A scenario generation method considering the uncertainty of distributed power generation and load output

Abstract

The present invention discloses a scenario generation method that takes into account the uncertainty of distributed power sources and load output. The method comprises the following steps: first, in view of the strong randomness and large volatility of wind power generation, distributed wind power is used as the research object of the uncertainty of power source output, and based on the binning theory and inverse transformation sampling method, multi-time section dynamic scenarios that take into account the randomness and volatility of power source output are generated; second, a copula function is introduced to generate spatial static scenarios with correlation; second, a scenario sorting method is used to couple and generate spatiotemporal dynamic and static scenarios that take into account the randomness, volatility and correlation of power source output; third, the above method is continued to process the uncertainty of load change, and the load characteristics are combined with the uncertainty to form corresponding typical scenarios; finally, the above-generated multiple scenarios are subjected to scenario reduction using the K-means clustering algorithm to generate typical scenarios, thereby improving the efficiency of subsequent calculations and scenario processing, and transforming the source-load output uncertainty problem into a probabilistic deterministic problem.

Description

Technical Field

[0001] The present invention belongs to the power industry and relates to power planning, and in particular to a scenario generation method taking into account the uncertainty of distributed power sources and load output. Background Art

[0002] Distributed power generation (DG) is a small-scale power supply device located close to the user, integrating local production, conversion, and use. The rational deployment and utilization of DGs can not only reduce network losses and the economic costs associated with traditional long-distance power transmission, but also improve system voltage distribution and power quality, providing users with a better electricity experience. Leveraging the green nature of renewable energy sources such as wind and solar power, they can also bring positive environmental benefits to society. Furthermore, because traditional energy sources are generally generated through centralized, large-scale grids, human error, equipment failure, and other incidents can easily cause widespread power outages within the power supply area. The introduction of DGs, however, creates power supply islands to meet load power demands. Due to its low cost, local consumption, reliable power supply, flexibility and efficiency, low-carbon and environmentally friendly operation, and multifaceted interactions, DGs are becoming an indispensable power supply model in the energy transition.

[0003] While distributed power generation (DGs) have been widely recognized and utilized internationally due to their numerous advantages, their integration into the power grid presents significant challenges. Traditional distribution networks have a single network topology and fixed power flow distribution. The integration of DGs injects new voltage and power into the distribution network, altering the network topology and power flow distribution. Improper integration can lead to reverse power flow in branches, voltage violations, and increased line losses, impacting system operation. Furthermore, wind and photovoltaic power generation are the primary generators of distributed generation. While these offer advantages such as renewable energy and environmental friendliness, wind power generation is affected by wind speed, while photovoltaic power generation is affected by sunlight intensity. Both wind speed and sunlight intensity fluctuate in real time. Low wind speed and sunlight intensity result in low output power, resulting in insufficient system power supply. High wind speed and sunlight intensity result in high output power, making it difficult to absorb the load and resulting in wasted resources. Therefore, the uncertainty in the output of distributed wind and photovoltaic power generation also places a burden on system operation.

[0004] In summary, given the current development trends of new power systems, my country's renewable energy power supply model will gradually replace the traditional fossil fuel power supply model, laying a solid foundation for achieving sustainable energy development. At the same time, the scale of distributed power generation (generated and consumed locally) integrated into the grid will gradually increase, and the proportion of distributed wind power and photovoltaic power generation will also increase. Therefore, considering the uncertainty of wind and solar power generation has important research and practical significance for the stable and efficient operation of distribution networks. Summary of the Invention

[0005] To address the deficiencies of the above-mentioned background technology, the present invention proposes a scenario generation method that takes into account the uncertainty of distributed power supply and load output.

[0006] The specific technical solution of the present invention is a scenario generation method taking into account the uncertainty of distributed power supply and load output, which specifically includes the following steps:

[0007] Step 1: Given the strong randomness and volatility of wind power generation, distributed wind power is used as the research object for the uncertainty of power output. Based on the binning theory and inverse transformation sampling method, a multi-time section dynamic scenario that takes into account the randomness and volatility of power output is generated;

[0008] Step 2: Introduce the Copula function to generate a spatial static scene with correlation;

[0009] Step 3: Using the scenario ranking method, generate coupled spatiotemporal dynamic and static scenarios that take into account the randomness, volatility, and correlation of power output;

[0010] Step 4: Continue the above method to deal with the uncertainty of load changes, and combine the load characteristics with the uncertainty to form the corresponding typical scenario;

[0011] Step 5: Use the K-means clustering algorithm to reduce the multiple scenarios generated above to generate typical scenarios, improve the efficiency of subsequent calculations and scenario processing, and transform the source-load output uncertainty problem into a probabilistic deterministic problem.

[0012] Step 1 First, the nonparametric kernel density estimation method can be used to effectively characterize the power distribution function. The process is as follows:

[0013] The nonparametric kernel density estimation method is as follows:

[0014]

[0015] Where x represents x i The i in the equation is a random variable; k is the sample size; h is the sample bandwidth, whose value affects the function fitting and is always greater than 0; K(x) is the kernel function, which is a probability density function that is always greater than 0, has an integral of 1, and a mean of 0. Its commonly used Gaussian function form is shown in the above formula.

[0016] Since wind speeds fluctuate in real time, a large amount of historical data is generated. Binning effectively integrates this data, reducing the time required for the forecast solution and increasing accuracy. The "bin" here refers to the wind power forecast bin. Historical wind power data for any timeframe consists of a day-ahead power forecast and the measured power values ​​sampled on the same day. Each predicted power value is matched by an actual power value, which serves as the data element within the bin.

[0017] Because the predicted power is arranged in ascending order, the difference between any adjacent data is not large. However, the corresponding actual power will produce large data discrepancies due to real-time changes in wind speed, which reflects the randomness of wind power generation. By fitting wind power with an appropriate probability distribution model and treating actual power as the conditional probability distribution of the predicted power, we can better solve and analyze the randomness of wind power generation.

[0018] Secondly, the concept of multivariate normal distribution is introduced to further analyze the volatility of wind power output;

[0019] The random output process P of the power supply side obtained based on the above random output processing is P = {P w ,w∈T} T When dealing with output fluctuation problems, the wind power of multiple time sections can be regarded as a multivariate random vector Y = (Y1, Y2, ..., Y K ) T , where K represents the number of time sections. Assume that the multivariate random variable Y follows a multivariate normal distribution Y~N(μ,∑). Where μ represents the expectation and is a K-dimensional zero vector. ∑ is the covariance matrix, which is expressed as follows:

[0020]

[0021] Where, δ i,j =cov(Y i ,Y j ) represents Y i ,Y j The covariance between .

[0022] The covariance in the above formula reflects the correlation between wind power output power at different time sections i and j, and the covariance matrix formed by the covariance can characterize the volatility of wind power output. From the above analysis, it can be seen that the key to generating dynamic scenarios that consider both the randomness and volatility of wind power output lies in the covariance matrix [δ i,j ] K×K ; Solve the covariance as follows:

[0023]

[0024] Where ε represents the range parameter, which is used to adjust the strength of the power correlation between different time sections i and j.

[0025] From the above formula, we can see that the farther the distance between different time sections is, the larger |ij| is, the greater the covariance δ i,j The smaller the value, the weaker the power correlation between time sections i and j; conversely, the closer the distance between different time sections, that is, the smaller |ij|, the smaller the covariance δ i,j The larger the value, the stronger the power correlation between time sections i and j. The changes in this model are consistent with reality, so the covariance δ can be effectively solved. i,j .

[0026] Assume the objective function f(ε) of the covariance range parameter ε is:

[0027]

[0028] Where N is the number of sampling populations; s represents the equally spaced sampling points; S represents the fluctuation range of wind power; pdf(s) represents the probability density function obtained by fitting the fluctuation of wind power in a randomly generated dynamic scenario; and pdf'(s) represents the probability density function obtained by fitting the fluctuation of historical wind power data.

[0029] The wind power fluctuation ΔP is defined as:

[0030] ΔP=P t -P t+1

[0031] Where, P t is the wind power at time t, P t+1 is the wind power at time t+1.

[0032] Since the probability of extreme wind power in the t location-scale distribution is greater than that in the normal distribution and is more consistent with the actual output power characteristics of wind power, this section uses the t location-scale distribution function to characterize short-term wind power fluctuations. The probability density function of the t location-scale distribution is as follows:

[0033]

[0034] Where μ represents the location parameter, k represents the scale parameter, and v represents the shape parameter.

[0035] pdf(s) and pdf'(s) represent the probability density functions corresponding to the location-scale distribution function. The solution to the randomly generated dynamic scene wind power fluctuation is as follows: when the ε value is determined, the covariance matrix [δ i,j ] K×Kis also determined at the same time, then the corresponding multivariate standard normal distribution random vector Y=(Y1,Y2,…,Y K ) T , then using the inverse transform for random sampling, we can generate n day-ahead wind power dynamic scenarios. For each dynamic scenario, we can also solve for the wind power dynamic scenario fluctuation data. Combining these acquired data, we can determine the covariance range parameter ε, which more accurately characterizes the volatility of wind power output.

[0036] Finally, in order to better analyze the randomness of wind power output, the inverse transform sampling method is used to generate a large number of dynamic samples that obey the probability distribution of the above steps. The process is as follows:

[0037] The principle formula of inverse transformation is shown as follows:

[0038] P w =F -1 (U)

[0039] Where, F -1 represents the inverse function of the cumulative probability distribution of the wind power output model that satisfies a certain distribution characteristic F; U represents the uniform distribution in the interval [0,1], i.e., U~Unif[0,1]; P w Represents the wind power output power obtained by random sampling.

[0040] In order to simplify the solution process, a function that obeys the standard normal distribution is introduced Will Substitute U, that is The expression is:

[0041]

[0042] From the above analysis, we can see that the inverse transform sampling process is first to generate a large number of random numbers that obey the normal distribution by computer, and then use the distribution function model to solve the inverse function to obtain the wind power output value, and then generate the wind power scenario.

[0043] Step 2 proposes a static scenario generation method based on Copula theory that takes into account power supply correlation. First, the marginal distribution function of historical wind power data is solved using the non-parametric kernel density estimation method.

[0044] Then, the maximum likelihood method is used to solve the variable correlation parameters of the following five Copula functions;

[0045] ①Clayton Copula function

[0046] The distribution function is:

[0047] C(x1,x2)=exp{-[(-lnx1)λ +(-lnx2)λ] 1 / λ},λ∈[0,∞]

[0048] Where x1, x2 represent the values ​​of two marginal distribution functions, and x1, x2∈[0,1]; λ represents the parameter describing the correlation between two random variables;

[0049] The probability density function is:

[0050] C(x1,x2)=exp{-[(-lnx1) λ +(-lnx2)λ] 1 / λ},λ∈[0,∞]

[0051] ②Frank Copula function

[0052] The distribution function is:

[0053]

[0054] The probability density function is:

[0055]

[0056] ③Gumbel Copula function

[0057] The distribution function is:

[0058] C(x1,x2)=exp{-[(-lnx1) 1 / λ +(-lnx2) 1 / λ ] λ},λ∈[1,∞]

[0059] The probability density function is:

[0060]

[0061] ④Normal Copula function

[0062] The distribution function is:

[0063]

[0064] Where, Represents the inverse of the cumulative distribution function of the standard normal distribution, where g is the set of two marginal distribution functions x1 and x2;

[0065] ⑤t-Copula function

[0066] Distribution function:

[0067]

[0068] Where n represents the degree of freedom, t n -1 Inverse the t-distribution function with n degrees of freedom.

[0069] Secondly, calculate the Euclidean distance between each Copula function and the empirical function. The calculation formula is as follows:

[0070] Where d represents the square of the Euclidean distance; C n (x 1i ,x 2i ) is expressed as the empirical Copula distribution function; C(x 1i ,x 2i ) is represented by the five Copula distribution functions mentioned above.

[0071] Secondly, the Copula function with the smallest Euclidean distance is selected as the fitting function;

[0072] Finally, a simple random sampling method is used to generate uniformly distributed random numbers u1, u2, ..., u in the interval [0, 1]. n , where n is the number of static scenes generated; the optimal fitting Copula function C(F1(x1),F2(x2),...,F n (x n )) Discretization, generate n discrete variables v1, v2, ..., v n ; Combined with the inverter sampling method, the wind power scenario x is obtained i =F i -1 (v i ),i=1,2,...,N。

[0073] Step 3 proposes a method for generating dynamic-static coupling scenarios that takes into account power supply volatility, randomness, and correlation;

[0074] First, generate correlated static scenarios. For multiple wind farms, use the Copula function to generate n sets of correlated static output scenarios, expressed as: (x0(m), y0(m)), where m = 1, 2, ..., n, and n is the number of generated scenarios.

[0075] Secondly, generate dynamic scenarios with uncertainty. Generate n groups of dynamic scenarios that take into account the randomness and volatility of wind farm output, expressed as: (x k (m),y k (m)), where m = 1, 2, ..., n, k = 1, 2, ..., K, K is a different time section;

[0076] Next, we dimensionally normalize the dynamic and static scenes. Since the static scenes with correlation generated above are randomly distributed on [0, 1], the distribution of the scenes generated with randomness and volatility is obviously not distributed on [0, 1] after normalization. Therefore, we dimensionally normalize the static scenes so that they are distributed on [0, 1] as the normalized result to support the subsequent scene coupling.

[0077] Finally, generate a coupled scenario that takes into account dynamic and static scenarios. Arrange the two random variables in (x0(m), y0(m)) from small to large at the same time, and record the scenarios after changing the order as x 0c (m), y 0c (m), and then (x0(m), y0(m)) is in the same time section k (m),y k The random variables in (m)) are arranged in the same order as (x0(m), y0(m)), and the scenes after changing the order are recorded as x kc (m), y kc (m). To simplify the calculation, this paper calculates the output scenarios at different time sections in proportion to the above-mentioned sequential scenario ratios, i.e., the scenario coupling coefficients. The calculation formula is as follows:

[0078]

[0079] Where x 0c (m), y 0c (m) represents the static scene after changing the order; x kc (m), y kc (m) represents the dynamic scene after changing the order; x jc (m), y jc (m) represents the dynamic scene generated at time section j; x j (m), y j (m) represents the coupled scenario taking into account dynamic and static scenarios at time section j.

[0080] Step 4: Propose a dynamic scenario generation method that takes into account load characteristics and uncertainty. First, the randomness of load changes is processed. The randomness of load changes is processed using the above-mentioned binning theory and inverse transform sampling. The processing steps are as follows:

[0081] 1) Divide a large number of historical load power predictions into N groups in ascending order with the same numerical scale interval. Then, evenly distribute the actual value corresponding to each predicted value into the N groups and place them into the N prediction boxes with the same scale interval.

[0082] 2) Fit the load power through an appropriate probability distribution model, regard the actual power as the conditional probability distribution of the predicted power, and still use the non-parametric kernel density estimation method mentioned in the previous step for fitting.

[0083] Secondly, load fluctuation processing is performed. The load fluctuation processing is the same as the above-mentioned distributed power generation processing method, which mainly considers the load power fluctuation amount in multiple time sections;

[0084] 1) Based on the above randomness processing, the random process P on the load side is obtained = {P L ,L∈T} T When dealing with fluctuation problems, the load power of multiple time sections can be regarded as a multivariate random vector L = (L1, L2, ..., L K ) T , where K represents the number of time sections. Assume that the multivariate random variable L obeys the multivariate normal distribution L~N(μ L ,∑ L );

[0085] 2) According to the above steps, determine the covariance parameters of the normal distribution of the multivariate random variable, and then obtain the covariance matrix to determine the multivariate normal distribution L~N(μ L ,∑ L ), paving the way for subsequent scene generation;

[0086] 3) Using the Matlab statistics toolbox, generate n numbers of L~N(μ L ,∑ L ) is a random sample of .

[0087] Finally, a scenario generation method considering load fluctuation and randomness is proposed;

[0088] 1) For each time section K of the random sample generated in the above steps, determine the prediction box it is in and obtain the probability distribution function within the corresponding prediction box;

[0089] 2) For the n numbers generated by the above steps that obey the multivariate normal distribution L~N(μ L ,∑ L ) is inversely transformed according to the probability distribution function to generate n scenarios that take into account load fluctuation and randomness.

[0090] Step 5: A source-load dynamic and static scenario reduction method based on the K-means clustering algorithm is proposed; the K-means partitioning clustering algorithm is used to perform a clustering analysis on the scenario. The scenario reduction clustering steps are as follows: (1) Input the original data sample and the required number of clusters K;

[0091] (2) Divide all sample data into K groups, select one random data sample from each group to form K cluster initial centers;

[0092] (3) Calculate the distance between the remaining sample data and each cluster center, and aggregate them into the cluster with the closest cluster center;

[0093] (4) Recalculate the average value of K groups of data samples and update the cluster center;

[0094] (5) Calculate the distance between the remaining samples and the cluster center again and re-cluster the data samples;

[0095] (6) Clustering is completed when the cluster center no longer changes or the criterion function converges. If not, return to step (3).

[0096] The advantages of the present invention are:

[0097] The method proposed in this invention can integrate multi-dimensional factors such as time scale, scenario properties, and research variables. Compared with the traditional method of generating a large number of initial scenarios based on wind power, the wind-electric static scenario coupling method adopted in this chapter introduces the concept of wind power fluctuation quantity on the basis of wind power, effectively reflects the multi-time section correlation of power output, and characterizes the randomness and volatility of wind power output. The introduction of the Copula function, on the basis of the characterization of output randomness and volatility, more comprehensively considers the correlation effect of output between multiple wind farms. The load uncertainty scenario characterization method based on binning theory and inverse transform sampling can generate a large number of effective scenarios that take into account load characteristics, load output randomness and volatility. The proposed initial scenario generation scheme is not constrained by the number of generated scenarios, and can generate a large number of accurate and effective scenarios that take into account the uncertainty of source-load output in a short period of time, with good operability. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 Generate a flow chart for source-side output uncertainty scenarios;

[0099] Figure 2 Generate a flow chart for load-side output uncertainty scenarios. DETAILED DESCRIPTION

[0100] The technical solution of the present invention is described in detail below through embodiments and in conjunction with the accompanying drawings:

[0101] First, the principle of the method of the present invention is introduced, which involves a scenario generation method taking into account the uncertainty of distributed power supply and load output.

[0102] The specific steps include:

[0103] Step 1: A case study was conducted using historical and forecast data from two Belgian wind farms in 2018. Both the measured and forecasted data were processed with a 15-minute time step. All data was normalized to the range [0, 1] pu. Due to the large number of historical data samples, the number of prediction bins was set to 50, with each bin having a data width of 0.02 pu. Dynamic scenarios were generated using the aforementioned formula to account for source-side random fluctuations.

[0104] Step 2: Generate a static scene of a wind farm with correlation based on the copula correlation theory, select the distribution function with the smallest Euclidean distance to the empirical distribution function for fitting, and use the inverse transform sampling method to generate a static scene.

[0105] Step 3: Sort the static scenes from small to large, and at the same time, arrange the dynamic scenes in the time section order obtained after sorting the static scenes, and generate the dynamic-static coupling scene taking into account the random fluctuation correlation on the source side according to the following formula. Sort the two random variables in (x0(m), y0(m)) from small to large at the same time, and record the scenes after changing the order as x 0c (m), y 0c (m), and then (x0(m), y0(m)) is in the same time section k (m),y k The random variables in (m)) are arranged in the same order as (x0(m), y0(m)), and the scenes after changing the order are recorded as x kc (m), y kc To simplify the calculation, the output scenarios at different time sections are calculated proportionally based on the above sequence scenario ratios, i.e., the scenario coupling coefficients. The calculation formula is as follows:

[0106]

[0107] Where x 0c (m), y 0c (m) represents the static scene after changing the order; x kc (m), y kc (m) represents the dynamic scene after changing the order; x jc (m), y jc (m) represents the dynamic scene generated at time section j; x j (m), y j (m) represents the coupled scenario taking into account dynamic and static scenarios at time section j.

[0108] Step 4: Scenario generation method taking into account load volatility, randomness and load characteristics;

[0109] First, input historical data and forecast data, and normalize the data using the following formula:

[0110]

[0111] Secondly, load data is processed by dividing it into industrial load and residential load according to load type;

[0112] Then, the binning theory is introduced to arrange the predicted data from small to large and put them into prediction boxes of the same size on average. The probability distribution function of the actual data corresponding to the predicted data in the box is solved using the non-parametric kernel density estimation method.

[0113] Finally, according to the steps and methods for solving the randomness and volatility of distributed power sources, n groups of scenarios considering the randomness, volatility and load characteristics of load power are obtained.

[0114] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Persons skilled in the art may make various modifications, additions, or substitutions to the described embodiments without departing from the spirit of the present invention or exceeding the scope of the appended claims.

Claims

1. A scenario generation method taking into account the uncertainty of distributed power supply and load output, characterized in that: The following steps are involved: Step 1: Given the strong randomness and volatility of wind power generation, distributed wind power is used as the research object for the uncertainty of power output. Based on the binning theory and inverse transformation sampling method, a multi-time section dynamic scenario that takes into account the randomness and volatility of power output is generated; Step 2: Introduce the Copula function to generate a spatial static scene with correlation; Step 3: Using the scenario ranking method, generate coupled spatiotemporal dynamic and static scenarios that take into account the randomness, volatility, and correlation of power output; Step 4: Continue the above method to deal with the uncertainty of load changes, and combine the load characteristics with the uncertainty to form the corresponding typical scenario; Step 5: Use the K-means clustering algorithm to reduce the multiple scenarios generated above to generate typical scenarios, improve the efficiency of subsequent calculations and scenario processing, and transform the source-load output uncertainty problem into a probabilistic deterministic problem.

2. The scenario generation method taking into account the uncertainty of distributed power supply and load output according to claim 1 is characterized in that: Step 1 First, the power distribution function can be effectively characterized by using the non-parametric kernel density estimation method. The process is as follows. The non-parametric kernel density estimation method is shown in the following formula: Where x represents x i The i random variables in the equation are: k is the sample size; h is the sample bandwidth, whose value affects the function fitting and is always greater than 0; K(x) is the kernel function, which is a probability density function that is always greater than 0, has an integral of 1, and a mean of 0. Its commonly used Gaussian function form is shown in the above formula. Then, since wind speed changes in real time, a large amount of historical data will be generated. The use of "binning" can effectively summarize and integrate a large amount of historical data, making the prediction and solution process time-saving and accurate. The "bin" here refers to the wind power prediction box. The historical wind power data of any time section is composed of the day-ahead power forecast value and the power measured value obtained by sampling on the same day. Each predicted power has a corresponding actual power value, and these actual powers are the data elements in the "bin"; Since the predicted power is arranged in ascending order, the difference between any adjacent data is not large. However, the corresponding actual power will produce large data differences due to the real-time changes in wind speed. This reflects the randomness of wind power generation. By fitting the wind power with an appropriate probability distribution model and treating the actual power as the conditional probability distribution of the predicted power, the randomness of wind power generation can be better solved and analyzed. Secondly, the concept of multivariate normal distribution is introduced to further analyze the volatility of wind power output; The random output process P of the power supply side obtained based on the above random output processing is P = {P w ,w∈T} T When dealing with output fluctuation problems, the wind power of multiple time sections can be regarded as a multivariate random vector Y = (Y1, Y2, ..., Y K ) T , where K represents the number of time sections. Assume that the multivariate random variable Y obeys the multivariate normal distribution Y~N(μ,Σ), where μ represents the expectation and is a K-dimensional zero vector, and Σ is the covariance matrix. The expression is as follows: Where, δ i,j =cov(Y i ,Y j ) represents Y i ,Y j The covariance between The covariance in the above formula reflects the correlation between wind power output power at different time sections i and j, and the covariance matrix formed by the covariance can characterize the volatility of wind power output. From the above analysis, it can be seen that the key to generating dynamic scenarios that consider both the randomness and volatility of wind power output lies in the covariance matrix [δ i,j ] K×K ; The covariance is solved as follows: Where ε represents the range parameter, which is used to adjust the strength of the power correlation between different time sections i and j; From the above formula, we can see that the farther the distance between different time sections is, the larger |ij| is, the greater the covariance δ i,j The smaller the value, the weaker the power correlation between time sections i and j; conversely, the closer the distance between different time sections, that is, the smaller |ij|, the smaller the covariance δ i,j The larger the value, the stronger the power correlation between time sections i and j. The change of the model is consistent with the actual situation, so the covariance δ can be effectively solved. i,j ; Assume the objective function f(ε) of the covariance range parameter ε is: Where N is the number of sampling populations; s represents the equally spaced sampling points; S represents the fluctuation range of wind power; pdf(s) represents the probability density function obtained by fitting the fluctuation of wind power in a randomly generated dynamic scenario; pdf'(s) represents the probability density function obtained by fitting the fluctuation of historical wind power data. The wind power fluctuation ΔP is defined as: ΔP=P t -P t+1 Where, P t is the wind power at time t, P t+1 is the wind power at time t+1; Since the probability of extreme wind power in the t location-scale distribution is greater than that in the normal distribution and is more consistent with the actual output power characteristics of wind power, this section uses the t location-scale distribution function to characterize short-term wind power fluctuations. The probability density function of the t location-scale distribution is as follows: In the formula, μ represents the location parameter; k represents the scale parameter; v represents the shape parameter; pdf(s) and pdf'(s) represent the probability density functions corresponding to the location-scale distribution function, and the solution to the randomly generated dynamic scene wind power fluctuation is as follows: when the ε value is determined, the covariance matrix [δ i,j ] K×K is also determined at the same time, then the corresponding multivariate standard normal distribution random vector Y=(Y1,Y2,…,Y K ) T , and then use the inverse transformation to perform random sampling to generate n day-ahead wind power dynamic scenarios. For each dynamic scenario, the wind power dynamic scenario fluctuation data can also be obtained by solving. Combining the above data, the covariance range parameter ε can be obtained, which more accurately characterizes the volatility of wind power output. Finally, in order to better analyze the randomness of wind power output, the inverse transform sampling method is used to generate a large number of dynamic samples that obey the probability distribution of the above steps. The process is as follows: The principle formula of inverse transformation is shown as follows: P w =F -1 (U) Where, F -1 represents the inverse function of the cumulative probability distribution of the wind power output model that satisfies a certain distribution characteristic F; U represents the uniform distribution in the interval [0,1], i.e., U~Unif[0,1]; P w represents the wind power output power obtained by random sampling; In order to simplify the solution process, a function that obeys the standard normal distribution is introduced Will Substitute U, that is The expression is: From the above analysis, we can see that the inverse transform sampling process is first to generate a large number of random numbers that obey the normal distribution by computer, and then use the distribution function model to solve the inverse function to obtain the wind power output value, and then generate the wind power scenario.

3. The scenario generation method taking into account the uncertainty of distributed power supply and load output according to claim 1 is characterized in that: Step 2 proposes a static scenario generation method based on Copula theory taking into account power supply correlation; Firstly, the marginal distribution function of historical wind power data is solved using nonparametric kernel density estimation method; Then, the maximum likelihood method is used to solve the variable correlation parameters of the following five Copula functions; ①Clayton Copula function The distribution function is: C(x1,x2)=exp{-[(-lnx1) λ +(-lnx2)λ] 1 / λ },λ∈[0,∞] Where x1, x2 represent the values ​​of two marginal distribution functions, and x1, x2∈[0,1]; λ represents the parameter describing the correlation between two random variables; The probability density function is: C(x1,x2)=exp{-[(-lnx1) λ +(-lnx2)λ] 1 / λ },λ∈[0,∞] ②Frank Copula function The distribution function is: The probability density function is: ③Gumbel Copula function The distribution function is: C(x1,x2)=exp{-[(-lnx1) 1 / λ +(-lnx2) 1 / λ ] λ },λ∈[1,∞] The probability density function is: ④Normal Copula function The distribution function is: Where, Represents the inverse of the cumulative distribution function of the standard normal distribution, where g is the set of two marginal distribution functions x1 and x2; ⑤t-Copula function Distribution function: Where n represents the degree of freedom, t n -1 Inverse of the t-distribution function representing n degrees of freedom; Secondly, calculate the Euclidean distance between each Copula function and the empirical function. The calculation formula is as follows: Where d represents the square of the Euclidean distance; C n (x 1i ,x 2i ) is expressed as the empirical Copula distribution function; C(x 1i ,x 2i ) is expressed as the five Copula distribution functions mentioned above; Secondly, the Copula function with the smallest Euclidean distance is selected as the fitting function; Finally, a simple random sampling method is used to generate uniformly distributed random numbers u1, u2, ..., u in the interval [0, 1]. n , where n is the number of static scenes generated; the optimal fitting Copula function C(F1(x1),F2(x2),...,F n (x n )) Discretization, generate n discrete variables v1, v2, ..., v n ; Combined with the inverter sampling method, the wind power scenario x is obtained i =F i -1 (v i ),i=1,2,...,N。 4. The scenario generation method taking into account the uncertainty of distributed power supply and load output according to claim 1 is characterized in that: Step 3 proposes a method for generating dynamic-static coupling scenarios that takes into account power supply volatility, randomness, and correlation; First, generate correlated static scenarios. For multiple wind farms, use the Copula function to generate n sets of correlated static output scenarios, expressed as: (x0(m), y0(m)), where m = 1, 2, ..., n, and n is the number of generated scenarios. Secondly, generate dynamic scenarios with uncertainty, and generate n groups of dynamic scenarios that take into account the randomness and volatility of wind farm output, expressed as: (x k (m),y k (m)), where m = 1, 2, ..., n, k = 1, 2, ..., K, K is a different time section; Then, the dynamic and static scenes are dimensioned. Since the static scenes with correlation generated above are randomly distributed on [0, 1], and the scenes generated considering randomness and volatility are obviously not distributed on [0, 1] after normalization, the static scenes are dimensioned so that they are distributed on [0, 1] as the normalized result to support the subsequent scene coupling. Finally, a coupled scenario is generated that takes into account dynamic and static scenarios. The two random variables in (x0(m), y0(m)) are arranged from small to large at the same time. The scenarios after changing the order are recorded as x 0c (m), y 0c (m), and then (x0(m), y0(m)) is in the same time section k (m),y k The random variables in (m)) are arranged in the same order as (x0(m), y0(m)), and the scenes after changing the order are recorded as x kc (m), y kc (m), in order to simplify the calculation, this paper calculates the output scenarios of different time sections in proportion to the above-mentioned sequential scenario ratios, i.e., the scenario coupling coefficients. The calculation formula is as follows: Where x 0c (m), y 0c (m) represents the static scene after changing the order; x kc (m), y kc (m) represents the dynamic scene after changing the order; x jc (m), y jc (m) represents the dynamic scene generated at time section j; x j (m), y j (m) represents the coupled scenario taking into account dynamic and static scenarios at time section j.

5. The scenario generation method taking into account the uncertainty of distributed power supply and load output according to claim 1 is characterized in that: In step 4, a dynamic scenario generation method is proposed that takes into account the load characteristics and uncertainties of the load side; First, the randomness of load changes is processed. The above-mentioned binning theory and inverse transformation sampling are used to process the randomness of load changes. The processing steps are as follows: 1) Divide a large number of historical load power predictions into N groups in ascending order with the same numerical scale interval. Then, evenly distribute the actual value corresponding to each predicted value into the N groups and place them into the N prediction boxes with the same scale interval. 2) Fit the load power through an appropriate probability distribution model, and regard the actual power as the conditional probability distribution of the predicted power. The non-parametric kernel density estimation method mentioned in the previous step is still used for fitting; Secondly, load fluctuation processing is carried out. The load fluctuation processing is the same as the distributed power generation processing method mentioned above, and the main consideration is the load power fluctuation amount in multiple time sections. 1) Based on the above randomness processing, the random process P on the load side is obtained = {P L ,L∈T} T When dealing with fluctuation problems, the load power of multiple time sections can be regarded as a multivariate random vector L = (L1, L2, ..., L K ) T , where K represents the number of time sections, and the multivariate random variable L obeys the multivariate normal distribution L~N(μ L ,∑ L ); 2) According to the above steps, determine the covariance parameters of the normal distribution of the multivariate random variable, and then obtain the covariance matrix to determine the multivariate normal distribution L~N(μ L ,∑ L ), paving the way for subsequent scene generation; 3) Using the Matlab statistics toolbox, generate n numbers of L~N(μ L ,∑ L ) a random sample of Finally, a scenario generation method considering load fluctuation and randomness is proposed; 1) For each time section K of the random sample generated in the above steps, determine the prediction box it is in and obtain the probability distribution function within the corresponding prediction box; 2) For the n numbers generated by the above steps that obey the multivariate normal distribution L~N(μ L ,∑ L ) is inversely transformed according to the probability distribution function to generate n scenarios that take into account load fluctuation and randomness.

6. The scenario generation method taking into account the uncertainty of distributed power supply and load output according to claim 1, characterized in that: Step 5 proposes a source load dynamic and static scenario reduction method based on K-means clustering algorithm; The K-means partitioning and clustering algorithm is used to perform cluster analysis on the scenes. The steps of scene reduction clustering are as follows: (1) Input the original data sample and the required number of clusters K; (2) Divide all sample data into K groups, select one random data sample from each group to form K cluster initial centers; (3) Calculate the distance between the remaining sample data and each cluster center, and aggregate them into the cluster with the closest cluster center; (4) Recalculate the average value of K groups of data samples and update the cluster center; (5) Calculate the distance between the remaining samples and the cluster center again and re-cluster the data samples; (6) Clustering is completed when the cluster center no longer changes or the criterion function converges. If not, return to step (3).

Images