Distributed photovoltaic load prediction method and device based on ISO and LightGBM, and medium

Abstract

The invention discloses a distributed photovoltaic load prediction method and device based on ISO and LightGBM, and a medium, and the method comprises the steps: carrying out the load prediction of meteorological load data through a LightGBM model if the data size of the meteorological load data is larger than a preset threshold range, and obtaining a load prediction value; if the data volume is within the threshold range, performing load prediction on the meteorological load data through a physical model to obtain a load prediction value; if the data volume is smaller than the threshold range, load prediction is carried out on the meteorological load data in a power curve fitting mode, and a load prediction value is obtained. According to the distributed photovoltaic load prediction method and device based on ISO and LightGBM and the medium provided by the invention, a machine learning model, a physical model and a statistical method are comprehensively applied, and an appropriate prediction method is selected according to different data conditions, so that the advantages of various methods can be brought into full play, the prediction accuracy and efficiency are improved, and the prediction efficiency is improved. The problem that the accuracy of distributed photovoltaic load prediction is difficult to improve can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system load analysis, and in particular to a distributed photovoltaic load prediction method, device and medium based on ISO and LightGBM. Background Art

[0002] Distributed photovoltaics, as a clean and renewable energy utilization method, is gradually becoming an important force in promoting energy transformation and promoting sustainable economic and social development; distributed photovoltaic systems, as a decentralized power system that directly converts solar energy into electrical energy, can reduce electricity bills, improve energy efficiency, and promote energy structure optimization by reducing power transmission losses; through load forecasting, we can know the electricity demand in the future in advance, so that distributed photovoltaic systems can reasonably arrange power generation plans and improve the stability and reliability of the system. Traditional load forecasting methods usually use statistical models for forecasting. These models include time series analysis, regression analysis, etc., and establish a mathematical relationship between load and influencing factors by fitting historical data and estimating parameters, so as to achieve the prediction of future loads.

[0003] However, photovoltaic power generation is affected by many complex factors, including light intensity, temperature, cloud cover, wind speed, and the performance of photovoltaic modules, which are highly random and unpredictable. Since traditional prediction methods often rely on historical data and simple statistical models, it is difficult to effectively capture complex nonlinear relationships. Therefore, the intermittent and uncertain nature of photovoltaic power generation poses severe challenges to load forecasting, resulting in low prediction accuracy. Summary of the invention

[0004] The present invention provides a distributed photovoltaic load forecasting method, device and medium based on ISO and LightGBM to solve the problem of difficulty in improving the accuracy of distributed photovoltaic load forecasting.

[0005] Obtain meteorological load data of the target PV site;

[0006] If the amount of the meteorological load data is greater than a preset threshold range, the meteorological load data is subjected to load forecasting by using the LightGBM model to obtain a load forecast value;

[0007] If the data volume of the meteorological load data is within the threshold range, load forecasting is performed on the meteorological load data through a physical model to obtain a load forecast value; wherein the physical model is established based on the meteorological load data of the neighboring photovoltaic sites of the target photovoltaic site;

[0008] If the data volume of the meteorological load data is less than the threshold range, load prediction is performed on the meteorological load data by fitting a power curve to obtain a load prediction value.

[0009] When the amount of meteorological load data is large enough, the present invention uses the LightGBM model for load forecasting; the LightGBM model is a data-driven machine learning model, which is widely used to process large-scale data sets and complex forecasting tasks. The performance of the model often improves with the increase of data volume, so it can well capture complex nonlinear relationships and feature interactions, and achieve high-precision load forecasting. When the amount of data is not enough to support a complex machine learning model, a method based on a physical model is used instead. Since the historical meteorological load data of the target photovoltaic site may also have insufficient samples, the meteorological load data of adjacent photovoltaic sites is used to establish a physical model, which can fully utilize the principle of spatial proximity to form data complementarity, thereby better reflecting the load characteristics of the target site and reducing the requirements for the amount of data. For situations less than the threshold range, the prediction is made by fitting the power curve. This method does not rely on a complete data set, but uses existing data trends and patterns to infer the value of the missing part, thereby improving the robustness of the prediction.

[0010] Compared with the existing technology, the present invention comprehensively applies machine learning models, physical models and statistical methods, and selects appropriate prediction methods according to different data conditions, which is conducive to giving full play to the advantages of various methods and improving the accuracy and efficiency of prediction. Therefore, it can solve the problem of difficulty in improving the accuracy of distributed photovoltaic load prediction.

[0011] As a preferred solution, the meteorological load data is subjected to load forecasting by the LightGBM model to obtain a load forecast value, specifically:

[0012] Calculating the correlation between each feature in the meteorological load data and the load forecast result to obtain a set of correlation coefficients;

[0013] According to the correlation coefficient set, data with correlation higher than a preset standard value is screened out from the meteorological load data to obtain a data set;

[0014] Performing clustering processing on the data set to obtain a feature data set;

[0015] The load prediction is performed on the feature data set through the LightGBM model to obtain the load prediction value.

[0016] This preferred solution can effectively remove those features that have little or no impact on the forecast results by calculating the correlation between each feature and the load forecast result and screening out data with a correlation higher than the preset standard value, thereby improving the quality of the data and the accuracy of the forecast. Clustering the screened data set can further group similar features together to form a more representative feature data set, thereby reducing the dimension of the data, and can better understand the intrinsic relationship between the data based on the clustering results, which helps to improve the interpretability and predictive ability of the model.

[0017] As a preferred solution, clustering is performed on the data set to obtain a feature data set, specifically:

[0018] Randomly select a number of data points from the data set as initial centroids;

[0019] By measuring the Euclidean distance between the data points and the cluster centers, each of the data points is assigned to the nearest cluster center among the cluster centers, thereby obtaining a first assignment result;

[0020] Calculate the centroids of the plurality of cluster centers based on the first allocation result to obtain a new centroid set;

[0021] According to the new centroid set, the corresponding clusters are merged or split until the cluster centers of the clusters no longer change or a preset number of iterations is reached, thereby obtaining a cluster division result;

[0022] Feature extraction is performed on basic statistics of the division result to obtain the feature data set.

[0023] This preferred solution randomly selects multiple initial centroids and iterates based on these centroids, which can increase the possibility of finding a global optimal solution and prevent the algorithm from falling into a local optimal solution. By continuously iteratively calculating cluster centers and merging or splitting clusters based on new centroid sets, the algorithm can automatically adjust the number of clusters according to the actual distribution of the data and continuously optimize the cluster division results, making the data points in each cluster closer and the distance between clusters farther, which helps to improve the quality and accuracy of clustering.

[0024] As a preferred solution, the LightGBM model is specifically:

[0025] According to the historical meteorological load samples, the residual of the mth iteration is calculated through the preset initial LightGBM model to obtain the current residual;

[0026] Establishing a decision tree based on the current residual;

[0027] In the case of maximizing the reduction of residual error through the decision tree, the initial LightGBM model is updated according to the learning rate to obtain the LightGBM model.

[0028] In each round of iteration of this preferred solution, the LightGBM model will build a new decision tree based on the residual of the current model and gradually update it through the learning rate. This gradient-based boosting strategy enables the model to continuously approach the true value, thereby improving the accuracy of the prediction; especially when processing complex meteorological load data, the LightGBM model can capture the nonlinear relationships and potential patterns in the data, thereby improving the accuracy of load forecasting.

[0029] As a preferred solution, the physical model is established based on the meteorological load data of the neighboring photovoltaic sites of the target photovoltaic site, specifically:

[0030] In the photovoltaic site set where the target photovoltaic site is located, a distance matrix between all photovoltaic sites is calculated according to the longitude and latitude of each photovoltaic site;

[0031] Finding a first photovoltaic site closest to the target photovoltaic site according to the distance matrix;

[0032] Calculate a historical data calculation coefficient set according to the historical meteorological load data of the first photovoltaic site and the target photovoltaic site;

[0033] A coefficient set is calculated based on the historical data, and the physical model is established according to the temperature, humidity and wind speed at different times.

[0034] This preferred solution calculates the distance matrix between photovoltaic sites and finds the first photovoltaic site closest to the target photovoltaic site, thereby making full use of the principle of spatial proximity. By introducing the data of neighboring sites, data complementarity can be formed to reduce dependence on a single data source. In addition, the establishment of the physical model takes into account meteorological factors such as temperature, humidity and wind speed at different times. By integrating these real-time meteorological data into the model, the model can dynamically adjust the prediction results to better adapt to actual meteorological conditions and load changes, thereby enhancing the adaptability and robustness of the model.

[0035] As a preferred solution, load forecasting is performed on the meteorological load data by fitting a power curve to obtain a load forecast value, specifically:

[0036] The meteorological load data is supplemented by a linear interpolation method to obtain complete data;

[0037] If the installed capacity of the system where the target photovoltaic site is located has not changed, then the load forecast value is calculated by fitting the theoretical basic power curve based on the complete data;

[0038] If the installed capacity of the system where the target photovoltaic site is located changes, the theoretical basic power curve is scaled to obtain a final theoretical basic power curve; based on the complete data, the load forecast value is calculated according to the final theoretical basic power curve.

[0039] This preferred solution uses linear interpolation to supplement the missing meteorological load data, which can ensure the integrity and continuity of the data and avoid incomplete data sets that may lead to inaccurate or biased prediction results. When the installed capacity of the system where the target photovoltaic site is located has not changed, the load forecast value is calculated directly based on the complete data to fit the theoretical basic power curve. This can be simple and direct and can better reflect the actual operating conditions of the system, thereby improving the accuracy of the prediction. When the installed capacity of the system changes, the theoretical basic power curve is scaled to adapt to the new capacity. This method allows the system to respond quickly to changes in system capacity and adjust the prediction results accordingly to ensure that the predicted value is consistent with the actual load conditions.

[0040] As a preferred solution, the final theoretical basic power curve is specifically:

[0041] P=S(aI 2 +bI+c)

[0042] Where I is the light intensity, a, b, and c are the least squares fitting parameters, and S is the scaling factor.

[0043] The present invention also provides a distributed photovoltaic load prediction device based on ISO and LightGBM, comprising a data module, a first prediction module, a second prediction module and a third prediction module;

[0044] Wherein, the data module is used to obtain the meteorological load data of the target photovoltaic site;

[0045] The first prediction module is used to perform load prediction on the meteorological load data through the LightGBM model to obtain a load prediction value if the data volume of the meteorological load data is greater than a preset threshold range;

[0046] The second prediction module is used to perform load prediction on the meteorological load data through a physical model to obtain a load prediction value if the data volume of the meteorological load data is within the threshold range; wherein the physical model is established based on the meteorological load data of the neighboring photovoltaic sites of the target photovoltaic site;

[0047] The third prediction module is used to perform load prediction on the meteorological load data by fitting a power curve to obtain a load prediction value if the data volume of the meteorological load data is less than the threshold range.

[0048] As a preferred solution, the first prediction module includes a coefficient unit, a data unit, a clustering unit and a prediction unit;

[0049] The coefficient unit is used to calculate the correlation between each feature in the meteorological load data and the load forecast result to obtain a set of correlation coefficients;

[0050] The data unit is used to filter out data with correlation higher than a preset standard value from the meteorological load data according to the correlation coefficient set to obtain a data set;

[0051] The clustering unit is used to perform clustering processing on the data set to obtain a feature data set;

[0052] The prediction unit is used to perform load prediction on the feature data set through the LightGBM model to obtain the load prediction value.

[0053] As a preferred solution, the clustering unit includes a centroid subunit, a distribution subunit, a calculation subunit, a result subunit and a feature subunit;

[0054] Wherein, the centroid subunit is used to randomly select a number of data points from the data set as initial centroids;

[0055] The allocation subunit is used to allocate each of the data points to the nearest cluster center among the cluster centers by measuring the Euclidean distance between the data points and the cluster centers, so as to obtain a first allocation result;

[0056] The calculation subunit is used to calculate the centroids of the plurality of cluster centers based on the first allocation result to obtain a new centroid set;

[0057] The result subunit is used to merge or split the corresponding clusters according to the new centroid set until the cluster centers of the clusters no longer change or a preset number of iterations is reached, thereby obtaining a cluster division result;

[0058] The feature subunit is used to extract features from the basic statistics of the segmentation result to obtain the feature data set.

[0059] As a preferred solution, the LightGBM model is specifically:

[0060] According to the historical meteorological load samples, the residual of the mth iteration is calculated through the preset initial LightGBM model to obtain the current residual;

[0061] Establishing a decision tree based on the current residual;

[0062] In the case of maximizing the reduction of residual error through the decision tree, the initial LightGBM model is updated according to the learning rate to obtain the LightGBM model.

[0063] As a preferred solution, the second prediction module includes a matrix unit, a site unit, a history unit and a physical unit;

[0064] The matrix unit is used to calculate the distance matrix between all photovoltaic sites in the photovoltaic site set where the target photovoltaic site is located according to the longitude and latitude of each photovoltaic site;

[0065] The site unit is used to find a first photovoltaic site closest to the target photovoltaic site according to the distance matrix;

[0066] The history unit is used to calculate a historical data calculation coefficient set according to the historical meteorological load data of the first photovoltaic site and the target photovoltaic site;

[0067] The physical unit is used to calculate the coefficient set based on the historical data and establish the physical model according to the temperature, humidity and wind speed at different times.

[0068] As a preferred solution, the third prediction module includes an interpolation unit, a capacity unit and a change unit;

[0069] Wherein, the interpolation unit is used to supplement the meteorological load data by linear interpolation to obtain complete data;

[0070] The capacity unit is used to calculate the load prediction value by fitting the theoretical basic power curve according to the complete data if the installed capacity of the system where the target photovoltaic site is located has not changed;

[0071] The change unit is used to scale the theoretical basic power curve to obtain a final theoretical basic power curve if the installed capacity of the system where the target photovoltaic site is located changes; based on the complete data, the load prediction value is calculated according to the final theoretical basic power curve.

[0072] As a preferred solution, the final theoretical basic power curve is specifically:

[0073] P=S(aI 2 +bI+c)

[0074] Where I is the light intensity, a, b, and c are the least squares fitting parameters, and S is the scaling factor.

[0075] The present application also provides a storage medium having a computer program stored thereon, and the computer program is called and executed by a computer to implement the distributed photovoltaic load forecasting method based on ISO and LightGBM as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow chart of a distributed photovoltaic load forecasting method based on ISO and LightGBM provided in an embodiment of the present application;

[0077] Figure 2 It is an overall flow chart provided by the embodiment of the present application;

[0078] Figure 3 It is a structural schematic diagram of a distributed photovoltaic load forecasting device based on ISO and LightGBM provided in an embodiment of the present application. DETAILED DESCRIPTION

[0079] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0080] In the description of the present application, it should be understood that the terms "first", "second" and "third" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first", "second" and "third" may explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, "several" means two or more.

[0081] The distributed photovoltaic load forecasting method based on ISO and LightGBM provided in the embodiment of the present application is mainly used in situations where it is necessary to capture the complex nonlinear relationship in meteorological load data and improve the accuracy of distributed photovoltaic load forecasting.

[0082] Embodiment 1:

[0083] See also Figure 1 The embodiment of the present application provides a distributed photovoltaic load forecasting method based on ISO and LightGBM, including S1 to S4, and the specific implementation steps are as follows:

[0084] S1. Obtain meteorological load data of the target photovoltaic site.

[0085] Step S1 of the embodiment of the present application is specifically as follows:

[0086] Obtaining meteorological load data of a target photovoltaic site; wherein the meteorological load data includes historical meteorological data and historical load data;

[0087] Meteorological load data are divided according to the number of data days. Data with more than 100 days are classified as large data volume type; data with more than or equal to 10 days and less than or equal to 100 days are classified as small data volume type; data with less than 10 days are classified as missing data type.

[0088] S2. If the amount of meteorological load data is greater than the preset threshold range, the LightGBM model is used to perform load forecasting on the meteorological load data to obtain the load forecast value.

[0089] Step S2 of the embodiment of the present application includes S2.1 to S2.3; wherein S2.1 is a process of obtaining a data set according to the Pearson correlation analysis method, S2.2 is a process of obtaining a feature data set by clustering according to the ISODATA algorithm, and S2.3 is a process of using the LightGBM model to perform load forecasting on the feature data set to obtain a load forecast value, specifically:

[0090] S2.1. If the number of days of meteorological load data is greater than 100 days, the Pearson correlation analysis method is used to calculate the correlation between each feature and the load forecast result of the meteorological load data in the three dimensions of time, weather, and historical load, and obtain a set of correlation coefficients; wherein, the correlation coefficient r in the set of correlation coefficients is used to measure the strength and direction of the linear relationship between two continuous variables, and its value is between -1 and 1; and, r = 1, indicating a complete positive correlation, there is a perfect linear relationship between the two variables; r = -1, indicating a complete negative correlation, there is a perfect negative linear relationship between the two variables; r = 0, indicating no linear correlation, there is no linear relationship between the two variables;

[0091] According to the correlation coefficient set, the data with correlation higher than the preset standard value is screened out from the meteorological load data to obtain a data set with higher correlation;

[0092] Among them, the calculation formula of the correlation coefficient r in the correlation coefficient set is:

[0093]

[0094] Among them, x i and i are the i-th value of feature X and the i-th value of load Y, respectively. and are the average values ​​of feature X and load Y respectively, and n is the data length.

[0095] S2.2, randomly select k data points from the data set as the initial centroids, and set the number of initial clusters k and other parameters, such as distance threshold, minimum cluster size, etc.;

[0096] By measuring the Euclidean distance between the data points and the cluster centers, each of the data points is assigned to the nearest cluster center among the cluster centers, thereby obtaining a first assignment result;

[0097] Based on the first allocation result, the centroids of several cluster centers are calculated using the weighted average method to obtain a new centroid set;

[0098] According to the new centroid set and pre-set conditions, the corresponding clusters are merged or split until the cluster centers of the clusters no longer change or the preset number of iterations is reached, and the cluster division results are obtained; wherein, the merging and splitting processing methods are specifically as follows: when the number of points in a cluster exceeds the size threshold of the maximum cluster, it is split, and the new cluster center can be determined by re-clustering the points in the cluster; when the distance between two clusters is less than the minimum distance threshold, the two clusters are merged into one cluster.

[0099] Feature extraction is performed on the basic statistics of the partitioning results to obtain a feature data set.

[0100] Among them, the Euclidean distance is:

[0101]

[0102] The new centroid in the new centroid set is:

[0103]

[0104] Where x is the data point, c j is the cluster center, n is the feature dimension, x i is the ith data point, is the cluster center corresponding to the i-th data point;

[0105] S j is the set of data points in the jth cluster, |S j | is the number of data points in the jth cluster.

[0106] Embodiment S2.2 of the present application is based on the ISODATA algorithm, randomly selects multiple initial centroids and iterates based on these centroids, which can increase the possibility of finding a global optimal solution and prevent the algorithm from falling into a local optimal solution. By continuously iteratively calculating cluster centers and merging or splitting clusters based on new centroid sets, the algorithm can automatically adjust the number of clusters according to the actual distribution of the data, and continuously optimize the cluster division results, so that the data points in each cluster are closer and the distance between clusters is farther, which helps to improve the quality and accuracy of clustering.

[0107] S2.3. Initialize the preset initial LightGBM model (Light Gradient Boosting Machine); the initial prediction value of the initial LightGBM model is the mean of the target variable;

[0108] According to the historical meteorological load samples, the residual of the mth iteration is calculated through the initial LightGBM model to obtain the current residual;

[0109] Build a decision tree f based on the current residual m , so that the decision tree f m Can minimize residual error;

[0110] In the decision tree f m While minimizing the residual error, the predicted value of the initial LightGBM model is updated according to the learning rate η to obtain the LightGBM model.

[0111] The LightGBM model is used to perform load forecasting on the feature data set to obtain the load forecast value.

[0112] The initial prediction value is:

[0113]

[0114] The residual of the mth iteration is:

[0115]

[0116] Update the prediction value of the initial LightGBM model, which can be expressed as:

[0117]

[0118] The prediction formula of the LightGBM model is:

[0119]

[0120] in, is the initial prediction value, N is the number of samples, y i is the true value of the i-th sample;

[0121] is the predicted value corresponding to the i-th sample in the m-1th iteration, η is the learning rate, and f m is the decision tree at the mth iteration, x i is the i-th value of feature X;

[0122] is the final prediction value, and M is the total number of decision trees.

[0123] In each round of iteration of Example S2.3 of the present application, the LightGBM model constructs a new decision tree based on the residual of the current model and gradually updates it through the learning rate. This gradient-based boosting strategy enables the model to continuously approach the true value, thereby improving the accuracy of the prediction; especially when processing complex meteorological load data, the LightGBM model can capture the nonlinear relationships and potential patterns in the data, thereby improving the accuracy of load forecasting.

[0124] Overall, by calculating the correlation between each feature and the load forecast result and screening out data with a correlation higher than a preset standard value, the embodiment S2 of the present application can effectively remove those features that have little or no impact on the forecast result, thereby improving the quality of the data and the accuracy of the forecast. Clustering the screened data set can further group similar features together to form a more representative feature data set, thereby reducing the dimension of the data, and can better understand the intrinsic relationship between the data based on the clustering results, which helps to improve the interpretability and predictive ability of the model;

[0125] In addition, the LightGBM model is a data-driven machine learning model that is widely used to process large-scale data sets and complex prediction tasks. The performance of the model often improves with the increase in the amount of data. A large amount of data can provide rich samples for the model, enabling it to better capture complex nonlinear relationships and feature interactions; therefore, using this model when you have a large amount of data can better achieve high-precision predictions.

[0126] S3. If the amount of meteorological load data is within a threshold range, load forecasting is performed on the meteorological load data through a physical model to obtain a load forecast value; wherein the physical model is established based on meteorological load data of adjacent photovoltaic sites of the target photovoltaic site.

[0127] Step S3 of the embodiment of the present application includes S3.1 to S3.2; wherein S3.1 is the process of establishing a physical model, and S3.2 is the process of performing load forecasting on meteorological load data through the physical model to obtain a load forecast value, specifically:

[0128] S3.1. In the photovoltaic site set where the target photovoltaic site is located, use the spherical distance formula (Haversine formula) to calculate the distance between the longitude and latitude points of two photovoltaic sites to obtain the site distance; where the longitude and latitude of the photovoltaic sites are (φ i ,λ i ), where i is the site index;

[0129] Traverse the photovoltaic sites in the photovoltaic site set to obtain a distance matrix D between all photovoltaic sites consisting of several site distances;

[0130] Find the first photovoltaic site closest to the target photovoltaic site according to the distance matrix D;

[0131] A historical data calculation coefficient set is obtained by calculating the historical meteorological load data of the first photovoltaic site and the target photovoltaic site;

[0132] The coefficient set is calculated based on historical data, and a physical model is established based on temperature, humidity and wind speed at different times.

[0133] The site distance is:

[0134]

[0135] The site formula corresponding to the first photovoltaic site is:

[0136] j min =argminD kj

[0137] The calculation formula of the historical data calculation coefficient in the historical data calculation coefficient set is:

[0138] β=(X T X) -1 X T P

[0139] The physical model is:

[0140]

[0141] Among them, d ij is the distance between site i and site j, R is the average radius of the earth; φ i and φ i are the longitude and latitude of site i, φ j and λ j are the longitude and latitude of station j, respectively, in radians;

[0142] D kj is the site distance corresponding to the target site k;

[0143] X is the characteristic matrix, which contains constant terms and all characteristic variables; P is the target variable vector, which contains the historical power generation load, and T represents the sign of the transposed matrix.

[0144] is the photovoltaic power generation load at the tth moment, T t , H t , R t and W t are the temperature, humidity, radiation intensity and wind speed at the tth moment; β 1 , β 2 , β 3 and β4 are the historical data calculation coefficients corresponding to temperature, humidity, radiation intensity and wind speed, β 0 The parameter is the photovoltaic power generation load data at time t-1.

[0145] Embodiment S3.1 of the present application calculates the distance matrix between photovoltaic sites and finds the first photovoltaic site closest to the target photovoltaic site, thereby making full use of the principle of spatial proximity. By introducing the data of neighboring sites, data complementarity can be formed to reduce dependence on a single data source. In addition, the establishment of the physical model takes into account meteorological factors such as temperature, humidity and wind speed at different times. By integrating these real-time meteorological data into the model, the model can dynamically adjust the prediction results to better adapt to actual meteorological conditions and load changes, thereby enhancing the adaptability and robustness of the model.

[0146] In addition, physical models rely on modeling of systems or existing theories and have less reliance on large-scale data. In the case of small data volumes, data-driven models such as the LightGBM model may be difficult to generalize correctly due to insufficient data, while physical models can make more accurate predictions.

[0147] S3.2. If the number of days of meteorological load data is greater than or equal to 10 days and less than or equal to 100 days, the meteorological load data is subjected to load forecasting through a physical model to obtain a load forecast value.

[0148] S4. If the amount of meteorological load data is less than a threshold range, load prediction is performed on the meteorological load data by fitting a power curve to obtain a load prediction value.

[0149] Step S4 of the embodiment of the present application is specifically as follows:

[0150] If the number of days of meteorological load data is less than 10 days, the meteorological load data is cleaned to remove obvious outliers and noise; and the meteorological load data is supplemented by linear interpolation to obtain complete data;

[0151] If the installed capacity of the system where the target PV site is located has not changed, the load forecast value is calculated by fitting the theoretical basic power curve based on the complete data;

[0152] If the installed capacity of the system where the target PV site is located changes, the theoretical basic power curve is adjusted to the new installed capacity level by scaling the theoretical basic power curve to obtain the final theoretical basic power curve;

[0153] Based on the complete data, the load forecast value is calculated according to the final theoretical basic power curve.

[0154] Among them, the theoretical basic power curve is:

[0155] P=aI 2 +bI+c

[0156] The final theoretical basic power curve is:

[0157] P=S(aI 2 +bI+c)

[0158] For parameter S:

[0159]

[0160] Where I is the light intensity, a, b, and c are the least squares fitting parameters, and S is the scaling factor.

[0161] C old is the original installed capacity of the system where the target PV site is located, C new is the new installed capacity of the system where the target PV site is located.

[0162] To apply this embodiment, please refer to Figure 2 , Figure 2 It is an overall flow chart provided in an embodiment of the present application, which shows the general process of obtaining the load forecast value as a forecast result in the first embodiment of the present application.

[0163] Embodiment S4 of the present application uses linear interpolation to supplement the missing meteorological load data, which can ensure the integrity and continuity of the data and avoid incomplete data sets that may cause inaccurate or biased prediction results. When the installed capacity of the system where the target photovoltaic site is located has not changed, directly fitting the theoretical basic power curve based on the complete data to calculate the load forecast value can be simple and direct and can better reflect the actual operating conditions of the system, thereby improving the accuracy of the prediction. When the installed capacity of the system changes, the theoretical basic power curve is scaled to adapt to the new capacity situation. This method allows the system to respond quickly to changes in system capacity and adjust the prediction results accordingly to ensure that the predicted value is consistent with the actual load situation;

[0164] Moreover, when data is missing, in order to achieve prediction results, it is easier to maintain performance stability by using a power curve model with a simpler structure.

[0165] Overall, the embodiments of the present application have the following beneficial effects:

[0166] When the amount of meteorological load data is large enough, the present invention uses the LightGBM model for load forecasting; the LightGBM model is a data-driven machine learning model, which is widely used to process large-scale data sets and complex forecasting tasks. The performance of the model often improves with the increase of data volume, so it can well capture complex nonlinear relationships and feature interactions, and achieve high-precision load forecasting. When the amount of data is not enough to support a complex machine learning model, a method based on a physical model is used instead. Since the historical meteorological load data of the target photovoltaic site may also have insufficient samples, the meteorological load data of adjacent photovoltaic sites is used to establish a physical model, which can fully utilize the principle of spatial proximity to form data complementarity, thereby better reflecting the load characteristics of the target site and reducing the requirements for the amount of data. In the case of missing data, the power curve is fitted for prediction. This method does not rely on a complete data set, but uses existing data trends and patterns to infer the value of the missing part, thereby improving the robustness of the prediction.

[0167] Embodiment 2:

[0168] See also Figure 3 , an embodiment of the present application provides a distributed photovoltaic load prediction device based on ISO and LightGBM, including a data module 10, a first prediction module 20, a second prediction module 30 and a third prediction module 40;

[0169] Wherein, the data module 10 is used to obtain the meteorological load data of the target photovoltaic site;

[0170] The first prediction module 20 is used to perform load prediction on the meteorological load data by using the LightGBM model to obtain a load prediction value if the data volume of the meteorological load data is greater than a preset threshold range;

[0171] The second prediction module 30 is used to perform load prediction on the meteorological load data through a physical model to obtain a load prediction value if the data volume of the meteorological load data is within a threshold range; wherein the physical model is established based on the meteorological load data of the neighboring photovoltaic sites of the target photovoltaic site;

[0172] The third prediction module 40 is used to perform load prediction on the meteorological load data by fitting the power curve to obtain a load prediction value if the data volume of the meteorological load data is less than a threshold range.

[0173] In one embodiment, the data module 10 is specifically:

[0174] Obtaining meteorological load data of a target photovoltaic site; wherein the meteorological load data includes historical meteorological data and historical load data;

[0175] Meteorological load data are divided according to the number of data days. Data with more than 100 days are classified as large data volume type; data with more than or equal to 10 days and less than or equal to 100 days are classified as small data volume type; data with less than 10 days are classified as missing data type.

[0176] In one embodiment, the first prediction module 20 includes a coefficient unit, a data unit, a centroid subunit, an allocation subunit, a calculation subunit, a result subunit, a feature subunit, a model building unit, and an actual prediction unit;

[0177] Among them, the coefficient unit and data unit are the process of obtaining the data set according to the Pearson correlation analysis method, the centroid subunit, allocation subunit, calculation subunit, result subunit and feature subunit are the process of clustering according to the ISODATA algorithm to obtain the feature data set, and the model building unit and actual prediction unit are the process of using the LightGBM model to perform load forecasting on the feature data set to obtain the load forecast value, specifically:

[0178] The coefficient unit is used to calculate the correlation between each feature and the load forecast result of the meteorological load data under the three dimensions of time, meteorology and historical load using the Pearson correlation analysis method if the number of days of meteorological load data is greater than 100, and obtain a set of correlation coefficients; wherein the correlation coefficient r in the set of correlation coefficients is used to measure the strength and direction of the linear relationship between two continuous variables, and its value is between -1 and 1; and, r = 1, indicating a complete positive correlation, there is a perfect linear relationship between the two variables; r = -1, indicating a complete negative correlation, there is a perfect negative linear relationship between the two variables; r = 0, indicating no linear correlation, there is no linear relationship between the two variables;

[0179] The data unit is used to screen out data with a correlation higher than a preset standard value from the meteorological load data according to the correlation coefficient set, so as to obtain a data set with a higher correlation;

[0180] Among them, the calculation formula of the correlation coefficient r in the correlation coefficient set is:

[0181]

[0182] Among them, x i and i are the i-th value of feature X and the i-th value of load Y, respectively. and are the average values ​​of feature X and load Y respectively, and n is the data length.

[0183] The centroid subunit is used to randomly select k data points from the dataset as the initial centroids and set the number of initial clusters k and other parameters such as the distance threshold, the size of the minimum cluster, etc.

[0184] an allocation subunit, configured to allocate each of the plurality of data points to a corresponding closest cluster center among the plurality of cluster centers by measuring the Euclidean distance between the plurality of data points and the plurality of cluster centers, so as to obtain a first allocation result;

[0185] A calculation subunit, configured to calculate the centroids of a plurality of cluster centers by using a weighted average method based on the first allocation result to obtain a new centroid set;

[0186] The result subunit is used to merge or split the corresponding clusters according to the new centroid set and the preset conditions, until the cluster centers of the clusters no longer change or the preset number of iterations is reached, so as to obtain the cluster division results; wherein, the merging and splitting processing methods are specifically as follows: when the number of points in a cluster exceeds the size threshold of the maximum cluster, it is split, and the new cluster center can be determined by re-clustering the points in the cluster; when the distance between two clusters is less than the minimum distance threshold, the two clusters are merged into one cluster.

[0187] The feature subunit is used to extract features from the basic statistics of the partitioning results to obtain a feature data set.

[0188] Among them, the Euclidean distance is:

[0189]

[0190] The new centroid in the new centroid set is:

[0191]

[0192] Where x is the data point, c j is the cluster center, n is the feature dimension, x i is the ith data point, is the cluster center corresponding to the i-th data point;

[0193] S j is the set of data points in the jth cluster, |S j | is the number of data points in the jth cluster.

[0194] The centroid subunit, allocation subunit, calculation subunit, result subunit and feature subunit of the embodiment of the present application are based on the ISODATA algorithm, and a plurality of initial centroids are randomly selected and iterated based on these centroids, which can increase the possibility of finding the global optimal solution and avoid the algorithm from falling into the local optimal solution. By continuously iteratively calculating the cluster center and merging or splitting the clusters based on the new centroid set, the algorithm can automatically adjust the number of clusters according to the actual distribution of the data, and continuously optimize the cluster division results, so that the data points in each cluster are closer and the distance between clusters is more distant, which helps to improve the quality and accuracy of clustering.

[0195] The model building unit is used to initialize the preset initial LightGBM model (Light Gradient Boosting Machine); wherein the initial prediction value of the initial LightGBM model is the mean of the target variable;

[0196] The model building unit is also used to calculate the residual of the mth iteration through the initial LightGBM model according to the historical meteorological load samples to obtain the current residual;

[0197] Model building unit, also used to build a decision tree f based on the current residual m , so that the decision tree f m Can minimize residual error;

[0198] Model building unit, also used in decision tree f m While minimizing the residual error, the predicted value of the initial LightGBM model is updated according to the learning rate η to obtain the LightGBM model.

[0199] The actual prediction unit is used to perform load prediction on the feature data set through the LightGBM model to obtain the load prediction value.

[0200] The initial prediction value is:

[0201]

[0202] The residual of the mth iteration is:

[0203]

[0204] Update the prediction value of the initial LightGBM model, which can be expressed as:

[0205]

[0206] The prediction formula of the LightGBM model is:

[0207]

[0208] in, is the initial prediction value, N is the number of samples, y i is the true value of the i-th sample;

[0209] is the predicted value corresponding to the i-th sample in the m-1th iteration, η is the learning rate, and f m is the decision tree at the mth iteration, x i is the i-th value of feature X;

[0210] is the final prediction value, and M is the total number of decision trees.

[0211] In each round of iteration of the model building unit and the actual prediction unit in the embodiment of the present application, the LightGBM model constructs a new decision tree based on the residual of the current model and gradually updates it through the learning rate. This gradient-based boosting strategy enables the model to continuously approach the true value, thereby improving the accuracy of the prediction; especially when processing complex meteorological load data, the LightGBM model can capture the nonlinear relationships and potential patterns in the data, thereby improving the accuracy of load forecasting.

[0212] Overall, the first prediction module 20 of the embodiment of the present application can effectively remove those features that have little or no impact on the prediction results by calculating the correlation between each feature and the load prediction result and filtering out the data with a correlation higher than the preset standard value, thereby improving the quality of the data and the accuracy of the prediction. Clustering the filtered data set can further group similar features together to form a more representative feature data set, thereby reducing the dimension of the data, and can better understand the intrinsic relationship between the data based on the clustering results, which helps to improve the interpretability and prediction ability of the model;

[0213] In addition, the LightGBM model is a data-driven machine learning model that is widely used to process large-scale data sets and complex prediction tasks. The performance of the model often improves with the increase in the amount of data. A large amount of data can provide rich samples for the model, enabling it to better capture complex nonlinear relationships and feature interactions; therefore, using this model when you have a large amount of data can better achieve high-precision predictions.

[0214] In one embodiment, the second prediction module 30 includes a matrix unit, a site unit, a history unit, a physical unit and a load prediction unit; wherein the matrix unit, the site unit, the history unit and the physical unit are the process of establishing a physical model, and the load prediction unit is the process of performing load prediction on meteorological load data through a physical model to obtain a load prediction value, specifically:

[0215] The matrix unit is used to calculate the distance between the longitude and latitude points of two photovoltaic sites in the photovoltaic site set where the target photovoltaic site is located using the spherical distance formula (Haversine formula) to obtain the site distance; where the longitude and latitude of the photovoltaic sites are (φ i ,λ i ), where i is the site index;

[0216] The matrix unit is also used to traverse the photovoltaic sites in the photovoltaic site set to obtain a distance matrix D between all photovoltaic sites composed of several site distances;

[0217] A site unit, used to find the first photovoltaic site closest to the target photovoltaic site according to the distance matrix D;

[0218] A history unit, used to calculate a historical data calculation coefficient set according to the historical meteorological load data of the first photovoltaic site and the target photovoltaic site;

[0219] The physical unit is used to calculate the coefficient set based on historical data and establish a physical model based on temperature, humidity and wind speed at different times.

[0220] The site distance is:

[0221]

[0222] The site formula corresponding to the first photovoltaic site is:

[0223] j min =argminD kj

[0224] The calculation formula of the historical data calculation coefficient in the historical data calculation coefficient set is:

[0225] β=(X T X) -1 X T P

[0226] The physical model is:

[0227]

[0228] Among them, d ij is the distance between site i and site j, R is the average radius of the earth; φ i and φ i are the longitude and latitude of site i, φ j and λ j are the longitude and latitude of station j, respectively, in radians;

[0229] D kj is the site distance corresponding to the target site k;

[0230] X is the characteristic matrix, which contains constant terms and all characteristic variables; P is the target variable vector, which contains the historical power generation load, and T represents the sign of the transposed matrix.

[0231] is the photovoltaic power generation load at the tth moment, T t , H t , R t and W t are the temperature, humidity, radiation intensity and wind speed at the tth moment; β 1 , β 2, β 3 and β 4 are the historical data calculation coefficients corresponding to temperature, humidity, radiation intensity and wind speed, β 0 The parameter is the photovoltaic power generation load data at time t-1.

[0232] The matrix unit, site unit, history unit and physical unit of the embodiment of the present application calculate the distance matrix between photovoltaic sites and find the first photovoltaic site closest to the target photovoltaic site, thereby making full use of the principle of spatial proximity. By introducing the data of neighboring sites, data complementarity can be formed to reduce the dependence on a single data source. In addition, the establishment of the physical model takes into account meteorological factors such as temperature, humidity and wind speed at different times. By integrating these real-time meteorological data into the model, the model can dynamically adjust the prediction results to better adapt to actual meteorological conditions and load changes, thereby enhancing the adaptability and robustness of the model.

[0233] In addition, physical models rely on modeling of systems or existing theories and have less reliance on large-scale data. In the case of small data volumes, data-driven models such as the LightGBM model may be difficult to generalize correctly due to insufficient data, while physical models can make more accurate predictions.

[0234] The load forecasting unit is used to perform load forecasting on the meteorological load data through a physical model to obtain a load forecast value if the number of days of data of the meteorological load data is greater than or equal to 10 days and less than or equal to 100 days.

[0235] In one embodiment, the third prediction module 40 includes an interpolation unit, a capacity unit, and a variation unit;

[0236] The interpolation unit is used to clean the meteorological load data to remove obvious outliers and noise if the number of days of meteorological load data is less than 10 days; and to supplement the meteorological load data by linear interpolation to obtain complete data;

[0237] The capacity unit is used to calculate the load forecast value by fitting the theoretical basic power curve based on the complete data if the installed capacity of the system where the target PV site is located has not changed;

[0238] A change unit is used for adjusting the theoretical basic power curve to the new installed capacity level by scaling the theoretical basic power curve if the installed capacity of the system where the target photovoltaic site is located changes, thereby obtaining a final theoretical basic power curve;

[0239] The variation unit is also used to calculate the load forecast value based on the complete data and the final theoretical basic power curve.

[0240] Among them, the theoretical basic power curve is:

[0241] P=aI 2 +bI+c

[0242] The final theoretical basic power curve is:

[0243] P=S(aI 2 +bI+c)

[0244] For parameter S:

[0245]

[0246] Where I is the light intensity, a, b, and c are the least squares fitting parameters, and S is the scaling factor.

[0247] C old is the original installed capacity of the system where the target PV site is located, C new is the new installed capacity of the system where the target PV site is located.

[0248] To apply this embodiment, please refer to Figure 2 , Figure 2 It is an overall flow chart provided in an embodiment of the present application, which shows the general process of obtaining the load forecast value as a forecast result in the second embodiment of the present application.

[0249] The third prediction module 40 of the embodiment of the present application uses linear interpolation to supplement the missing meteorological load data, which can ensure the integrity and continuity of the data and avoid incomplete data sets that may cause inaccurate or biased prediction results. When the installed capacity of the system where the target photovoltaic site is located has not changed, directly fitting the theoretical basic power curve based on the complete data to calculate the load forecast value can be simple and direct and can better reflect the actual operating conditions of the system, thereby improving the accuracy of the prediction. When the installed capacity of the system changes, the theoretical basic power curve is scaled to adapt to the new capacity situation. This method allows the system to quickly respond to changes in system capacity and adjust the prediction results accordingly to ensure that the predicted value is consistent with the actual load situation;

[0250] Moreover, when data is missing, in order to achieve prediction results, it is easier to maintain performance stability by using a power curve model with a simpler structure.

[0251] Overall, the embodiments of the present application have the following beneficial effects:

[0252] When the amount of meteorological load data is large enough, the present invention uses the LightGBM model for load forecasting; the LightGBM model is a data-driven machine learning model, which is widely used to process large-scale data sets and complex forecasting tasks. The performance of the model often improves with the increase of data volume, so it can well capture complex nonlinear relationships and feature interactions, and achieve high-precision load forecasting. When the amount of data is not enough to support a complex machine learning model, a method based on a physical model is used instead. Since the historical meteorological load data of the target photovoltaic site may also have insufficient samples, the meteorological load data of adjacent photovoltaic sites is used to establish a physical model, which can fully utilize the principle of spatial proximity to form data complementarity, thereby better reflecting the load characteristics of the target site and reducing the requirements for the amount of data. In the case of missing data, the power curve is fitted for prediction. This method does not rely on a complete data set, but uses existing data trends and patterns to infer the value of the missing part, thereby improving the robustness of the prediction.

[0253] Embodiment three:

[0254] An embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the distributed photovoltaic load prediction method based on ISO and LightGBM;

[0255] Among them, the distributed photovoltaic load forecasting method based on ISO and LightGBM, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0256] The above are preferred embodiments of the present invention. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A distributed photovoltaic load forecasting method based on ISO and LightGBM, characterized in that: include: Obtain meteorological load data of the target PV site; If the amount of the meteorological load data is greater than a preset threshold range, the meteorological load data is subjected to load forecasting by using the LightGBM model to obtain a load forecast value; If the data volume of the meteorological load data is within the threshold range, load forecasting is performed on the meteorological load data through a physical model to obtain a load forecast value; wherein the physical model is established based on the meteorological load data of the neighboring photovoltaic sites of the target photovoltaic site; If the data volume of the meteorological load data is less than the threshold range, load prediction is performed on the meteorological load data by fitting a power curve to obtain a load prediction value.

2. A distributed photovoltaic load forecasting method based on ISO and LightGBM as claimed in claim 1, characterized in that: The LightGBM model is used to perform load forecasting on the meteorological load data to obtain the load forecast value, which is specifically: Calculating the correlation between each feature in the meteorological load data and the load forecast result to obtain a set of correlation coefficients; According to the correlation coefficient set, data with correlation higher than a preset standard value is screened out from the meteorological load data to obtain a data set; Performing clustering processing on the data set to obtain a feature data set; The load prediction is performed on the feature data set through the LightGBM model to obtain the load prediction value.

3. A distributed photovoltaic load forecasting method based on ISO and LightGBM as claimed in claim 2, characterized in that: The data set is clustered to obtain a feature data set, specifically: Randomly select a number of data points from the data set as initial centroids; By measuring the Euclidean distance between the data points and the cluster centers, each of the data points is assigned to the nearest cluster center among the cluster centers, thereby obtaining a first assignment result; Calculate the centroids of the plurality of cluster centers based on the first allocation result to obtain a new centroid set; According to the new centroid set, the corresponding clusters are merged or split until the cluster centers of the clusters no longer change or a preset number of iterations is reached, thereby obtaining a cluster division result; Feature extraction is performed on basic statistics of the division result to obtain the feature data set.

4. A distributed photovoltaic load forecasting method based on ISO and LightGBM as claimed in claim 1, characterized in that: The LightGBM model is specifically: According to the historical meteorological load samples, the residual of the mth iteration is calculated through the preset initial LightGBM model to obtain the current residual; Establishing a decision tree based on the current residual; In the case of maximizing the reduction of residual error through the decision tree, the initial LightGBM model is updated according to the learning rate to obtain the LightGBM model.

5. A distributed photovoltaic load forecasting method based on ISO and LightGBM as claimed in claim 1, characterized in that: The physical model is established based on the meteorological load data of the neighboring photovoltaic sites of the target photovoltaic site, specifically: In the photovoltaic site set where the target photovoltaic site is located, a distance matrix between all photovoltaic sites is calculated according to the longitude and latitude of each photovoltaic site; Finding a first photovoltaic site closest to the target photovoltaic site according to the distance matrix; Calculate a historical data calculation coefficient set according to the historical meteorological load data of the first photovoltaic site and the target photovoltaic site; A coefficient set is calculated based on the historical data, and the physical model is established according to the temperature, humidity and wind speed at different times.

6. A distributed photovoltaic load forecasting method based on ISO and LightGBM as claimed in claim 1, characterized in that: The load forecast is performed on the meteorological load data by fitting the power curve to obtain the load forecast value, which is specifically: The meteorological load data is supplemented by a linear interpolation method to obtain complete data; If the installed capacity of the system where the target photovoltaic site is located has not changed, then the load forecast value is calculated by fitting the theoretical basic power curve based on the complete data; If the installed capacity of the system where the target photovoltaic site is located changes, the theoretical basic power curve is scaled to obtain a final theoretical basic power curve; Based on the complete data, the load forecast value is calculated according to the final theoretical basic power curve.

7. A distributed photovoltaic load forecasting method based on ISO and LightGBM as claimed in claim 6, characterized in that: The final theoretical basic power curve is specifically: P=S(aI 2 +bI+c) Where I is the light intensity, a, b, and c are the least squares fitting parameters, and S is the scaling factor.

8. A distributed photovoltaic load forecasting device based on ISO and LightGBM, characterized in that: It includes a data module, a first prediction module, a second prediction module and a third prediction module; Wherein, the data module is used to obtain the meteorological load data of the target photovoltaic site; The first prediction module is used to perform load prediction on the meteorological load data through the LightGBM model to obtain a load prediction value if the data volume of the meteorological load data is greater than a preset threshold range; The second prediction module is used to perform load prediction on the meteorological load data through a physical model to obtain a load prediction value if the data volume of the meteorological load data is within the threshold range; wherein the physical model is established based on the meteorological load data of the neighboring photovoltaic sites of the target photovoltaic site; The third prediction module is used to perform load prediction on the meteorological load data by fitting a power curve to obtain a load prediction value if the data volume of the meteorological load data is less than the threshold range.

9. A distributed photovoltaic load prediction device based on ISO and LightGBM as claimed in claim 8, characterized in that: The first prediction module includes a coefficient unit, a data unit, a clustering unit and a prediction unit; The coefficient unit is used to calculate the correlation between each feature in the meteorological load data and the load forecast result to obtain a set of correlation coefficients; The data unit is used to filter out data with correlation higher than a preset standard value from the meteorological load data according to the correlation coefficient set to obtain a data set; The clustering unit is used to perform clustering processing on the data set to obtain a feature data set; The prediction unit is used to perform load prediction on the feature data set through the LightGBM model to obtain the load prediction value.

10. A distributed photovoltaic load prediction device based on ISO and LightGBM as claimed in claim 9, characterized in that: The clustering unit includes a centroid subunit, a distribution subunit, a calculation subunit, a result subunit and a feature subunit; Wherein, the centroid subunit is used to randomly select a number of data points from the data set as initial centroids; The allocation subunit is used to allocate each of the data points to the nearest cluster center among the cluster centers by measuring the Euclidean distance between the data points and the cluster centers, so as to obtain a first allocation result; The calculation subunit is used to calculate the centroids of the plurality of cluster centers based on the first allocation result to obtain a new centroid set; The result subunit is used to merge or split the corresponding clusters according to the new centroid set until the cluster centers of the clusters no longer change or a preset number of iterations is reached, thereby obtaining a cluster division result; The feature subunit is used to extract features from the basic statistics of the segmentation result to obtain the feature data set.

11. A distributed photovoltaic load forecasting device based on ISO and LightGBM as claimed in claim 8, characterized in that: The LightGBM model is specifically: According to the historical meteorological load samples, the residual of the mth iteration is calculated through the preset initial LightGBM model to obtain the current residual; Establishing a decision tree based on the current residual; In the case of maximizing the reduction of residual error through the decision tree, the initial LightGBM model is updated according to the learning rate to obtain the LightGBM model.

12. A distributed photovoltaic load forecasting device based on ISO and LightGBM as claimed in claim 8, characterized in that: The second prediction module includes a matrix unit, a site unit, a history unit and a physical unit; The matrix unit is used to calculate the distance matrix between all photovoltaic sites in the photovoltaic site set where the target photovoltaic site is located according to the longitude and latitude of each photovoltaic site; The site unit is used to find a first photovoltaic site closest to the target photovoltaic site according to the distance matrix; The history unit is used to calculate a historical data calculation coefficient set according to the historical meteorological load data of the first photovoltaic site and the target photovoltaic site; The physical unit is used to calculate the coefficient set based on the historical data and establish the physical model according to the temperature, humidity and wind speed at different times.

13. A distributed photovoltaic load forecasting device based on ISO and LightGBM as claimed in claim 8, characterized in that: The third prediction module includes an interpolation unit, a capacity unit and a change unit; Wherein, the interpolation unit is used to supplement the meteorological load data by linear interpolation to obtain complete data; The capacity unit is used to calculate the load prediction value by fitting the theoretical basic power curve according to the complete data if the installed capacity of the system where the target photovoltaic site is located has not changed; The change unit is used to scale the theoretical basic power curve to obtain a final theoretical basic power curve if the installed capacity of the system where the target photovoltaic site is located changes; based on the complete data, the load prediction value is calculated according to the final theoretical basic power curve.

14. A distributed photovoltaic load prediction device based on ISO and LightGBM as claimed in claim 13, characterized in that: The final theoretical basic power curve is specifically: P=S(aI 2 +bI+c) Where I is the light intensity, a, b, and c are the least squares fitting parameters, and S is the scaling factor.

15. A storage medium, characterized in that: The storage medium stores a computer program, which is called and executed by a computer to implement a distributed photovoltaic load forecasting method based on ISO and LightGBM as described in any one of claims 1 to 7 above.

Images