble learning based on the data set to obtain a trained ensemble model as a horizontal surface irradiance prediction model includes: Divide the training set constructed by the regional power characteristic data, inverter characteristic data, weather characteristic data, seasonal characteristic data of the photovoltaic power station and the corresponding measured horizontal plane irradiance into multiple data subsets; Randomly inputting multiple data subsets into multiple learners, and training each learner separately, wherein the number of learners is equal to the number of data subsets; The integrated weighted model obtained based on the trained learners is used as the horizontal plane irradiance prediction model, and the prediction result of the horizontal plane irradiance prediction model is the weighted average of the prediction results output by each trained learner.

8. A distributed photovoltaic horizontal surface irradiance prediction system, characterized in that: include: A sample photovoltaic power station selection module is used to select multiple distributed photovoltaic power stations in different geographical locations and climatic conditions as sample photovoltaic power stations; The sample data acquisition module is used to obtain the monitoring data, weather data, seasonal data and measured horizontal irradiance of all photovoltaic power stations within the preset area of ​​each sample photovoltaic power station; A data set construction module is used to extract characteristic data corresponding to monitoring data, weather data and seasonal data of the photovoltaic power station, and form a data set based on the characteristic data and the measured horizontal plane irradiance; A model training module, used for performing model training by using ensemble learning based on the data set to obtain a trained ensemble model as a horizontal surface irradiance prediction model; The prediction module is used to input the monitoring data, weather data and season data of the photovoltaic power station to be predicted into the horizontal plane irradiance prediction model to obtain the horizontal plane irradiance prediction result.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the distributed photovoltaic horizontal plane irradiance prediction method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the distributed photovoltaic horizontal plane irradiance prediction method according to any one of claims 1-7.

11. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to enable a computer to execute the distributed photovoltaic horizontal plane irradiance prediction method according to any one of claims 1 to 7.

Images