Wind and light load joint prediction method and device, storage medium and computer equipment

Abstract

The invention discloses a wind-light-load joint prediction method and device, a storage medium and computer equipment, and the method comprises the steps: obtaining a plurality of data combination scenes obeying the probability distribution condition through employing an inverse transformation sampling method for the probability distribution condition of any power generation data of a distributed power grid in any historical month, determining typical data combination scenes with different data combination modes; performing four-dimensional Markov chain modeling based on the typical data combination scenes of the various power generation data, predicting a four-dimensional power generation data combination mode of each day in a month, and determining respective target typical data combination scenes of the various power generation data in the predicted four-dimensional power generation data combination mode of each day in the month, and obtaining a monthly four-dimensional power generation data combined prediction result, and further obtaining an annual four-dimensional power generation data combined prediction result. The coupling relation of wind power, photovoltaic power and net load can be effectively considered, and more accurate wind and light load combined prediction under the condition of a whole-year hour-by-hour time sequence is realized.

Description

Technical Field

[0001] The present application relates to the technical field of wind-solar load prediction, and in particular to a wind-solar load joint prediction method and device, storage medium, and computer equipment. Background Art

[0002] In recent years, with the increasing attention paid to environmental protection and sustainable development around the world, my country's new energy industry has ushered in unprecedented development opportunities. As an important form of new energy utilization, the status and role of distributed generation systems in the power system have become increasingly prominent. Distributed generation systems, with their flexibility, high efficiency and environmental protection, provide strong support for the stable operation of the power system and the optimization and adjustment of the energy structure.

[0003] However, with the large-scale access of new energy sources and the development of new power systems, the operating characteristics of power systems have changed significantly. Traditional power system load forecasting methods can no longer meet current needs. Summary of the invention

[0004] In view of this, the present application provides a wind-solar-load joint prediction method and device, storage medium, and computer equipment. For the probability distribution of any kind of power generation data in any historical month of the distributed power grid, the inverse transform sampling method is used to obtain a variety of data combination scenarios that obey the probability distribution, and determine the typical data combination scenarios with different data combination methods; based on the typical data combination scenarios of various power generation data, four-dimensional Markov chain modeling is performed to predict the four-dimensional power generation data combination method of each day in the month, and in the predicted four-dimensional power generation data combination method of each day in the month, the target typical data combination scenario of each power generation data in the month is determined to obtain the monthly four-dimensional power generation data joint prediction result, and then the annual four-dimensional power generation data joint prediction result is obtained. It can effectively consider the coupling relationship between wind power, photovoltaic power generation and net load, and realize more accurate wind-solar-load joint prediction under the hourly long time series throughout the year.

[0005] According to one aspect of the present application, a wind-solar-load joint prediction method is provided, which is applied to a distributed power grid; the method comprises:

[0006] Obtaining the power generation data of each historical month in the distributed power grid, wherein the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data and net load data, and each type of power generation data includes multiple power generation data values ​​with a time resolution of hourly level;

[0007] For any kind of power generation data in any historical month, obtain the probability distribution of each power generation data value in the power generation data of this kind, use the inverse transform sampling method to obtain multiple data combination scenarios that conform to the probability distribution, and among the multiple data combination scenarios, obtain typical data combination scenarios with different data combination methods. Among them, in each typical data combination scenario, the time resolution of the typical power generation data value is daily;

[0008] Based on the typical data combination scenarios corresponding to various power generation data in the month, perform four-dimensional Markov chain modeling to obtain the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix of the month. Based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix, predict the four-dimensional power generation data combination methods for each day in the month. Among them, the four-dimensional power generation data combination method is combined by the daily predicted data values of various power generation data;

[0009] Among the four-dimensional power generation data combination methods predicted for each day in the month, respectively determine the predicted data combination methods of various power generation data on a monthly basis based on the daily predicted data values of various power generation data, and determine the target typical data combination scenarios that match the predicted data combination methods of various power generation data, to obtain the monthly four-dimensional power generation data joint prediction result of the month. Based on the monthly four-dimensional power generation data joint prediction results of each month, obtain the annual four-dimensional power generation data joint prediction result.

[0010] Optionally, the predicting the four-dimensional power generation data combination methods for each day in the month based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix includes:

[0011] In the four-dimensional power generation data combination probability matrix, sample the four-dimensional power generation data combination method on the initial day. Based on the four-dimensional power generation data combination probability transition matrix, starting from the four-dimensional power generation data combination method on the initial day, sequentially predict the four-dimensional power generation data combination methods transferred out on the next day until the four-dimensional power generation data combination methods for the total number of days in the month are obtained, to obtain the four-dimensional power generation data combination methods for each day in the month.

[0012] Optionally, the obtaining the probability distribution of each power generation data value in the power generation data of this kind includes:

[0013] Based on the kernel density estimation method, obtain the probability distribution of each power generation data value in the power generation data of this kind.

[0014] Optionally, the using the inverse transform sampling method to obtain multiple data combination scenarios that conform to the probability distribution includes:

[0015] Based on the cumulative distribution function calculation formula, determine the cumulative distribution function corresponding to the probability distribution, and use the inverse transform sampling method and the cumulative distribution function to obtain multiple data combination scenarios that follow the probability distribution, where the cumulative distribution function is:

[0016]

[0017] F e (x) is the cumulative distribution function, is the probability distribution, X h are the probability values in the probability distribution.

[0018] Optionally, the step of using the inverse transform sampling method and the cumulative distribution function to obtain multiple data combination scenarios that follow the probability distribution includes:

[0019] Substitute the random numbers that follow the standard normal distribution and the inverse function of the cumulative distribution function into the inverse transform sampling formula to obtain multiple data combination scenarios that follow the probability distribution, where the inverse transform sampling formula is:

[0020]

[0021] P t is the inverse transform sampling result, t is time, F e is the cumulative distribution function, is the inverse function of the cumulative distribution function, Z t are the random numbers that follow the standard normal distribution, and the standard normal distribution function values φ(Z t ) are uniformly distributed within a preset numerical interval.

[0022] Optionally, the step of obtaining typical data combination scenarios with different data combination methods among multiple data combination scenarios includes:

[0023] Among multiple data combination scenarios, based on the scenario set retention and deletion metric formula, determine one deleted data combination scenario each time, calculate the scenario closest to the deleted data combination scenario based on the scenario distance similarity calculation formula, and delete the calculated scenario until the number of remaining typical data combination scenarios meets the preset number, where the scenario set retention and deletion metric formula is:

[0024]

[0025] The scenario distance similarity calculation formula is:

[0026]

[0027] D k(C i , C' i ) is the set C of typical data combination scenarios i and the set C' of data combination scenarios to be deleted i The metric relationship between them, N i is the total number of typical data combination scenarios, N' i is the total number of data combination scenarios to be deleted, v is the index of the typical data combination scenario, and in the set C of typical data combination scenarios i the initial probability of each typical data combination scenario is 1 / N i In the set C' of data combination scenarios to be deleted i the initial probability of each data combination scenario to be deleted is 1 / N' i , s' v is the data combination scenario to be deleted, s v is the scenario closest to the data combination scenario s' to be deleted v .

[0028] Optionally, before obtaining the probability distribution of each power generation data value in the power generation data of the type, the method further includes:

[0029] Obtain the first power generation data value at the first preset percentile and the second power generation data value at the second preset quantity percentile in the power generation data of the type, and determine the outlier range based on the first power generation data value and the second power generation data value;

[0030] Based on the determined outlier range, remove the power generation data outliers in the power generation data of the type, and fill the missing power generation data values in the power generation data of the type by linear interpolation.

[0031] According to another aspect of the present application, a wind-solar-load joint prediction device is provided, and the device includes:

[0032] A wind-solar-load data acquisition module, configured to acquire the power generation data of each historical month in a distributed power grid, where the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data, and each power generation data includes multiple power generation data values with a time resolution of hourly level;

[0033] The typical data combination scenario determination module is used to obtain the probability distribution of each power generation data value in any kind of power generation data for any historical month, and use the inverse transform sampling method to obtain multiple data combination scenarios that conform to the probability distribution. Among the multiple data combination scenarios, typical data combination scenarios with different data combination methods are obtained. Among the typical data combination scenarios, the time resolution of the typical power generation data value is daily level;

[0034] The wind-solar-load combined combination mode prediction module is used to perform four-dimensional Markov chain modeling based on the typical data combination scenarios corresponding to various power generation data under the month, obtain the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix of the month, and predict the four-dimensional power generation data combination mode of each day in the month based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix. Among them, the four-dimensional power generation data combination mode is combined by the daily prediction data values of various power generation data to obtain the monthly four-dimensional power generation data combination mode of the month;

[0035] The annual wind-solar-load prediction result generation module is used to determine the predicted data combination mode of each kind of power generation data on a monthly basis respectively based on the daily prediction data values of various power generation data in the four-dimensional power generation data combination modes predicted for each day in the month, and determine the target typical data combination scenario that matches the predicted data combination mode of each kind of power generation data, obtain the monthly four-dimensional power generation data joint prediction result of the month, and obtain the annual four-dimensional power generation data joint prediction result based on the monthly four-dimensional power generation data joint prediction results of each month.

[0036] Optionally, the wind-solar-load combined combination mode prediction module is further used for:

[0037] In the four-dimensional power generation data combination probability matrix, sample the four-dimensional power generation data combination mode of the initial day, and based on the four-dimensional power generation data combination probability transition matrix, starting from the four-dimensional power generation data combination mode of the initial day, predict the four-dimensional power generation data combination mode transferred to the next day in sequence until the four-dimensional power generation data combination mode of the total number of days in the month is obtained, and obtain the four-dimensional power generation data combination mode of each day in the month.

[0038] Optionally, the typical data combination scenario determination module is further used for:

[0039] Based on the kernel density estimation method, obtain the probability distribution of each power generation data value in the kind of power generation data.

[0040] Optionally, the typical data combination scenario determination module is further used for:

[0041] Based on the cumulative distribution function calculation formula, determine the cumulative distribution function corresponding to the probability distribution, and use the inverse transform sampling method and the cumulative distribution function to obtain multiple data combination scenarios that follow the probability distribution, where the cumulative distribution function is:

[0042]

[0043] F e (x) is the cumulative distribution function, is the probability distribution, X h are the probability values in the probability distribution.

[0044] Optionally, the typical data combination scenario determination module is further configured to:

[0045] Substitute the random number that follows the standard normal distribution and the inverse function corresponding to the cumulative distribution function into the inverse transform sampling formula to obtain multiple data combination scenarios that follow the probability distribution, where the inverse transform sampling formula is:

[0046]

[0047] P t is the inverse transform sampling result, t is time, F e is the cumulative distribution function, is the inverse function of the cumulative distribution function, Z t is the random number that follows the standard normal distribution, and the standard normal distribution function value φ(Z t ) is uniformly distributed within a preset numerical interval.

[0048] Optionally, the typical data combination scenario determination module is further configured to:

[0049] In multiple data combination scenarios, based on the scenario set retention and deletion metric formula, determine one deleted data combination scenario each time, and based on the scenario distance similarity calculation formula, calculate the scenario closest to the deleted data combination scenario and delete the calculated scenario until the number of remaining typical data combination scenarios meets the preset number, where the scenario set retention and deletion metric formula is:

[0050]

[0051] The scenario distance similarity calculation formula is:

[0052]

[0053] D k (C i , C′ i ) is the set C of typical data combination scenariosi The measurement relationship between the set C' of data combination scenarios to be deleted i is N i where N' is the total number of typical data combination scenarios i and N is the total number of data combination scenarios to be deleted. v is the index of a typical data combination scenario, and for the set C of typical data combination scenarios i the initial probability of each typical data combination scenario is 1 / N i For the set C' of data combination scenarios to be deleted i the initial probability of each data combination scenario to be deleted is 1 / N' i where s' v is a data combination scenario to be deleted, and s v is the scenario closest to the data combination scenario s' to be deleted v in distance.

[0054] Optionally, the device further includes: a preprocessing module for wind, light, and load data, configured to:

[0055] Obtain a first power generation data value at a first preset percentile and a second power generation data value at a second preset quantity percentile in the power generation data of the type, and determine an outlier range based on the first power generation data value and the second power generation data value;

[0056] Based on the determined outlier range, remove the power generation data outliers in the power generation data of the type, and fill the missing power generation data values in the power generation data of the type by linear interpolation.

[0057] According to another aspect of the present application, there is provided a storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned wind-light-load joint prediction method is implemented.

[0058] According to still another aspect of the present application, there is provided a computer device including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, and when the processor executes the program, the above-mentioned wind-light-load joint prediction method is implemented.

[0059] With the above technical solution, a method and device for combined prediction of wind, light and load, a storage medium, and a computer device provided by this application are directed to the probability distribution of any type of power generation data in any historical month of a distributed power grid. The inverse transform sampling method is used to obtain multiple data combination scenarios that conform to the probability distribution, and typical data combination scenarios with different data combination methods are determined; based on the typical data combination scenarios of various power generation data, a four-dimensional Markov chain is modeled to predict the four-dimensional power generation data combination methods for each day within a month. Among the predicted four-dimensional power generation data combination methods for each day within a month, the target typical data combination scenarios for various power generation data within the month are determined to obtain the combined prediction result of monthly four-dimensional power generation data, and then the combined prediction result of annual four-dimensional power generation data is obtained. It can effectively consider the coupling relationship between wind power, photovoltaic power and net load, and realize more accurate combined prediction of wind, light and load under the condition of annual hourly long time series.

[0060] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the specific implementation manners of this application are specifically given below. Brief Description of the Drawings

[0061] The drawings described herein are used to provide a further understanding of this application and constitute a part of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0062] Figure 1 A flowchart showing a method for combined prediction of wind, light and load provided by an embodiment of this application is shown;

[0063] Figure 2 A flowchart showing another method for combined prediction of wind, light and load provided by an embodiment of this application is shown;

[0064] Figure 3 A structural diagram showing a device for combined prediction of wind, light and load provided by an embodiment of this application is shown;

[0065] Figure 4 A structural diagram showing another device for combined prediction of wind, light and load provided by an embodiment of this application is shown. Detailed Description of the Embodiment

[0066] Hereinafter, this application will be described in detail with reference to the drawings and in combination with the embodiments. It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other.

[0067] In this embodiment, a method for combined prediction of wind, light and load is provided, as Figure 1As shown in the figure, it is applied to a distributed power grid; the method includes:

[0068] Step 101, obtain the power generation data of each historical month in the distributed power grid, where the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data, and each type of power generation data includes multiple power generation data values with an hourly time resolution.

[0069] Currently, the demand scope of load forecasting has expanded from the traditional typical daily curve to 8760 hours throughout the year, that is, long-term sequence load curve forecasting. The 8760-hour net load forecasting, which comprehensively considers the future annual load level and new energy output and forecasts the hourly net load of the power system within a year, has become one of the key tasks in the long-term forecasting of the power grid system. As the boundary condition for medium- and long-term time series simulation, the 8760-hour net load forecasting is of great significance for the optimal operation of the power system and the rational allocation of energy resources. It can provide a scientific basis for the optimal dispatching of power generation resources, effectively reduce the phenomena of wind and light abandonment, improve the utilization rate of new energy, and thus promote the clean, low-carbon, and efficient development of the power system.

[0070] However, in the power grid with extensive participation of distributed power sources, the spatio-temporal distribution characteristics of wind power, photovoltaic power, and net load are highly correlated, showing a complex coupling relationship. Traditional single-variable forecasting methods, such as independently forecasting only for load, wind power, or photovoltaic power, are difficult to accurately capture the dynamic characteristics of multi-source spatio-temporal coupling and cannot meet the requirements of joint forecasting of wind, light, and load.

[0071] In the above embodiments of the present application, based on the coupling state probability distribution of wind power, photovoltaic power, and net load, the state transition probability matrix of wind, light, and load is calculated through a Markov chain model to realize the forecasting of monthly scenarios. In addition, each of the twelve months is modeled separately, and the forecasting results are spliced in chronological order to obtain the annual 8760-hour forecasting result, thereby realizing the joint forecasting of wind, light, and load in the distributed power grid under long-term sequence. Specifically, obtain the centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and system load data of historical months in the distributed power grid, and preprocess all the data respectively. Then, calculate the net load data. The concept of net load is: net load = system load - centralized wind power - centralized photovoltaic power - distributed photovoltaic power. Then, divide the centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data by month, and respectively construct historical data sets from January to December, that is, obtain the power generation data of each historical month in the distributed power grid. In particular, data for three consecutive historical years can be obtained to improve the data forecasting accuracy.

[0072] Step 102: For any type of power generation data in any historical month, obtain the probability distribution of each power generation data value in the type of power generation data, use the inverse transform sampling method to obtain multiple data combination scenarios that follow the probability distribution, and among the multiple data combination scenarios, obtain typical data combination scenarios with different data combination methods. Among each typical data combination scenario, the time resolution of the typical power generation data value is daily level.

[0073] Next, the KDE (Kernel Density Estimation) method can be used to calculate the probability distribution of centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data respectively. Then, inverse transform sampling is performed on the calculated probability distribution of centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data respectively to obtain several scenarios of centralized wind power curves, several scenarios of centralized photovoltaic power curves, several scenarios of distributed photovoltaic power curves, and several scenarios of net load curves, that is, multiple data combination scenarios of various power generation data respectively. Through the synchronous back substitution elimination method, scenario reduction is performed on several scenarios (multiple data combination scenarios) of the same type of power generation data to obtain a set of typical data combination scenarios that can represent the original scenario set of power generation data to the greatest extent in terms of probability. In particular, each typical data value in each typical data combination scenario can also be encoded to obtain a combination identification sequence for each typical data combination scenario, so as to perform scenario "restoration" according to various predicted data combination methods in the future.

[0074] Regarding the determination of the combination identification sequence, specifically, for the centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data of the historical month obtained in Step 101, the typical scenario with the closest data curve distance for each day of each type of power generation data can be judged through the Euclidean distance. Then, the typical data combination scenario is stateified with historical power generation data. For example, finally, the historical daily state sequence of the net load data (the combination identification sequence of the typical data combination scenario corresponding to the net load data) is:

[0075] {a 1 ,a 3 ,a 3 ,a 2 ,……},

[0076] The historical daily state sequence of the distributed photovoltaic power data is:

[0077] {b 1 ,b 2 ,b 2 ,b 1 ,……},

[0078] The historical daily status sequence of centralized wind power data is as follows:

[0079] {c 3 , c 3 , c 4 , c 3 , ……},

[0080] The historical daily status sequence of centralized photovoltaic power data is as follows:

[0081] {d 2 , d 2 , d 3 , d 4 , ……}.

[0082] Step 103: Based on the typical data combination scenarios corresponding to various power generation data in the month, perform four-dimensional Markov chain modeling to obtain the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix of the month. Based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix, predict the four-dimensional power generation data combination modes of each day in the month, where the four-dimensional power generation data combination modes are combined by the daily prediction data values of various power generation data.

[0083] Next, through four-dimensional Markov chain modeling and the typical data combination scenarios corresponding to various power generation data in the month, obtain the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix. Based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix, predict the four-dimensional power generation data combination modes of each day of the aforementioned types of power generation data in the month. Therefore, through four-dimensional Markov chain modeling, it is possible to couple the four types of power generation data of "wind, light, and load".

[0084] Step 104: Among the four-dimensional power generation data combination modes of each day predicted in the month, respectively determine the predicted data combination modes of various power generation data on a monthly basis based on the daily prediction data values of various power generation data, and determine the target typical data combination scenarios matching the predicted data combination modes of various power generation data to obtain the monthly four-dimensional power generation data joint prediction result of the month. Based on the monthly four-dimensional power generation data joint prediction results of each month, obtain the annual four-dimensional power generation data joint prediction result.

[0085] Finally, among the combined ways of four-dimensional power generation data for each day predicted in a month, based on the daily prediction data values of various power generation data respectively, determine the combined ways of the predicted data of various power generation data on a monthly basis, and determine the target typical data combination scenarios matching the combined ways of the predicted data of various power generation data respectively, so as to obtain the combined prediction result of monthly four-dimensional power generation data. Specifically, according to the data combination way of each typical data combination scenario, find the corresponding combined identification sequence, and then restore the target typical data combination scenario of any kind of power generation data in the month. Combining the target typical data combination scenarios of the four kinds of power generation data respectively, the combined prediction result of monthly four-dimensional power generation data is obtained. Based on the combined prediction results of monthly four-dimensional power generation data for each month, the combined prediction result of annual four-dimensional power generation data is obtained. For this reason, the prediction results for 12 months are obtained, spliced in chronological order, and finally the combined prediction result of annual 8760-hour wind-solar load is obtained.

[0086] By applying the technical solution of this embodiment, the joint probability distribution is calculated by the KDE method, and the inverse transform sampling is adopted, so that multiple possible scenarios can be generated, thus more comprehensively depicting the uncertainty of the wind-solar-load system, and being more able to take into account the randomness and uncertainty of the system compared with the traditional point estimation. At the same time, through the state processing of historical data and the combination of four-dimensional Markov chain modeling, the coupling relationship between wind power, photovoltaic power and load can be taken into account. Through the calculation of net load and the modeling of probability distribution, the mutual influence and change characteristics between wind-solar load are reflected, and through the separate processing of the data sets for 12 months of the whole year, a comprehensive and accurate combined prediction of wind-solar load for the annual 8760-hour power demand can be made. It is especially suitable for systems with large fluctuations in wind-solar resources and complex changes in load demand.

[0087] Further, as a refinement and extension of the specific implementation manner of the above embodiment, in order to completely illustrate the specific implementation process of this embodiment, another wind-solar-load combined prediction method is provided, as Figure 2 shown, this method includes:

[0088] Step 201, obtain the power generation data of each historical month in the distributed power grid. For any kind of power generation data in any historical month, obtain the first power generation data value located at the first preset percentile and the second power generation data value located at the second preset quantity percentile in the kind of power generation data, and determine the outlier range based on the first power generation data value and the second power generation data value.

[0089] Step 202, based on the determined outlier range, remove the power generation data outliers in the kind of power generation data, and fill the missing values in the kind of power generation data by the linear interpolation method.

[0090] In the above embodiments of the present application, in the distributed power grid, the power generation data of each historical month is obtained. Then, data preprocessing can be performed on the power generation data. The data preprocessing can include null data detection, out-of-limit data detection, error data detection, duplicate data detection, data filling or deletion. For example, for any kind of power generation data in any historical month, the 25th percentile and the 75th percentile in the dataset of the aforementioned type of power generation data are calculated. The outlier range is set to twice the difference between the 75th percentile and the 25th percentile. The data points in the dataset that are lower than the lower quartile minus the outlier range or higher than the upper quartile plus the outlier range are regarded as outliers, and then these outliers are removed from the dataset. Then, the missing values are filled by the linear interpolation method, which can improve the accuracy of subsequent data prediction.

[0091] Step 203: Based on the kernel density estimation method, obtain the probability distribution of each power generation data value in the type of power generation data, where the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data, and each type of power generation data includes multiple power generation data values with a time resolution of hourly level.

[0092] Next, the Kernel Density Estimation method is a common non-parametric method in statistics for estimating the probability density function of a random variable. This method applies kernel smoothing to probability density estimation and estimates the probability density function by placing kernel functions on each data point and summing them. The specific calculation formula of the kernel density estimation method is as follows:

[0093]

[0094] represents the estimated probability distribution, that is, the probability distribution of each power generation data value in the power generation data. n is the number of samples, that is, the total number of each power generation data value in the power generation data. x represents the input data, that is, the input power generation data value. h is the window width parameter of KDE, and K(u) is the kernel function. The selection of the window width h and the kernel function K(u) determines the performance of KDE.

[0095] Step 204: Based on the cumulative distribution function calculation formula, determine the cumulative distribution function corresponding to the probability distribution, and substitute the random number obeying the standard normal distribution and the inverse function corresponding to the cumulative distribution function into the inverse transform sampling formula to obtain multiple data combination scenarios obeying the probability distribution, where the cumulative distribution function is:

[0096]

[0097] F e(x) is the cumulative distribution function, is the probability distribution, X h are the probability values in the probability distribution;

[0098] The inverse transform sampling formula is:

[0099]

[0100] P t is the inverse transform sampling result, t is time, F e is the cumulative distribution function, is the inverse function of the cumulative distribution function, Z t is a random number following the standard normal distribution, and the standard normal distribution function values φ(Z t ) are uniformly distributed within a preset numerical interval.

[0101] Next, inverse transform sampling is a commonly used probability distribution sampling method that can generate random samples conforming to a given probability distribution according to the given probability distribution function. Its basic idea is to achieve sampling through the inverse function of the cumulative distribution function (CDF). Specifically, in the inverse transform sampling formula: P t is the possible situation at time t; is the inverse function of the cumulative distribution function F e ; Z t is a random number following the standard normal distribution, and the standard normal distribution function values φ(Z t ) follow the uniform distribution between [0, 1]. When sampling a certain random variable (i.e., any type of power generation data), a large number of random numbers following the [0, 1] distribution can be generated, and these random numbers correspond to the vertical axis values of the cumulative probability distribution curve of the random variable. Finding the inverse function values corresponding to the above values is the horizontal axis corresponding power value (or load value). For this reason, instead of directly operating on the empirical distribution of the random variable, a random scenario conforming to a certain distribution can be simply generated through the above method.

[0102] Step 205, in multiple data combination scenarios, based on the scenario set retention and deletion metric formula, each time a deleted data combination scenario is determined, based on the scenario distance similarity calculation formula, calculate the scenario closest to the deleted data combination scenario, and delete the calculated scenario until the number of remaining typical data combination scenarios meets the preset number, where the scenario set retention and deletion metric formula is:

[0103]

[0104] The scenario distance similarity calculation formula is:

[0105]

[0106] D k (C i ,C′ i ) is the set C of typical data combination scenarios i and the measurement relationship with the set C′ of deleted data combination scenarios, N i is the total number of typical data combination scenarios, N′ i is the total number of deleted data combination scenarios, v is the index of the typical data combination scenario, and the set C of typical data combination scenarios i the initial probability of each typical data combination scenario i is 1 / N In the set C′ of deleted data combination scenarios i the initial probability of each deleted data combination scenario i is 1 / N′ s′ i is the deleted data combination scenario, s v is the scenario closest to the deleted data combination scenario s′ v v i Distance to the nearest scenario

[0107] Next, through the synchronous back substitution elimination method, a large number of photovoltaic, wind power, and net load scenarios can be reduced to a small number of scenarios that can represent the initial scenarios. Specifically:

[0108] Use C i to represent N i reserved scenarios, that is, the set of typical data combination scenarios, C′ i to represent N′ i the set of deleted data combination scenarios. Determine the initial probability of the typical data combination scenarios in the set C i is 1 / N In the set C′ of deleted data combination scenarios i the initial probability of the deleted data combination scenarios i is 1 / N′ i v .

[0109] Each time, a scenario s′ that meets the conditions is removed v and placed in the set C′ of deleted data combination scenarios i That is;

[0110]

[0111] Change the number N of typical data combination scenarios i = N i -1,

[0112] Change the number N′ of deleted data combination scenariosi = N′ i + 1, and by using the scene distance similarity calculation formula, select the scene s v that is the closest to the scene s′ v .

[0113]

[0114] Change the probability of the scene s v that is the closest to the deleted scene s′ v to ensure that the sum of the probabilities of all scenes in the set C of reserved scenes (i.e., typical data combination scenes) is 1, and then update the probabilities of each scene in the set of deleted data combination scenes to 1 / N′ . Repeat the iteration until the number of scenes in the set C of typical data combination scenes i meets the set quantity i requirement i .

[0115] Step 206: Based on the typical data combination scenes corresponding to various power generation data in the month, perform four-dimensional Markov chain modeling to obtain the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix for the month. Among them, in each typical data combination scene, the time resolution of the typical power generation data value is daily level.

[0116] Step 207: In the four-dimensional power generation data combination probability matrix, sample the four-dimensional power generation data combination mode of the initial day. Based on the four-dimensional power generation data combination probability transition matrix, starting from the four-dimensional power generation data combination mode of the initial day, sequentially predict the four-dimensional power generation data combination mode transferred out on the next day until the four-dimensional power generation data combination modes for the total number of days in the month are obtained, and obtain the four-dimensional power generation data combination modes for each day in the month. Among them, the four-dimensional power generation data combination mode is combined by the daily prediction data values of various power generation data.

[0117] Step 208: In the four-dimensional power generation data combination modes predicted for each day in the month, respectively determine the predicted data combination modes of various power generation data on a monthly basis based on the daily prediction data values of various power generation data, and determine the target typical data combination scenes that match the predicted data combination modes of various power generation data, to obtain the monthly four-dimensional power generation data joint prediction result for the month. Based on the monthly four-dimensional power generation data joint prediction results of each month, obtain the annual four-dimensional power generation data joint prediction result.

[0118] Next, the specific steps of the four-dimensional Markov chain modeling are as follows:

[0119] ​In the typical data combination scenarios of payload data, after arranging each typical data value at a daily time resolution, for example:

[0120] {a 1 ,a 3 ,a 3 ,a 2 ,……},

[0121] Correspondingly, the typical data combination scenario of distributed photovoltaic power data is:

[0122] {b 1 ,b 2 ,b 2 ,b 1 ,……},

[0123] The typical data combination scenario of centralized wind power data is:

[0124] {c 3 ,c 3 ,c 4 ,c 3 ,……},

[0125] The typical data combination scenario of centralized photovoltaic power data is:

[0126] {d 2 ,d 2 ,d 3 ,d 4 ,……},

[0127] Coupling the four types of power generation data to obtain the wind-solar-load scenario state sequence (i.e., the four-dimensional power generation data combination method), which is expressed as:

[0128] {{a 1 ,b 1 ,c 3 ,d 2},{a 3 ,b 2 ,c 3 ,d 2},

[0129] {a 3 ,b 2 ,c 4 ,d 3},{a 2 ,b 1 ,c 3 ,d 4},……},

[0130] Calculate the occurrence probability of each wind-solar load scenario state (four-dimensional power generation data combination method), that is, the wind-solar load coupling state probability, and obtain the four-dimensional power generation data combination probability matrix. The calculation formula is as follows:

[0131]

[0132] In the formula, P(a = i, b = j, c = k, d = l) is the probability of different scenario states (four-dimensional power generation data combination method), and N(a i , b j , c k , d l ) is the number of occurrences of the (a i , b j , c k , d l ) wind-solar load scenario (four-dimensional power generation data combination method), and N is the total number of wind-solar load scenario states (four-dimensional power generation data combination method).

[0133] Statistically analyze the wind-solar load scenario state of each day and the wind-solar load scenario state of the next day, and calculate the wind-solar load state transition probability matrix (four-dimensional power generation data combination probability transition matrix). The calculation formula is as follows:

[0134]

[0135] In the formula, P{(a = i, b = j, c = k, d = l) → (a = z, b = x, c = n, d = m)} is the probability of different scenario state transitions, and N{(a = i, b = j, c = k, d = l) → (a = z, b = x, c = n, d = m)} is:

[0136] The number of times the (a i , b j , c k , d l ) wind-solar load scenario transfers to the (a z , b x , c n , d m ) wind-solar load scenario on the next day, and N is the total number of wind-solar load scenario transfer situations.

[0137] By performing inverse transformation sampling based on the coupled state probability of wind-solar load, the predicted initial coupled state of wind-solar load (the combined mode of four-dimensional power generation data for the initial day) is obtained through sampling. In particular, it is also possible to make a judgment based on the coupled state of wind-solar load in the same period of previous years for the predicted date. If it is similar to the coupled state in the same period of previous years, it is set as the initial coupled state of wind-solar load. If there is a large difference from the coupled state in the same period of previous years, resampling is performed. Find the wind-solar load state transition probability corresponding to the initial coupled state of wind-solar load in the wind-solar load state transition probability matrix, continue inverse transformation sampling to obtain the predicted coupled state of wind-solar load for the next day, and cycle in turn to complete the joint prediction of the coupled state of wind-solar load for thirty days (or thirty-one days, or twenty-eight days or twenty-nine days). Next, perform scenario restoration according to the combination identification sequence in the typical data combination scenario to obtain the joint prediction result of monthly four-dimensional power generation data. When performing the above inverse transformation sampling, the specific steps are as follows:

[0138] Initialize the cumulative distribution function CDF i = 0, and calculate the cumulative probability of each scenario in turn:

[0139]

[0140] Finally, an increasing array is obtained: CDF = {CDF 1 , CDF 2 , …, CDF n}.

[0141] Generate a uniform random number u ~ U(0, 1) in the interval [0, 1].

[0142] Find the smallest i in the cumulative distribution array that satisfies CDF i ≥ u. The corresponding scenario S i is the sampling result.

[0143] By applying the technical solution of this embodiment, the joint probability distribution is calculated by the KDE method, and inverse transformation sampling is used to generate multiple possible scenarios, thus more comprehensively characterizing the uncertainty of the wind-solar-load system. This method can take into account the randomness and uncertainty of the system better than the traditional point estimation. Through the state processing of historical data and the combination of four-dimensional Markov chain modeling, this method can take into account the coupling relationship between wind power, photovoltaic power and load. Through the calculation of net load and the modeling of joint probability distribution, it can more accurately reflect the mutual influence and change characteristics between wind-solar-load. By processing the data sets for 12 months of the whole year respectively, this method can make a comprehensive and accurate joint prediction of wind-solar load for 8760 hours of electricity demand throughout the year. It is especially suitable for systems with large fluctuations in wind-solar resources and complex changes in load demand.

[0144] Further, as Figure 1For the specific implementation of the method, an embodiment of the present application provides a combined prediction device for wind, light, and load, as Figure 3 shown. The device includes:

[0145] A wind-light-load data acquisition module 301, configured to acquire the power generation data of each historical month in a distributed power grid, where the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data, and each type of power generation data includes multiple power generation data values with a time resolution of hourly level;

[0146] A typical data combination scenario determination module 302, configured to, for any type of power generation data in any historical month, obtain the probability distribution of each power generation data value in the type of power generation data, use the inverse transform sampling method to obtain multiple data combination scenarios that conform to the probability distribution, and obtain typical data combination scenarios with different data combination methods among the multiple data combination scenarios, where in each typical data combination scenario, the time resolution of the typical power generation data value is daily level;

[0147] A wind-light-load combined combination method prediction module 303, configured to perform a four-dimensional Markov chain modeling based on the typical data combination scenarios corresponding to each type of power generation data in the month, obtain a four-dimensional power generation data combination probability matrix and a four-dimensional power generation data combination probability transition matrix of the month, and predict the four-dimensional power generation data combination method of each day in the month based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix, where the four-dimensional power generation data combination method is combined by the daily prediction data values of each type of power generation data to obtain the monthly four-dimensional power generation data combination method of the month;

[0148] An annual wind-light-load prediction result generation module 304, configured to, in the four-dimensional power generation data combination methods of each day predicted in the month, respectively determine the predicted data combination method of each type of power generation data on a monthly basis based on the daily prediction data values of each type of power generation data, and determine the target typical data combination scenario that matches the predicted data combination method of each type of power generation data, obtain the monthly four-dimensional power generation data combined prediction result of the month, and obtain the annual four-dimensional power generation data combined prediction result based on the monthly four-dimensional power generation data combined prediction results of each month.

[0149] Optionally, the wind-light-load combined combination method prediction module 303 is further configured to:

[0150] In the four-dimensional power generation data combination probability matrix, sample the four-dimensional power generation data combination method on the initial day of sampling. Based on the four-dimensional power generation data combination probability transition matrix, starting from the four-dimensional power generation data combination method on the initial day, sequentially predict the four-dimensional power generation data combination method transferred out on the next day until the four-dimensional power generation data combination methods for the total number of days in the month are obtained, and obtain the four-dimensional power generation data combination methods for each day in the month.

[0151] Optionally, the typical data combination scenario determination module 302 is further configured to:

[0152] Based on the kernel density estimation method, obtain the probability distribution of each power generation data value in the power generation data of the type.

[0153] Optionally, the typical data combination scenario determination module 302 is further configured to:

[0154] Based on the cumulative distribution function calculation formula, determine the cumulative distribution function corresponding to the probability distribution, and use the inverse transform sampling method and the cumulative distribution function to obtain multiple data combination scenarios that follow the probability distribution, where the cumulative distribution function is:

[0155]

[0156] F e (x) is the cumulative distribution function, is the probability distribution, X h are the probability values in the probability distribution.

[0157] Optionally, the typical data combination scenario determination module 302 is further configured to:

[0158] Substitute the random number that follows the standard normal distribution and the inverse function of the cumulative distribution function into the inverse transform sampling formula to obtain multiple data combination scenarios that follow the probability distribution, where the inverse transform sampling formula is:

[0159]

[0160] P t is the inverse transform sampling result, t is time, F e is the cumulative distribution function, is the inverse function of the cumulative distribution function, Z t is the random number that follows the standard normal distribution, and the standard normal distribution function value φ(Z t ) is uniformly distributed within a preset numerical interval.

[0161] Optionally, the typical data combination scenario determination module 302 is further configured to:

[0162] In multiple data combination scenarios, based on the scenario set retention and deletion metric formula, each time a data combination scenario to be deleted is determined. Based on the scenario distance similarity calculation formula, the scenario closest to the data combination scenario to be deleted is calculated, and the calculated scenario is deleted until the number of remaining typical data combination scenarios meets the preset number. Among them, the scenario set retention and deletion metric formula is:

[0163]

[0164] The scenario distance similarity calculation formula is:

[0165]

[0166] D k (C i , C′ i ) is the metric relationship between the typical data combination scenario set C i and the data combination scenario set C′ to be deleted i , N i is the total number of typical data combination scenarios, N′ i is the total number of data combination scenarios to be deleted, v is the index of the typical data combination scenario, and in the typical data combination scenario set C i , the initial probability of each typical data combination scenario is 1 / N i , and in the data combination scenario set C′ to be deleted i , the initial probability of each data combination scenario to be deleted is 1 / N′ i , s′ v is the data combination scenario to be deleted, and s v is the scenario closest to the data combination scenario s′ to be deleted v .

[0167] Furthermore, the embodiment of the present application provides another wind-solar-load joint prediction device, as shown in Figure 4 . This device includes:

[0168] A wind-solar-load data acquisition module 401, which is used to acquire the historical monthly power generation data in a distributed power grid. Among them, the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data, and net load data, and each type of power generation data includes multiple power generation data values with a time resolution of hourly level;

[0169] The typical data combination scenario determination module 402 is configured to obtain the probability distribution of each power generation data value in any type of power generation data for any historical month, use the inverse transform sampling method to obtain multiple data combination scenarios that conform to the probability distribution, and obtain typical data combination scenarios with different data combination methods from the multiple data combination scenarios. Among them, in each typical data combination scenario, the time resolution of the typical power generation data value is daily;

[0170] The wind-solar-load combined combination method prediction module 403 is configured to perform a four-dimensional Markov chain modeling based on the typical data combination scenarios corresponding to various power generation data under the month, obtain the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix of the month, and predict the four-dimensional power generation data combination method for each day in the month based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transition matrix. Among them, the four-dimensional power generation data combination method is combined by the daily predicted data values of various power generation data to obtain the monthly four-dimensional power generation data combination method of the month;

[0171] The annual wind-solar-load prediction result generation module 404 is configured to, in the four-dimensional power generation data combination methods for each day predicted in the month, respectively determine the predicted data combination methods of various power generation data on a monthly basis based on the daily predicted data values of various power generation data, and determine the target typical data combination scenarios that match the predicted data combination methods of various power generation data, obtain the monthly four-dimensional power generation data joint prediction result of the month, and obtain the annual four-dimensional power generation data joint prediction result based on the monthly four-dimensional power generation data joint prediction results of each month;

[0172] The wind-solar-load data preprocessing module 405 is configured to: obtain the first power generation data value at the first preset percentile and the second power generation data value at the second preset quantity percentile in the type of power generation data, and determine the outlier range based on the first power generation data value and the second power generation data value; remove the power generation data outliers in the type of power generation data based on the determined outlier range, and fill the power generation data missing values in the type of power generation data by linear interpolation method.

[0173] Optionally, the wind-solar-load combined combination method prediction module 403 is further configured to:

[0174] In the four-dimensional power generation data combination probability matrix, sample the four-dimensional power generation data combination method of the initial day, and based on the four-dimensional power generation data combination probability transition matrix, starting from the four-dimensional power generation data combination method of the initial day, sequentially predict the four-dimensional power generation data combination method transferred out on the next day until the four-dimensional power generation data combination methods for the total number of days in the month are obtained, and obtain the four-dimensional power generation data combination methods for each day in the month.

[0175] Optionally, the typical data combination scenario determination module 402 is further configured to:

[0176] Based on the kernel density estimation method, obtain the probability distribution of each power generation data value in the power generation data of each type.

[0177] Optionally, the typical data combination scenario determination module 402 is further configured to:

[0178] Based on the cumulative distribution function calculation formula, determine the cumulative distribution function corresponding to the probability distribution, and use the inverse transform sampling method and the cumulative distribution function to obtain multiple data combination scenarios that follow the probability distribution, where the cumulative distribution function is:

[0179]

[0180] F e (x) is the cumulative distribution function, is the probability distribution, X h are the probability values in the probability distribution.

[0181] Optionally, the typical data combination scenario determination module 402 is further configured to:

[0182] Substitute a random number that follows the standard normal distribution and the inverse function of the cumulative distribution function into the inverse transform sampling formula to obtain multiple data combination scenarios that follow the probability distribution, where the inverse transform sampling formula is:

[0183]

[0184] P t is the inverse transform sampling result, t is time, F e is the cumulative distribution function, is the inverse function of the cumulative distribution function, Z t is a random number that follows the standard normal distribution, and the standard normal distribution function value φ(Z t ) is uniformly distributed within a preset numerical interval.

[0185] Optionally, the typical data combination scenario determination module 402 is further configured to:

[0186] Among multiple data combination scenarios, based on the scenario set retention and deletion metric formula, determine one deleted data combination scenario each time, calculate the scenario that is closest to the deleted data combination scenario based on the scenario distance similarity calculation formula, and delete the calculated scenario until the number of remaining typical data combination scenarios meets the preset number, where the scenario set retention and deletion metric formula is:

[0187]

[0188] The formula for calculating the scene distance similarity is as follows:

[0189]

[0190] D k (C i , C' i ) is the metric relationship between the set C of typical data combination scenarios i and the set C' of deleted data combination scenarios i , N i is the total number of typical data combination scenarios, N' i is the total number of deleted data combination scenarios, v is the index of the typical data combination scenario, and in the set C of typical data combination scenarios i , the initial probability of each typical data combination scenario is 1 / N i , and in the set C' of deleted data combination scenarios i , the initial probability of each deleted data combination scenario is 1 / N' i , s' v is the deleted data combination scenario, s v is the scenario closest to the deleted data combination scenario s' v .

[0191] It should be noted that for other corresponding descriptions of each functional unit involved in the wind-solar-load joint prediction device provided in the embodiments of the present application, reference can be made to Figures 1 to 2 the corresponding descriptions in the method, which will not be elaborated here.

[0192] Based on the method as described above Figures 1 to 2 shown, correspondingly, the embodiments of the present application also provide a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the wind-solar-load joint prediction method as described above Figures 1 to 2 shown is implemented.

[0193] Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various implementation scenarios of the present application.

[0194] Based on the method as described above Figures 1 to 2 shown, and Figure 3 、 Figure 4In the virtual device embodiment shown, to achieve the above object, an embodiment of the present application further provides a computer device, which may specifically be a personal computer, a server, a network device, etc. The computer device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the Figures 1 to 2 wind-solar-load combined prediction method shown above.

[0195] Optionally, the computer device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, sensors, an audio circuit, a WI-FI module, etc. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc. Optionally, the user interface may further include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.

[0196] Those skilled in the art can understand that the structure of a computer device provided in this embodiment does not limit the computer device, and it may include more or fewer components, or combine some components, or have different component arrangements.

[0197] The storage medium may further include an operating system and a network communication module. The operating system is a program for managing and storing the hardware and software resources of the computer device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to implement communication between components inside the storage medium, and communication between the storage medium and other hardware and software in the entity device.

[0198] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, or can be implemented by hardware to obtain the probability distribution of any kind of power generation data for any historical month of a distributed power grid, and use the inverse transform sampling method to obtain multiple data combination scenarios that conform to the probability distribution, and determine typical data combination scenarios with different data combination methods; based on the typical data combination scenarios of various power generation data, perform a four-dimensional Markov chain modeling to predict the four-dimensional power generation data combination methods for each day within the predicted month. Among the four-dimensional power generation data combination methods for each day within the predicted month, determine the target typical data combination scenarios for various power generation data within the month to obtain the monthly four-dimensional power generation data joint prediction result, and further obtain the annual four-dimensional power generation data joint prediction result. It can effectively consider the coupling relationship between wind power, photovoltaic power and net load, and achieve a more accurate wind-solar-load joint prediction in the case of annual hourly long time series.

[0199] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the drawings are not necessarily essential for implementing the present application. Those skilled in the art can understand that the modules in the devices in the implementation scenario can be distributed in the devices in the implementation scenario according to the description of the implementation scenario, or can be correspondingly changed and located in one or more devices different from the present implementation scenario. The modules in the above implementation scenario can be combined into one module, or further split into multiple sub-modules.

[0200] The above serial numbers of the present application are only for description and do not represent the advantages or disadvantages of the implementation scenario. The above disclosure is only several specific implementation scenarios of the present application. However, the present application is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present application.

Claims

1. A wind-solar-load joint prediction method, characterized in that: Applied to a distributed power grid; the method comprises: Obtaining the power generation data of each historical month in the distributed power grid, wherein the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data and net load data, and each type of power generation data includes multiple power generation data values ​​with a time resolution of hourly level; For any type of power generation data in any historical month, obtain the probability distribution of each power generation data value in the type of power generation data, use the inverse transformation sampling method to obtain multiple data combination scenarios that obey the probability distribution, and obtain typical data combination scenarios with different data combination methods in the multiple data combination scenarios, wherein the time resolution of the typical power generation data value in each typical data combination scenario is at the day level; Based on the typical data combination scenarios corresponding to various power generation data in the month, four-dimensional Markov chain modeling is performed to obtain a four-dimensional power generation data combination probability matrix and a four-dimensional power generation data combination probability transfer matrix for the month, and based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transfer matrix, the four-dimensional power generation data combination mode of each day in the month is predicted, wherein the four-dimensional power generation data combination mode is combined by combining the daily-level prediction data values ​​of various power generation data; Among the four-dimensional power generation data combination methods predicted for each day of the month, the monthly forecast data combination methods of various power generation data are determined based on the daily forecast data values ​​of various power generation data, and the target typical data combination scenarios matching the forecast data combination methods of various power generation data are determined to obtain the monthly four-dimensional power generation data joint forecast result of the month, and based on the monthly four-dimensional power generation data joint forecast results of each month, the annual four-dimensional power generation data joint forecast result is obtained.

2. The method according to claim 1, characterized in that The method of predicting the four-dimensional power generation data combination mode of each day in the month based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transfer matrix includes: In the four-dimensional power generation data combination probability matrix, the four-dimensional power generation data combination mode of the initial day is sampled, and based on the four-dimensional power generation data combination probability transfer matrix, starting with the four-dimensional power generation data combination mode of the initial day, the four-dimensional power generation data combination mode transferred out on the next day is predicted in sequence until the four-dimensional power generation data combination mode of the total number of days in the month is obtained, and the four-dimensional power generation data combination mode of each day in the month is obtained.

3. The method according to claim 1, characterized in that The obtaining of the probability distribution of each power generation data value in the type of power generation data includes: Based on the kernel density estimation method, the probability distribution of each power generation data value in the type of power generation data is obtained.

4. The method according to claim 1, characterized in that The method of using the inverse transform sampling method to obtain multiple data combination scenarios that obey the probability distribution includes: Based on the cumulative distribution function calculation formula, the cumulative distribution function corresponding to the probability distribution is determined, and the inverse transformation sampling method and the cumulative distribution function are used to obtain multiple data combination scenarios that obey the probability distribution, wherein the cumulative distribution function is: F e (x) is the cumulative distribution function, is the probability distribution, X h are the probability values ​​in the probability distribution.

5. The method according to claim 4, characterized in that The inverse transformation sampling method and the cumulative distribution function are used to obtain multiple data combination scenarios that obey the probability distribution, including: Substitute the random numbers that obey the standard normal distribution and the inverse function corresponding to the cumulative distribution function into the inverse transformation sampling formula to obtain multiple data combination scenarios that obey the probability distribution, where the inverse transformation sampling formula is: P t is the inverse transform sampling result, t is the time, F e is the cumulative distribution function, is the inverse function of the cumulative distribution function, Z t is a random number that obeys a standard normal distribution, and the standard normal distribution function value φ(Z t ) are evenly distributed within the preset value range.

6. The method according to claim 1, characterized in that In the multiple data combination scenarios, typical data combination scenarios with different data combination methods are obtained, including: In multiple data combination scenarios, based on the scenario set retention and deletion metric formula, a deleted data combination scenario is determined each time, and based on the scenario distance similarity calculation formula, the scene closest to the deleted data combination scenario is calculated, and the calculated scene is deleted until the number of remaining typical data combination scenarios meets the preset number, wherein the scenario set retention and deletion metric formula is: The scene distance similarity calculation formula is: D k (C i , C i ′) is a typical data combination scene set C i Combine the scene set C with the deleted data i The metric relationship between ′, N i is the total number of typical data combination scenarios, N i is the total number of deleted data combination scenarios, v is the index of the typical data combination scenario, and the typical data combination scenario set C i in, each typical Initial probability of data combination scenario 1 / N i , the deleted data combination scene set C i In each Initial probability of deleting data combination scenario 1 / N i ,s v Combined scene for deleted data, s v To combine scenes with deleted data v The closest scene.

7. The method according to claim 1, characterized in that Before obtaining the probability distribution of each power generation data value in the type of power generation data, the method further includes: Obtaining a first power generation data value at a first preset percentile and a second power generation data value at a second preset number percentile in the power generation data of the type, and determining an abnormal value range based on the first power generation data value and the second power generation data value; Based on the determined abnormal value range, the abnormal values ​​of the power generation data in the type of power generation data are removed, and the missing values ​​of the power generation data in the type of power generation data are filled by a linear interpolation method.

8. A wind-solar-load joint prediction device, characterized in that: The device comprises: The wind, photovoltaic and load data acquisition module is used to obtain the power generation data of each month in the distributed power grid, wherein the power generation data includes centralized wind power data, centralized photovoltaic power data, distributed photovoltaic power data and net load data, and each type of power generation data includes multiple power generation data values ​​with a time resolution of hourly level; A typical data combination scenario determination module is used to obtain the probability distribution of each power generation data value in any type of power generation data in any historical month, obtain multiple data combination scenarios that obey the probability distribution using an inverse transformation sampling method, and obtain typical data combination scenarios with different data combination methods in the multiple data combination scenarios, wherein the time resolution of the typical power generation data value in each typical data combination scenario is at the daily level; A wind-solar-load joint combination mode prediction module is used to perform four-dimensional Markov chain modeling based on typical data combination scenarios corresponding to various power generation data in the month, obtain a four-dimensional power generation data combination probability matrix and a four-dimensional power generation data combination probability transfer matrix for the month, and predict the four-dimensional power generation data combination mode of each day in the month based on the four-dimensional power generation data combination probability matrix and the four-dimensional power generation data combination probability transfer matrix, wherein the four-dimensional power generation data combination mode is combined by combining the daily-level prediction data values ​​of various power generation data to obtain the monthly four-dimensional power generation data combination mode of the month; The module for generating the wind and solar load forecast results throughout the year is used to determine the monthly forecast data combination methods of various power generation data based on the daily forecast data values ​​of various power generation data in the four-dimensional power generation data combination methods predicted for each day of the month, and determine the target typical data combination scenarios that match the forecast data combination methods of various power generation data, so as to obtain the monthly four-dimensional power generation data joint forecast result of the month, and obtain the annual four-dimensional power generation data joint forecast result based on the monthly four-dimensional power generation data joint forecast result of each month.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for wind-solar-load joint prediction as described in any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method for wind-solar-load joint prediction described in any one of claims 1 to 7 is implemented.

Images