Photovoltaic short-term power probability prediction method of dynamic combination copula function

Abstract

The present application relates to photovoltaic power prediction technical field, specifically points to a kind of dynamic combination Copula function's photovoltaic short-term power probability prediction method, device and equipment, comprising: obtaining all historical data and its edge distribution in adjacent time window before prediction period, multiple three-dimensional Copula functions of current time window are obtained by fitting;The correlation coefficient between historical predicted power, historical prediction error and historical forecast irradiance two two variables in current time window is converted into the dependence parameter of each three-dimensional Copula function of current time window, to obtain each target three-dimensional Copula function of current time window, and then obtain the upper and lower bounds of each power error of current time window under pre-set confidence, and then it is weighted and summed with its target weight coefficient, to obtain the upper and lower bounds of power error of prediction period under pre-set confidence, then it is added with the predicted power in prediction period, and then the power probability prediction interval of prediction period under pre-set confidence is obtained.The present application improves the accuracy of photovoltaic power station power probability prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power prediction, in particular to a photovoltaic short-term power probability prediction method, device and equipment dynamically combining Copula functions. BACKGROUND

[0002] With the acceleration of photovoltaic power station construction, a large number of photovoltaic power stations connected to the power grid will affect power quality and grid operation stability, therefore, it is necessary to accurately predict photovoltaic power in order to predict future photovoltaic power generation power in advance, make good grid operation and dispatching plan, and reduce the adverse effects of a large number of photovoltaic grid connections.

[0003] Existing researches mainly focus on the deterministic prediction of photovoltaic power, however, the probabilistic prediction can quantify the uncertainty of the prediction and provide more comprehensive prediction information for decision makers; the existing probabilistic prediction method usually inputs historical meteorological data or historical power data or historical meteorological data and historical power data into the existing power prediction model to obtain the deterministic prediction power at the prediction time; then, according to the historical power data and meteorological data, a probability interval prediction model of power prediction error is constructed to obtain the probability prediction interval of the power prediction error at the prediction time; finally, the probability prediction interval of the power prediction error corresponding to the prediction time is combined with the deterministic prediction power to obtain the power probability prediction interval at the prediction time; however, in the construction of the probability interval prediction model of the power prediction error, only the kernel density estimation method is used to fit the prediction power error sequence obtained from the historical power data to obtain the historical power prediction error density function, and then the probability prediction interval of the power prediction error is determined, and since the kernel density estimation method can only fit the probability distribution of a single variable sequence, it cannot establish the correlation between the prediction power error and the prediction power and the meteorological data, so that the influence of the dependence relationship between the prediction power error and the prediction power and the meteorological data changing with time on the prediction interval cannot be considered, resulting in insufficient description of the diversified prediction scene of the photovoltaic power station, and the method cannot adapt to the changing multi-scene in the actual operation of the photovoltaic power station, so that the accuracy of the final photovoltaic power station power probability prediction is low. SUMMARY

[0004] Therefore, the technical problem to be solved by the present application is to overcome the problem that the correlation between the prediction power error and the prediction power and the meteorological data is not established in the prior art, resulting in low accuracy of photovoltaic power station power probability prediction.

[0005] To solve the above technical problems, the present application provides a photovoltaic short-term power probability prediction method dynamically combining Copula functions, comprising:

[0006] obtaining a historical actual power sequence, a historical predicted power sequence and a historical forecast irradiance sequence in a previous time window before a prediction period, and calculating a historical prediction error sequence, fitting a plurality of different three-dimensional Copula functions of the current time window in combination with the marginal distribution of the obtained historical predicted power, historical prediction error and historical forecast irradiance;

[0007] transforming the correlation coefficients between the calculated historical predicted power, historical prediction error and historical forecast irradiance in the current time window into dependent parameters of each three-dimensional Copula function of the current time window, to obtain each target three-dimensional Copula function of the current time window;

[0008] obtaining each power error upper and lower bound of the current time window under the preset confidence level by using each target three-dimensional Copula function of the current time window in combination with the historical predicted power sequence and the historical forecast irradiance sequence of the current time window; obtaining a target weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence level; and obtaining the power error upper and lower bound of the prediction period under the preset confidence level by weighted sum of each power error upper and lower bound of the current time window under the preset confidence level and the target weight coefficient thereof;

[0009] adding the power error upper and lower bound of the prediction period under the preset confidence level to the predicted power at each time point in the prediction period to obtain the upper and lower bounds of the predicted power at each time point in the prediction period under the preset confidence level, and further obtaining the power probability prediction interval of the prediction period under the preset confidence level.

[0010] Preferably, the calculation of the correlation coefficients between the historical predicted power, the historical prediction error and the historical forecast irradiance in the current time window comprises:

[0011] based on the historical predicted power sequence, the historical prediction error sequence and the historical forecast irradiance sequence of the current time window, constructing a generalized autoregressive conditional heteroscedasticity model of the historical predicted power, a generalized autoregressive conditional heteroscedasticity model of the historical prediction error and a generalized autoregressive conditional heteroscedasticity model of the historical forecast irradiance of the current time window, and then calculating the conditional variance sequence of the historical predicted power, the conditional variance sequence of the historical prediction error and the conditional variance sequence of the historical forecast irradiance of the current time window;

[0012] dividing the historical predicted power sequence, the historical prediction error sequence and the historical forecast irradiance sequence of the current time window by the conditional variance sequence thereof to obtain the standardized residual sequence of the historical predicted power, the standardized residual sequence of the historical prediction error and the standardized residual sequence of the historical forecast irradiance of the current time window;

[0013] Pearson correlation coefficients between the standardized residual series of the historical predicted power, the standardized residual series of the historical prediction error and the standardized residual series of the historical forecasted irradiance in the current time window are calculated, as the correlation coefficients between the corresponding two variables of the historical predicted power, the historical prediction error and the historical forecasted irradiance in the current time window.

[0014] Preferably, the three different three-dimensional Copula functions of the current time window include a three-dimensional Gaussian Copula function, a three-dimensional Clayton Copula function and a three-dimensional Gumbel Copula function.

[0015] Preferably, the transforming the calculated correlation coefficients between the two variables of the historical predicted power, the historical prediction error and the historical forecasted irradiance in the current time window into the dependence parameters of each three-dimensional Copula function of the current time window includes:

[0016] The expressions of the first dependence parameter , the second dependence parameter and the third dependence parameter of the three-dimensional Gaussian Copula function of the current time window are respectively: ; ; ;

[0017] The expressions of the first dependence parameter , the second dependence parameter and the third dependence parameter of the three-dimensional Clayton Copula function of the current time window are respectively: ; ; ;

[0018] The expressions of the first dependence parameter , the second dependence parameter and the third dependence parameter of the three-dimensional Gumbel Copula function of the current time window are respectively: ; ; ;

[0019] wherein, represents the correlation coefficient between the historical predicted power and the historical prediction error in the current time window; represents the correlation coefficient between the historical predicted power and the historical forecasted irradiance in the current time window; represents the correlation coefficient between the historical prediction error and the historical forecasted irradiance in the current time window.

[0020] Preferably, the obtaining of each power error upper and lower bound of the current time window under the preset confidence level comprises:

[0021] The target three-dimensional Copula function of each target three-dimensional Copula function of the current time window is used to randomly generate a prediction power sample data set, a prediction error sample data set and a forecast irradiance sample data set corresponding to each target three-dimensional Copula function of the current time window.

[0022] According to the mapping formula, the historical prediction power sequence and the historical forecast irradiance sequence of the current time window are mapped to the target three-dimensional Copula function space of the current time window to obtain a prediction power mapping sequence and a forecast irradiance mapping sequence corresponding to each target three-dimensional Copula function of the current time window.

[0023] Based on the prediction power sample data set and the forecast irradiance sample data set corresponding to each target three-dimensional Copula function of the current time window, and the prediction power mapping sequence and the forecast irradiance mapping sequence, a plurality of prediction error samples satisfying the preset tolerance are selected from the prediction error sample data set corresponding to each target three-dimensional Copula function of the current time window to construct a prediction error conditional sample sequence corresponding to each target three-dimensional Copula function of the current time window.

[0024] After obtaining the historical prediction error edge inverse function of the current time window, the prediction error conditional sample sequence corresponding to each target three-dimensional Copula function of the current time window is mapped back to the original data space to obtain a target prediction error sequence corresponding to each target three-dimensional Copula function of the current time window, and the preset confidence level is combined to calculate each power error upper and lower bound of the current time window under the preset confidence level.

[0025] Preferably, the obtaining of the target weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence level comprises:

[0026] The average proportional deviation and the average coverage error of the power probability prediction interval of the prediction period corresponding to the current time window are minimized as the objective function, the weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence level is combined to constrain the particle swarm optimization algorithm, and the weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence level is optimized to obtain the target weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence level.

[0027] Preferably, the expression of the objective function corresponding to the current time window is:

[0028] ;

[0029] ;

[0030] wherein, denotes the average coverage error of the power probability prediction interval of the prediction period corresponding to the current time window; denotes the empirical coverage probability that the actual power value falls in the power probability prediction interval of the prediction period corresponding to the current time window, ; denotes the actual power at the th time instant within the current time window; denotes the power probability prediction interval of the prediction period corresponding to the current time window; denotes whether the event that the actual power value falls in the power probability prediction interval of the prediction period corresponding to the current time window occurs, if the event occurs, , otherwise ; denotes the nominal coverage probability, ; denotes the average proportional bias of the prediction period corresponding to the current time window; denotes the sequence length of the current time window; denotes the prediction power upper bound at the th time instant within the prediction period under the preset confidence level, ; denotes the prediction power lower bound at the th time instant within the prediction period under the preset confidence level, ; denotes the prediction power at the th time instant within the current time window; denotes the power error upper bound of the prediction period under the preset confidence level, ; denotes the power error lower bound of the prediction period under the preset confidence level, ; denotes the th power error upper bound of the current time window under the preset confidence level; denotes the th power error lower bound of the current time window under the preset confidence level;

[0031] The expression of the weight coefficient constraint corresponding to the th power error upper and lower bound of the current time window under the preset confidence level is:

[0032] ;​​

[0033] wherein, represents the weight coefficient corresponding to each power error upper and lower bound in the current time window under the preset confidence level; represents the number of target three-dimensional Copula functions.

[0034] Preferably, the obtaining of the marginal distribution of the historical predicted power, the historical prediction error and the historical forecast irradiance comprises:

[0035] By using kernel density estimation, the marginal distribution of the historical predicted power, the marginal distribution of the historical prediction error and the marginal distribution of the historical forecast irradiance are obtained based on the historical predicted power, the historical prediction error and the historical forecast irradiance at all times.

[0036] The application also provides a photovoltaic short-term power probability prediction device dynamically combining Copula functions, comprising:

[0037] The Copula function fitting module: obtains the historical actual power sequence, the historical predicted power sequence and the historical forecast irradiance sequence in the adjacent time window before the prediction period, and calculates to obtain the historical prediction error sequence, and combines the obtained marginal distribution of the historical predicted power, the historical prediction error and the historical forecast irradiance to fit a plurality of different three-dimensional Copula functions of the current time window;

[0038] The target Copula function determination module: converts the correlation coefficients between the calculated historical predicted power, the historical prediction error and the historical forecast irradiance in the current time window into the dependence parameters of each three-dimensional Copula function of the current time window, to obtain each target three-dimensional Copula function of the current time window;

[0039] The power error upper and lower bound calculation module: under the preset confidence level, by using each target three-dimensional Copula function of the current time window, combining the historical predicted power sequence and the historical forecast irradiance sequence of the current time window obtains each power error upper and lower bound of the current time window under the preset confidence level; obtains the target weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence level; and performs weighted summation on each power error upper and lower bound of the current time window under the preset confidence level and its target weight coefficient, to obtain the power error upper and lower bound of the prediction period under the preset confidence level;

[0040] The power probability prediction interval calculation module: adds the power error upper and lower bound of the prediction period under the preset confidence level to the predicted power at each time in the prediction period to obtain the upper and lower bounds of the predicted power at each time in the prediction period under the preset confidence level, and further obtains the power probability prediction interval of the prediction period under the preset confidence level. ​

[0041] The application further provides a photovoltaic short-term power probability prediction device dynamically combining a Copula function, comprising:

[0042] a memory for storing a computer program;

[0043] a processor for implementing the steps of the above-mentioned photovoltaic short-term power probability prediction method dynamically combining a Copula function when executing the computer program.

[0044] The above-mentioned technical solution of the application has the following beneficial effects compared with the prior art:

[0045] (1) The photovoltaic short-term power probability prediction method dynamically combining a Copula function considers that historical prediction power is an estimation of photovoltaic power station power, contains part of regular information of power station operation, and since photovoltaic power is affected by multiple factors, prediction has errors, historical prediction errors can reflect the difference between predicted values and actual values, and this difference contains information of influencing factors that are not completely captured by a prediction model. Meanwhile, historical forecast irradiance is a key external factor affecting photovoltaic power, and illumination intensity directly determines the potential of photovoltaic power generation and is a key index of historical meteorological data. By analyzing the three variables, the characteristics of photovoltaic power station power variation can be comprehensively reflected from different angles, which provides a basis for accurately predicting short-term power probability and helps to establish a more comprehensive and accurate photovoltaic power station short-term power probability prediction model. Meanwhile, by constructing a joint probability distribution model based on a Copula function from prediction power, prediction error and forecast irradiance, the dependency structure between random variables can be flexibly modeled, the conditional relationship between multiple variables is fully considered, the relevance between prediction power, prediction error and forecast irradiance is established, the photovoltaic power station multi-prediction scene is better described, and the actual situation of prediction is more in line with, thereby improving the reliability and accuracy of photovoltaic power station power probability prediction;

[0046] (2) The photovoltaic short-term power probability prediction method of the dynamic combination Copula function, considering that the correlation coefficient between variables will change over time, the correlation coefficient between the historical prediction power, the historical prediction error and the historical prediction irradiance in the current time window is converted into three parameters of each three-dimensional Copula function in the current time window, so that a plurality of different target three-dimensional Copula functions of the current time window are obtained, so that the target Copula function obtained over time constitutes a dynamic Copula function, the dependence relationship between variables can be dynamically updated, the model can adapt to the scene change, the adaptability to the scene change is improved, and the accuracy of photovoltaic power probability prediction is improved, so that the prediction result can more accurately reflect the actual power change; meanwhile, considering that different Copula function models have different advantages and disadvantages and applicable scenarios, and the particle swarm optimization algorithm is used to optimize the weight, the upper and lower bounds of the power error obtained by the plurality of different Copula functions are weighted and summed to obtain more accurate upper and lower bounds of the error, the advantages of different Copula functions are complementary, the accuracy of the power prediction interval is improved, the prediction result is more valuable, and more reliable photovoltaic power prediction results are provided for power system operation and dispatching;

[0047] (3) The photovoltaic short-term power probability prediction method of the dynamic combination Copula function, the prediction power, the prediction error and the prediction irradiance data of the photovoltaic power station often have heteroscedasticity, that is, the fluctuation degree of data changes over time; by fitting a GARCH model (generalized autoregressive conditional heteroscedasticity model), the heteroscedasticity characteristics can be effectively captured, and the time-varying nature of data fluctuation can be accurately described; the fitted GARCH type is used to obtain the conditional variance of each variable, the original data is divided by the conditional variance, and standardized residuals are obtained, the standardized residuals eliminate the influence of data heteroscedasticity, so that the data of different variables have comparability under the same scale, and the subsequent analysis requirements are met, which provides a stable and reliable data basis for calculating the Pearson correlation coefficient between variables, ensures the accuracy of the correlation analysis, and lays a foundation for constructing a more accurate Copula function model; the Pearson correlation coefficient of the standardized residual sequence is calculated, the dynamic relationship between the prediction power, the prediction error and the prediction irradiance is represented, and the correlation between the variables will change under different meteorological conditions and power station operating states, the GARCH model provides an effective means for timely tracking and reflecting the change, and helps to improve the accuracy of photovoltaic power probability prediction;

[0048] (4) The photovoltaic short-term power probability prediction method of dynamically combining the Copula function, because the Gaussian Copula function is suitable for describing the linear correlation between variables, in the photovoltaic power prediction, if there is an approximate linear relationship between the predicted power, the prediction error and the predicted irradiance, the Gaussian Copula function can better depict the joint distribution between multiple variables, thereby providing an accurate basis for probability prediction, at the same time, the calculation of the Gaussian Copula function is relatively simple, has good mathematical properties, and is convenient for parameter estimation and model solving; the Clayton Copula function has a lower tail correlation characteristic, that is, when one variable takes a smaller value, the other variable also tends to take a smaller value, in the operation of the photovoltaic power station, some extreme conditions may occur, such as in cloudy or night conditions, the predicted irradiance is low, at this time, the predicted power is also low, and the prediction error may be relatively large, the Clayton Copula function can better describe the lower tail correlation, so that the model describes the variable relationship in the extreme condition more accurately, and improves the reliability of the probability prediction in the low power interval; the Gumbel Copula function is suitable for describing the upper tail correlation characteristic, when one variable takes a larger value, the other variable also tends to take a larger value, in the case of sufficient light, the predicted irradiance is high, the output power of the photovoltaic cell is also high, and at the same time, the prediction error may be relatively small, the Gumbel Copula function can capture the upper tail correlation, which is helpful to more accurately describe the joint distribution of variables in the high power interval, and provide more reasonable results for the probability prediction in the high power condition; by combining the three Copula functions, the dependence relationship between the predicted power, the prediction error and the predicted irradiance in different value ranges can be comprehensively described, from linear to nonlinear, from lower tail correlation to upper tail correlation, and the complex relationship between variables is described in more detail, so as to improve the accuracy and reliability of the photovoltaic power probability prediction. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to make the content of the present application more easily understood, the present application will be further described in detail below according to specific embodiments of the present application and in conjunction with the drawings, in which:

[0050] Figure 1 is a flow chart of the photovoltaic short-term power probability prediction method of dynamically combining the Copula function provided by the present application;

[0051] Figure 2 is a schematic diagram of the photovoltaic short-term power probability prediction device provided by the present application. DETAILED DESCRIPTION

[0052] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0053] Reference Figure 1 As shown, Figure 1 This is a flow chart of a photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions provided by the present invention; specifically, it includes:

[0054] S1: Obtain the historical actual power sequence, historical predicted power sequence and historical forecast irradiance sequence in the adjacent time window before the forecast period, and calculate the historical forecast error sequence. Historical forecast error at each moment The expression is: ;

[0055] in, Indicates the The historical actual power at a certain moment; Indicates the The historical forecast power at each moment;

[0056] Based on the historical predicted power series, historical prediction error series, and historical forecast irradiance series of the current time window, and combined with the obtained marginal distributions of historical predicted power, historical prediction error, and historical forecast irradiance, multiple different three-dimensional Copula functions of the current time window are fitted;

[0057] In a specific embodiment of the present invention, kernel density estimation is used to obtain the marginal distribution of historical predicted power, the marginal distribution of historical prediction error, and the marginal distribution of historical predicted irradiance based on the historical predicted power, historical prediction error, and historical predicted irradiance at all times; wherein the marginal distribution is expressed as follows:

[0058] ;

[0059] in, Indicates at point The density estimate at ; Indicates the sequence length of the current time window; Indicates bandwidth parameter; Indicates the The sample data at the moment can be replaced by the corresponding data of the predicted power, prediction error and predicted irradiance at the moment; represents the kernel function;

[0060] S2: Calculate the correlation coefficient between the historical forecast power, historical forecast error and historical forecast irradiance in the current time window, that is:

[0061] Based on the historical forecast power series, historical forecast error series, and historical forecast irradiance series of the current time window, a generalized autoregressive conditional heteroskedasticity model of the historical forecast power, a generalized autoregressive conditional heteroskedasticity model of the historical forecast error, and a generalized autoregressive conditional heteroskedasticity model of the historical forecast irradiance of the current time window are constructed. Then, the conditional variance series of the historical forecast power, the conditional variance series of the historical forecast error, and the conditional variance series of the historical forecast irradiance of the current time window are calculated.

[0062] Divide the historical prediction power sequence of the current time window by its conditional variance sequence, the historical prediction error sequence by its conditional variance sequence, and the historical forecast irradiance sequence by its conditional variance sequence to obtain the standardized residual sequence of the historical prediction power of the current time window, the standardized residual sequence of the historical prediction error, and the standardized residual sequence of the historical forecast irradiance;

[0063] Calculate the Pearson correlation coefficient between the standardized residual sequence of historical predicted power, the standardized residual sequence of historical prediction error, and the standardized residual sequence of historical forecast irradiance in the current time window as the correlation coefficient between the pairwise variables of historical predicted power, historical prediction error, and historical forecast irradiance in the current time window;

[0064] Among them, the expression of the generalized autoregressive conditional heteroskedasticity model is:

[0065] ;

[0066] ;

[0067] in, Indicates the The actual value at the moment; Indicates the The conditional mean at each moment; Indicates the The conditional standard deviation at each moment; represents the residual term, ; Indicates the The white noise at each moment obeys the standard normal distribution; represents a constant term; represents the error hysteresis influence coefficient; represents the fluctuation lag influence coefficient; Indicates the Order error hysteresis influence coefficient; Indicates the order of the error lag influence coefficient; Indicates the The order represents the coefficient of influence of fluctuation lag; represents the order of the fluctuation lag influence coefficient; when , =1 is a special case of the GARCH model, that is, only the variance and residual square of the previous moment are considered, and it is used to describe the volatility aggregation phenomenon of the forecast error time series;

[0068] Normalized residual series The expression is: ;

[0069] in, Indicates the of a moment The corresponding standardized residuals; Indicates the of a moment The corresponding conditional variance; Indicates the The sample data at the moment can be replaced by the corresponding data of the predicted power, prediction error and predicted irradiance at the moment;

[0070] Pearson correlation coefficient The expression is:

[0071] ;

[0072] in, Indicates the first standardized residual sequence Data at a moment; represents the mean of the first standardized residual series; Indicates the first part of the second standardized residual sequence Data at a moment; represents the mean of the second standardized residual sequence; substitute any two standardized residual sequences in the standardized residual sequence of historical predicted power, the standardized residual sequence of historical prediction error, and the standardized residual sequence of historical forecast irradiance in the current time window in turn to obtain their corresponding Pearson correlation coefficients;

[0073] The calculated correlation coefficients between the historical predicted power, historical prediction error, and historical forecast irradiance in the current time window are converted into dependent parameters of each three-dimensional Copula function in the current time window, and multiple target three-dimensional Copula functions in the current time window are obtained;

[0074] In a specific embodiment of the present invention, the three different three-dimensional Copula functions obtained by fitting the current time window include: a three-dimensional Gaussian Copula function, a three-dimensional Clayton Copula function, and a three-dimensional Gumbel Copula function;

[0075] The three-dimensional Gaussian Copula function The expression of the three-dimensional Gaussian Copula function is:

[0076] ;

[0077] wherein, denotes a variable, ; denote a historical predicted power variable, a historical prediction error variable and a historical forecast irradiance variable, respectively; denotes the inverse cumulative distribution function of a standard normal distribution; denotes the cumulative distribution function of a three-dimensional joint normal distribution with the covariance matrix:

[0078] ;

[0079] The expressions of the first dependency parameter , the second dependency parameter and the third dependency parameter of the three-dimensional Gaussian Copula function of the current time window are: ; ; ;

[0080] The expression of the three-dimensional Clayton Copula function is:

[0081] ;

[0082] wherein, ; the expressions of the first dependency parameter , the second dependency parameter and the third dependency parameter of the three-dimensional Clayton Copula function of the current time window are: ; ; ;

[0083] The expression of the three-dimensional Gumbel Copula function is:

[0084] ;

[0085] wherein, ; the expressions of the first dependency parameter , the second dependency parameter and the third dependency parameter of the three-dimensional Gumbel Copula function of the current time window are: ; ; ;

[0086] wherein, represents a correlation coefficient between the historical predicted power and the historical prediction error in the current time window; represents a correlation coefficient between the historical predicted power and the historical predicted irradiance in the current time window; represents a correlation coefficient between the historical prediction error and the historical predicted irradiance in the current time window;

[0087] S3: under the preset confidence level, using the plurality of target three-dimensional Copula functions of the current time window, combining the historical predicted power sequence and the historical predicted irradiance sequence of the current time window, obtaining a plurality of power error upper and lower bounds of the current time window under the preset confidence level, including:

[0088] using each target three-dimensional Copula function of the current time window, randomly generating a predicted power sample data set, a prediction error sample data set and a predicted irradiance sample data set corresponding to each target three-dimensional Copula function of the current time window;

[0089] According to the mapping formula, the historical predicted power sequence and the historical predicted irradiance sequence of the current time window are mapped to each target three-dimensional Copula function space of the current time window to obtain a predicted power mapping sequence and a predicted irradiance mapping sequence corresponding to each target three-dimensional Copula function of the current time window; wherein the mapping formula is: represents the predicted power mapping sequence corresponding to the first target three-dimensional Copula function ; represents the historical predicted power sequence of the current time window; represents the predicted irradiance mapping sequence corresponding to the first target three-dimensional Copula function ; represents the historical predicted irradiance sequence of the current time window;

[0090] Based on the predicted power sample data set and the predicted irradiance sample data set corresponding to each target three-dimensional Copula function of the current time window, and the predicted power mapping sequence and the predicted irradiance mapping sequence, a plurality of prediction error samples satisfying the preset tolerance are selected from the prediction error sample data set corresponding to each target three-dimensional Copula function of the current time window, and a prediction error conditional sample sequence corresponding to each target three-dimensional Copula function of the current time window is constructed, and the expression is: ; wherein, represents the prediction error conditional sample sequence corresponding to the first target three-dimensional Copula function ; ​​​represents a preset tolerance, in the process of solving the probability interval, the tolerance refers to when there is a difference between two or more variables, how much the value of one variable can change without affecting the similarity or difference between the two variables, in simple terms, the tolerance can measure the fault tolerance between two or more variables; 、 and respectively represent the first target three-dimensional Copula function corresponding to the first predicted power condition sample, the first prediction error condition sample and the first forecast irradiance condition sample;

[0091] According to the historical prediction error marginal distribution of the current time window, the historical prediction error marginal inverse function of the current time window is obtained, and then the prediction error condition sample sequence corresponding to each target three-dimensional Copula function of the current time window is mapped back to the original data space to obtain the target prediction error sequence corresponding to each target three-dimensional Copula function of the current time window, and its expression is: ; Wherein, represents the target prediction error sequence corresponding to the first target three-dimensional Copula function; represents the historical prediction error marginal inverse function of the current time window;

[0092] Based on the target prediction error sequence corresponding to each target three-dimensional Copula function of the current time window, the upper and lower bounds of each power error under the preset confidence are calculated, and its expression is:

[0093] ;

[0094] ;

[0095] Wherein, represents the upper bound of the first power error in the current time window under the preset confidence; represents the lower bound of the first power error in the current time window under the preset confidence; represents the significance level; represents the preset confidence;

[0096] The target weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence is obtained, including:

[0097] The average proportional deviation and the minimum average coverage error of the power probability prediction interval corresponding to the prediction period of the current time window are taken as the objective function, the weight coefficient constraint corresponding to each power error upper and lower bound of the current time window under the preset confidence is combined, the particle swarm optimization algorithm is used to optimize the weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence, and the target weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence is obtained.

[0098] The expression of the objective function corresponding to the current time window is:

[0099] ;

[0100] ;

[0101] The average proportional deviation of the power probability prediction interval corresponding to the prediction period of the current time window is represented by . The empirical coverage probability of the actual power value falling in the power probability prediction interval corresponding to the prediction period of the current time window is represented by . ; The actual power at the i th moment in the current time window is represented by . The power probability prediction interval corresponding to the prediction period of the current time window is represented by . The event of the actual power value falling in the power probability prediction interval corresponding to the prediction period of the current time window is represented by . The nominal coverage probability is represented by . The average proportional deviation of the power probability prediction interval corresponding to the prediction period of the current time window is represented by . The sequence length of the current time window is represented by . The prediction power upper bound at the i th moment in the prediction period under the preset confidence is represented by . The prediction power lower bound at the i th moment in the prediction period under the preset confidence is represented by . The prediction power at the i th moment in the current time window is represented by . The power error upper bound of the prediction period under the preset confidence is represented by . The power error lower bound of the prediction period under the preset confidence is represented by = ; represents the upper bound of the power error of the current time window under the preset confidence level; represents the lower bound of the power error of the current time window under the preset confidence level; represents the upper bound of the power error of the current time window under the preset confidence level; represents the lower bound of the power error of the current time window under the preset confidence level;

[0102] the upper and lower bounds of the power error of the current time window under the preset confidence level The expression of the weight coefficient constraint corresponding to the upper and lower bounds of the power error of the current time window under the preset confidence level is:

[0103] ;

[0104] wherein, represents the weight coefficient corresponding to the upper and lower bounds of the power error of the current time window under the preset confidence level; represents the number of target three-dimensional Copula functions;

[0105] The upper and lower bounds of the power error of the prediction period under the preset confidence level are obtained by weighting and summing each upper and lower bound of the power error of the current time window under the preset confidence level and the target weight coefficient thereof;

[0106] In one specific embodiment of the present application, the initial weight coefficients corresponding to the upper and lower bounds of the power error obtained based on the three-dimensional Gaussian Copula function, the three-dimensional Clayton Copula function and the three-dimensional Gumbel Copula function are 0.3, 0.3 and 0.4 in turn; on this basis, the corresponding target weight coefficients are obtained by optimization using the steps in S3;

[0107] S4: The upper and lower bounds of the predicted power at each time point in the prediction period under the preset confidence level are obtained by adding the upper and lower bounds of the power error of the prediction period under the preset confidence level to the predicted power at each time point in the prediction period obtained; and then the power probability prediction interval of the prediction period under the preset confidence level is obtained; wherein, the power probability prediction interval at each time point in the prediction period under the preset confidence level is obtained based on the upper and lower bounds of the predicted power at each time point in the prediction period under the preset confidence level; and finally, the power probability prediction interval of the prediction period under the preset confidence level is obtained by integrating the power probability prediction intervals of all time points in the prediction period under the preset confidence level.

[0108] In verifying the effectiveness of the photovoltaic short-term power probability prediction method provided by the present application, all historical data in the preset number of days are obtained, including historical measured data, historical prediction data and historical forecast irradiance data at each time point, and historical prediction error data at each time point is calculated; wherein, the time resolution when obtaining the related data each day is set to 15 minutes, so that each variable each day corresponds to 96 data points; ​

[0109] The size of the time window is set to 672, the sliding step is set to 96, the sliding time window is set, and based on all the historical data in the preset days, all the historical data in the plurality of time windows corresponding to the preset days is obtained; the prediction period is set to one day after the time window, and then based on all the historical data in 7 days in each time window, the power probability prediction interval of the next day of each time window is obtained according to steps S1 to S4, and the accuracy of the dynamic combination Copula function photovoltaic short-term power probability prediction method is verified in combination with the actual power of the next day of each time window.

[0110] Therefore, in the actual application process, when a plurality of prediction periods are set, the historical actual power sequence, the historical predicted power sequence and the historical predicted irradiance sequence in the adjacent time window before each prediction period can be used to obtain the power probability prediction interval of each prediction period based on steps S1 to S4.

[0111] In summary, the present application has the following advantages:

[0112] (1) The present application firstly uses kernel density estimation to fit the marginal distribution of the predicted power, the prediction error and the predicted irradiance of the photovoltaic power station; then GARCH model is fitted for the three variables, the standardized residual and the Pearson correlation coefficient are calculated for each day by setting the sliding window; finally, the Gaussian Copula function, the Clayton Copula function and the Gumbel Copula function are fitted according to the related data in the entire data set and the fitted marginal probability distribution, and the inter-variable correlation coefficient is converted into the dependence parameter in the Copula function for each time window, and the joint probability distribution model based on the Copula function is established, which considers the correlation and dependence relationship between the three variables, can better describe the multi-variable prediction scenario, is more consistent with the actual situation of prediction, improves the generalization ability of the probability prediction model, and is more suitable for the actual photovoltaic power station scenario.

[0113] (2) The present application selects Gaussian Copula function, Clayton Copula function and Gumbel Copula function, calculates the upper and lower bounds of the prediction error under a certain confidence level, then takes 0.3, 0.3 and 0.4 as the weight initial value, and performs weighted summation on the upper and lower bounds of the intervals obtained by the three Copula functions to obtain the error upper and lower bounds; then the particle swarm optimization algorithm is used to optimize the weight to obtain the final error upper and lower bounds; different Copula function models are combined, which have different advantages and disadvantages and suitable scenarios, realize the complementary advantages of different Copula functions, solve the defects of the model constructed by a single Copula function, and adapt to the multi-variable prediction scenario, thereby improving the accuracy of the power prediction interval.

[0114] (3) The application sets a sliding time window, fits a GRACH model of predicted power, prediction error and forecast irradiance for each time window, converts the mutual correlation coefficient between the three variables calculated for each time window into a parameter in a Copula function, thereby obtaining a dynamically combined target Copula function, and dynamically updates the dependency relationship between the variables through the sliding time window, thereby solving the problem that the influence of the change of the dependency relationship between the variables of the photovoltaic power station over time on the final probability interval prediction is difficult to consider in the prior art, improving the adaptability of the model to scene changes, and thereby improving the accuracy of photovoltaic power probability prediction.

[0115] Referring to Figure 2 as shown, Figure 2 is a schematic diagram of a photovoltaic short-term power probability prediction device of a dynamically combined Copula function provided by the application; specifically comprising:

[0116] The Copula function fitting module 100: obtains the historical actual power sequence, the historical predicted power sequence and the historical forecast irradiance sequence in the adjacent time window before the prediction period, and calculates the historical prediction error sequence, and combines the edge distribution of the historical predicted power, the historical prediction error and the historical forecast irradiance to fit a plurality of different three-dimensional Copula functions of the current time window;

[0117] The target Copula function determination module 200: converts the correlation coefficient between the historical predicted power, the historical prediction error and the historical forecast irradiance in the current time window into the dependency parameter of each three-dimensional Copula function of the current time window, and obtains each target three-dimensional Copula function of the current time window;

[0118] The power error upper and lower bound calculation module 300: under a preset confidence level, uses each target three-dimensional Copula function of the current time window, combines the historical predicted power sequence and the historical forecast irradiance sequence of the current time window to obtain each power error upper and lower bound of the current time window under the preset confidence level; obtains the target weight coefficient corresponding to each power error upper and lower bound of the current time window under the preset confidence level; and performs weighted summation on each power error upper and lower bound of the current time window under the preset confidence level and the target weight coefficient thereof, to obtain the power error upper and lower bound of the prediction period under the preset confidence level.

[0119] The power probability prediction interval calculation module 400: adds the power error upper and lower bound of the prediction period under the preset confidence level to the predicted power at each time point in the prediction period to obtain the upper and lower bounds of the predicted power at each time point in the prediction period under the preset confidence level, and thereby obtains the power probability prediction interval of the prediction period under the preset confidence level.

[0120] The device of the embodiment is used to implement the photovoltaic short-term power probability prediction method of dynamically combining Copula functions, and therefore the specific embodiments of the photovoltaic short-term power probability prediction device of dynamically combining Copula functions can be found in the embodiment part of the photovoltaic short-term power probability prediction method of dynamically combining Copula functions in the foregoing. For example, the Copula function fitting module 100, the target Copula function determination module 200, the power error upper and lower bound calculation module 300, and the power probability prediction interval calculation module 400 are respectively used to implement S1 to S4 in the photovoltaic short-term power probability prediction method of dynamically combining Copula functions. Therefore, the specific embodiments can refer to the description of the respective embodiments of the corresponding parts, and will not be described here again.

[0121] In one specific embodiment of the application, a photovoltaic short-term power probability prediction device of dynamically combining Copula functions is also provided, comprising:

[0122] The memory is configured to store a computer program, and the processor is configured to implement the steps of the photovoltaic short-term power probability prediction method of dynamically combining Copula functions when executing the computer program.

[0123] Those skilled in the art should understand that the embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can adopt a computer program product in the form of being implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0124] The application is described with reference to flowcharts and / or block diagrams according to the method, device (system), and computer program product of the embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks.

[0125] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or multiple blocks.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or multiple blocks.

[0127] Obviously, the above-described embodiments are only examples and are not intended to limit the present application. Based on the above description, one of ordinary skill in the art can make other variations and changes of different forms. Here, it is not necessary and impossible to enumerate all the embodiments. The obvious variations and changes derived therefrom are still within the scope of the present application.

Claims

1. A photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions, characterized in that: include: Obtain the historical actual power series, historical predicted power series, and historical forecast irradiance series in the adjacent time windows before the forecast period, and calculate the historical forecast error series. Combined with the obtained marginal distributions of historical predicted power, historical prediction error, and historical forecast irradiance, multiple different three-dimensional Copula functions for the current time window are fitted. The calculated correlation coefficients between the historical predicted power, historical prediction error, and historical forecast irradiance in the current time window are converted into dependent parameters of each three-dimensional Copula function in the current time window, and each target three-dimensional Copula function in the current time window is obtained; Under the preset confidence, using each target three-dimensional Copula function of the current time window, combined with the historical predicted power sequence and the historical predicted irradiance sequence of the current time window, the upper and lower bounds of each power error in the current time window under the preset confidence are obtained; the target weight coefficient corresponding to each upper and lower bound of the power error in the current time window under the preset confidence is obtained; each upper and lower bound of the power error in the current time window under the preset confidence is weighted and summed with its target weight coefficient to obtain the upper and lower bounds of the power error in the prediction period under the preset confidence; The upper and lower bounds of the power error of the prediction period under the preset confidence level are added to the predicted power at each moment in the obtained prediction period to obtain the upper and lower bounds of the predicted power at each moment in the prediction period under the preset confidence level, and then the power probability prediction interval of the prediction period under the preset confidence level is obtained.

2. The photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions according to claim 1 is characterized in that: Calculating the correlation coefficients between the historical forecast power, historical forecast error, and historical forecast irradiance in the current time window includes: Based on the historical forecast power series, historical forecast error series, and historical forecast irradiance series of the current time window, a generalized autoregressive conditional heteroskedasticity model of the historical forecast power, a generalized autoregressive conditional heteroskedasticity model of the historical forecast error, and a generalized autoregressive conditional heteroskedasticity model of the historical forecast irradiance of the current time window are constructed. Then, the conditional variance series of the historical forecast power, the conditional variance series of the historical forecast error, and the conditional variance series of the historical forecast irradiance of the current time window are calculated. Divide the historical prediction power sequence of the current time window by its conditional variance sequence, the historical prediction error sequence by its conditional variance sequence, and the historical forecast irradiance sequence by its conditional variance sequence to obtain the standardized residual sequence of the historical prediction power of the current time window, the standardized residual sequence of the historical prediction error, and the standardized residual sequence of the historical forecast irradiance; The Pearson correlation coefficients between the standardized residual sequence of the historical predicted power, the standardized residual sequence of the historical prediction error, and the standardized residual sequence of the historical predicted irradiance in the current time window are calculated as the correlation coefficients between the corresponding pairwise variables of the historical predicted power, historical prediction error, and historical predicted irradiance in the current time window.

3. The photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions according to claim 1 is characterized in that: The three different three-dimensional Copula functions obtained by fitting the current time window include: a three-dimensional Gaussian Copula function, a three-dimensional Clayton Copula function, and a three-dimensional Gumbel Copula function.

4. The photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions according to claim 3 is characterized in that: The calculation of the correlation coefficients between the historical predicted power, historical prediction error, and historical forecast irradiance in the current time window is converted into dependent parameters of each three-dimensional Copula function in the current time window, including: The first dependent parameter of the three-dimensional Gaussian Copula function of the current time window , the second dependent parameter and the third dependent parameter The expressions are: ; ; ; The first dependent parameter of the three-dimensional Clayton Copula function of the current time window , the second dependent parameter and the third dependent parameter The expressions are: ; ; ; The first dependent parameter of the three-dimensional Gumbel Copula function of the current time window , the second dependent parameter and the third dependent parameter The expressions are: ; ; ; in, represents the correlation coefficient between the historical prediction power and the historical prediction error in the current time window; Indicates the correlation coefficient between the historical predicted power and the historical forecast irradiance in the current time window; Represents the correlation coefficient between the historical forecast error and the historical forecast irradiance in the current time window.

5. The photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions according to claim 1 is characterized in that: The three-dimensional Copula function of each target in the current time window is used to combine the historical predicted power sequence and the historical forecast irradiance sequence of the current time window to obtain the upper and lower bounds of each power error in the current time window under the preset confidence level, including: Using each target three-dimensional Copula function in the current time window, randomly generate the predicted power sample data set, prediction error sample data set, and predicted irradiance sample data set corresponding to each target three-dimensional Copula function in the current time window; According to the mapping formula, the historical predicted power sequence and the historical forecast irradiance sequence of the current time window are mapped to each target three-dimensional Copula function space of the current time window, and the predicted power mapping sequence and the forecast irradiance mapping sequence corresponding to each target three-dimensional Copula function of the current time window are obtained; Based on the predicted power sample dataset and the predicted irradiance sample dataset corresponding to each target three-dimensional Copula function in the current time window, as well as the predicted power mapping sequence and the predicted irradiance mapping sequence, multiple prediction error samples that meet the preset tolerance are selected from the prediction error sample dataset corresponding to each target three-dimensional Copula function in the current time window to construct a prediction error conditional sample sequence corresponding to each target three-dimensional Copula function in the current time window; According to the marginal distribution of historical prediction errors in the current time window, after obtaining the inverse marginal function of historical prediction errors in the current time window, the prediction error conditional sample sequence corresponding to each target three-dimensional Copula function in the current time window is mapped back to the original data space to obtain the target prediction error sequence corresponding to each target three-dimensional Copula function in the current time window. Combined with the preset confidence, the upper and lower bounds of each power error in the current time window under the preset confidence are calculated.

6. The photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions according to claim 1 is characterized in that: The step of obtaining the target weight coefficient corresponding to each upper and lower bound of the power error in the current time window under the preset confidence level includes: Taking the minimum average proportional deviation and the minimum average coverage error of the power probability prediction interval of the prediction period corresponding to the current time window as the objective function, combined with the weight coefficient constraints corresponding to the upper and lower bounds of each power error in the current time window under the preset confidence, the particle swarm optimization algorithm is used to optimize the weight coefficient corresponding to the upper and lower bounds of each power error in the current time window under the preset confidence, and the target weight coefficient corresponding to the upper and lower bounds of each power error in the current time window under the preset confidence is obtained.

7. The photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions according to claim 6 is characterized in that: The expression of the objective function corresponding to the current time window is: ； ； in, Indicates the average coverage error of the power probability prediction interval corresponding to the prediction period of the current time window; Indicates the empirical coverage probability that the actual power value falls within the power probability prediction interval corresponding to the prediction period of the current time window, = ; Indicates the number of The actual power at a moment; Indicates the power probability prediction interval of the current time window corresponding to the prediction period; Indicates whether the event that the actual power value falls within the power probability prediction interval of the prediction period corresponding to the current time window occurs. If the actual power value falls within the power probability prediction interval of the prediction period corresponding to the current time window occurs, then , otherwise ; represents the nominal coverage probability, ; Indicates the average proportional deviation of the current time window corresponding to the forecast period; Indicates the sequence length of the current time window; Indicates the first The upper bound of the predicted power at the moment, ; Indicates the first The lower bound of the predicted power at the moment, ; Indicates the number of The predicted power at each moment; represents the upper bound of the power error in the prediction period under the preset confidence level, ; It represents the lower bound of the power error in the prediction period under the preset confidence level, = ; Indicates the first An upper bound on the power error; Indicates the first A lower bound on the power error; The current time window under the preset confidence The expressions of the weight coefficient constraints corresponding to the upper and lower bounds of the power error are: ； in, Indicates the first The weight coefficients corresponding to the upper and lower bounds of the power error; Indicates the number of target three-dimensional copula functions.

8. The photovoltaic short-term power probability prediction method based on a dynamic combination of Copula functions according to claim 1 is characterized in that: The obtaining of the marginal distribution of the historical predicted power, the historical predicted error and the historical predicted irradiance includes: Using kernel density estimation, based on the historical predicted power, historical prediction error and historical predicted irradiance at all times, the marginal distribution of historical predicted power, the marginal distribution of historical prediction error and the marginal distribution of historical predicted irradiance are obtained.

9. A photovoltaic short-term power probability prediction device based on a dynamic combination of Copula functions, characterized in that: include: Copula function fitting module: obtains the historical actual power series, historical predicted power series, and historical forecast irradiance series in the adjacent time windows before the forecast period, and calculates the historical forecast error series. Combined with the obtained marginal distribution of historical predicted power, historical prediction error, and historical forecast irradiance, multiple different three-dimensional Copula functions of the current time window are fitted; Target Copula function determination module: converts the calculated correlation coefficients between the historical forecast power, historical forecast error, and historical forecast irradiance in the current time window into dependent parameters of each three-dimensional Copula function in the current time window, and obtains each target three-dimensional Copula function in the current time window; Power error upper and lower bound calculation module: Under the preset confidence, use the three-dimensional Copula function of each target in the current time window, combined with the historical predicted power series and historical forecast irradiance series of the current time window Obtaining each upper and lower bound of the power error in the current time window under a preset confidence level; obtaining the target weight coefficient corresponding to each upper and lower bound of the power error in the current time window under a preset confidence level; performing a weighted summation of each upper and lower bound of the power error in the current time window under a preset confidence level and its target weight coefficient to obtain the upper and lower bounds of the power error in the prediction period under the preset confidence level; Power probability prediction interval calculation module: Add the upper and lower bounds of the power error of the prediction period under the preset confidence level to the predicted power at each moment in the obtained prediction period to obtain the upper and lower bounds of the predicted power at each moment in the prediction period under the preset confidence level, and then obtain the power probability prediction interval of the prediction period under the preset confidence level.

10. A photovoltaic short-term power probability prediction device based on a dynamic combination of Copula functions, characterized in that: include: Memory for storing computer programs; A processor is configured to implement the steps of a photovoltaic short-term power probability prediction method using a dynamic combined Copula function as described in any one of claims 1 to 8 when executing the computer program.

Images