New energy prediction error sample generation method and system based on sample deletion technology

Abstract

The invention discloses a new energy prediction error sample generation method and system based on a sample deletion technology. The method comprises the following steps: establishing a new energy prediction error distribution model by using a GMM (Gaussian Mixture Model) method; generating a new energy prediction error sample according to the new energy prediction error distribution model; according to a Gaussian component with a negative weight coefficient in the new energy prediction error distribution model, carrying out sample deletion on the new energy prediction error sample to obtain a deleted new energy prediction error sample; the method has the advantage that the new energy prediction error sample can be quickly generated.

Description

Technical Field

[0001] The present invention relates to the field of new energy probability simulation, and particularly to a method and system for generating new energy prediction error samples based on sample deletion technology. Background Art

[0002] The proportion of new energy represented by wind power and photovoltaic power in the power structure has been increasing year by year. However, the uncertainty of new energy output also brings severe challenges to the stable operation of the power system. In order to accurately measure the uncertainty of new energy, it is usually necessary to establish the difference between the actual output and the predicted output of new energy, that is, the probability distribution model of new energy output prediction error. Based on the probability distribution model, samples of new energy prediction error are obtained by sampling. Using the samples of new energy prediction error, the reliability level of the power system containing new energy, the spinning reserve demand, etc. can be evaluated.

[0003] Accurately generating new energy prediction error samples can clarify the impact of the new energy prediction error probability distribution on the operating characteristics of the power system. At the same time, the efficiency of sample generation directly affects the efficiency of the operating characteristics analysis of the power system containing new energy. Therefore, it is necessary to accurately and efficiently generate prediction error samples.

[0004] Common probability modeling methods for new energy prediction error include distribution function fitting methods, such as normal distribution, Beta distribution, etc.; and non-parametric modeling methods, such as Gaussian mixture model (GMM), kernel density estimation, support vector machine, etc.

[0005] Based on the GMM method, establishing the probability distribution model of new energy prediction error has received extensive attention in the academic and industrial fields. The sampling method that can quickly and conveniently sample and generate a large number of samples by combining the sampling of each Gaussian component respectively is recorded in the literature "Zhao Yuan, Wang Jie, Xiong Yanjiao, etc. Gaussian mixture model of random variables in power grid reliability assessment [J]. Automation of Electric Power Systems, 2016, 40(1): 66-71.". However, the conventional GMM method is limited by the algorithm definition, and the accuracy of probability modeling is limited. For this reason, experts and scholars have introduced the concept of negative weights, allowing the weight coefficients of some Gaussian components to be negative, which improves the accuracy of probability modeling. However, the introduction of negative weights makes it necessary to solve the inverse function of the complex probability distribution function many times during the sampling process, resulting in a slow and cumbersome sampling process, making the GMM method unable to achieve rapid sample generation. Summary of the Invention

[0006] The technical problem to be solved by the present invention is that the existing GMM method cannot achieve rapid generation of new energy prediction error samples.

[0007] The present invention solves the above technical problems by the following technical means: A method for generating new energy prediction error samples based on sample deletion technology, including:

[0008] S1. Establish a new - energy prediction error distribution model using the GMM method;

[0009] S2. Generate new - energy prediction error samples according to the new - energy prediction error distribution model;

[0010] S3. Carry out sample deletion on the new - energy prediction error samples according to the Gaussian components with negative weight coefficients in the new - energy prediction error distribution model to obtain the new - energy prediction error samples after deletion.

[0011] The present invention makes full use of the modeling accuracy of the generalized GMM method. According to the new - energy prediction error distribution model, new - energy prediction error samples are generated. Then, according to the Gaussian components with negative weight coefficients in the new - energy prediction error distribution model, sample deletion is carried out on the new - energy prediction error samples, so that it is not necessary to repeatedly solve the inverse function of a complex distribution function, solving the problem of repeatedly solving the inverse function of a complex distribution function in the sample generation process of the generalized GMM method, improving the sampling efficiency, and realizing the rapid generation of new - energy prediction error samples.

[0012] Further, S1 includes:

[0013] Let the number of Gaussian components be M + N, the weight coefficient of the k - th Gaussian component be α k , the mean be μ k , the standard deviation be σ k . The weight coefficients of the first to the M - th Gaussian components are positive, and the weight coefficients of the (M + 1) - th to the (M + N) - th Gaussian components are negative. The new - energy prediction error distribution model established using the GMM method is

[0014]

[0015] where g k (x|μ k ,σ k ) is the k - th Gaussian component and x represents the numerical value of the new - energy prediction error, and f(x) represents the probability density function of the new - energy prediction error, that is, the new - energy prediction error distribution model.

[0016] Furthermore, the new - energy prediction error distribution model satisfies the following relationship:

[0017] -∞ < α k < +∞

[0018]

[0019] Furthermore, S2 includes:

[0020] Let the length of the new - energy prediction error sample be \(H\). For the \(k\) - th Gaussian component, where \(k = 1,\cdots,M\), allocate the sample number \(L\). k , \(L\) k =\(\alpha\) k \(H\). Generate a Gaussian - component sample of length \(\alpha\) k \(H\) according to the mean \(\mu\) k and the standard deviation \(\sigma\) k . Denote it as the \(k\) - th Gaussian - component sample \(S\) k ; Combine the Gaussian - component samples to obtain the new - energy prediction error sample \(S'\), \(S'=\{S\) 1 ,\cdots,S\) k ,\cdots,S\) M \}, \(k = 1,\cdots,M\). Denote the elements in \(S'\) as \(S'\) i ,

[0021] Furthermore, the method for generating the Gaussian - component sample is: According to the Gaussian components with positive weight coefficients in the new - energy prediction error distribution model, use the normrnd function in matlab, the standard_normal function in the numpy extension library of Python, or the std::normal_distribution function in C++ to generate the Gaussian - component sample.

[0022] Furthermore, \(S3\) includes:

[0023] \(S31\): Determine the sample - deletion intervals respectively in the case of only 1 negative - coefficient Gaussian component and the case of multiple negative - coefficient Gaussian components, and calculate the number of negative - coefficient samples corresponding to the sample - deletion intervals;

[0024] \(S32\): First, establish a deletion counter \(D\) t , \(t = 1,\cdots,T\). The initial value of \(D\) t is taken as 0. Establish a set \(S\) of samples after deletion, and its elements are empty; where \(T\) represents the total number of deletion intervals. Then, traverse each element \(S'\) in the new - energy prediction error sample \(S'\) obtained from \(S2\). i Judge whether \(S'\) i is located in the \(t\) - th deletion interval and whether the value of the deletion counter of the \(t\) - th deletion interval is less than or equal to the number of negative - coefficient samples; If both conditions are met, the value of \(D\) t is incremented by 1, and the element \(S'\) i will not be stored in the set \(S\) of samples after deletion. If the value of \(D\) t exceeds the number of negative - coefficient samples, or the element \(S'\) i does not belong to any deletion interval, then the element \(S'\) iStore the pruned sample set S. In this way, the obtained pruned sample set S is the pruned new energy prediction error sample.

[0025] Furthermore, S31 includes:

[0026] If there is only one Gaussian component with a negative coefficient, i.e., N = 1, then for the (M + N)-th Gaussian component, the lower and upper bounds a0 and b0 of its corresponding distribution range are expressed as

[0027]

[0028] Evenly divide the distribution range into T pruning intervals. Then the lower bound a of the t-th pruning interval t and the upper bound b t are respectively

[0029]

[0030] The number of negative coefficient samples corresponding to the t-th pruning interval is

[0031]

[0032] Furthermore, S31 also includes:

[0033] If there are multiple Gaussian components with negative coefficients, i.e., N ≥ 2, then for the (M + 1)-th to the (M + N)-th negative coefficient Gaussian components, the lower and upper bounds a0 and b0 of their superimposed distribution range are expressed as

[0034]

[0035] Evenly divide the superimposed distribution range into T pruning intervals. Then the lower bound a of the t-th pruning interval t and the upper bound b t are respectively

[0036]

[0037] The number of negative coefficient samples corresponding to the t-th pruning interval is

[0038]

[0039] The present invention also provides a new energy prediction error sample generation system based on the sample pruning technology, including:

[0040] A model construction module for establishing a new energy prediction error distribution model using the GMM method;

[0041] A sample generation module for generating new energy prediction error samples according to the new energy prediction error distribution model;

[0042] A sample deletion module, configured to perform sample deletion on new energy prediction error samples according to Gaussian components with negative weight coefficients in the new energy prediction error distribution model, so as to obtain the new energy prediction error samples after deletion.

[0043] Further, the model construction module is further configured to:

[0044] Assume that the number of Gaussian components is M + N, and the weight coefficient α of the k-th Gaussian component k , mean μ k , standard deviation σ k . The weight coefficients of the first to the M-th Gaussian components are positive, and the weight coefficients of the (M + 1)-th to the (M + N)-th Gaussian components are negative. The new energy prediction error distribution model established using the GMM method is

[0045]

[0046] where g k (x|μ k , σ k ) is the k-th Gaussian component and x represents the numerical value of the new energy prediction error.

[0047] Even further, the new energy prediction error distribution model satisfies the following relationship:

[0048] -∞ < α k < +∞

[0049]

[0050] Even further, the sample generation module is further configured to:

[0051] Assume that the length of the new energy prediction error sample is H. For the k-th Gaussian component, where k = 1,..., M, allocate the sample number L k , L k = α k H. Generate a Gaussian component sample with a length of α k H according to the mean μ k and the standard deviation σ k H, denoted as the k-th Gaussian component sample S k ; Combine the Gaussian component samples to obtain the new energy prediction error sample S`, S` = {S 1 ,..., S k ,..., S M}, k = 1,..., M. The elements in S` are denoted as S` i ,

[0052] Further, the method for generating the Gaussian component samples is as follows: According to the Gaussian components with positive weight coefficients in the new energy prediction error distribution model, use the normrnd function in matlab, the standard_normal function in the pythond extension library numpy, or the std::normal_distribution function in C++ to generate Gaussian component samples.

[0053] Further, the sample deletion module is also used for:

[0054] S31. Determine the sample deletion intervals respectively in the case of only one Gaussian component with a negative coefficient and in the case of multiple Gaussian components with negative coefficients, and calculate the number of negative coefficient samples corresponding to the sample deletion intervals;

[0055] S32. First, establish a deletion counter D for each sample deletion interval t , t = 1, …, T, the initial value of D t is taken as 0, establish a set S of samples after deletion, and its elements are empty; where T represents the total number of deletion intervals; then, traverse each element S` in the new energy prediction error sample S` obtained by the sample generation module i , and judge whether S` i is located in the t-th deletion interval and whether the value of the deletion counter of the t-th deletion interval is less than or equal to the number of negative coefficient samples; if both conditions are met, the value of D t is incremented by 1, and the element S` i will not be stored in the set S of samples after deletion. If the value of D t exceeds the number of negative coefficient samples, or the element S` i does not belong to any deletion interval, then the element S` i is stored in the set S of samples after deletion. In this way, the obtained set S of samples after deletion is the new energy prediction error sample after deletion.

[0056] Further, S31 includes:

[0057] If there is only one Gaussian component with a negative coefficient, that is, N = 1, then for the (M + N)-th Gaussian component, represent the upper and lower bounds a0 and b0 of its corresponding distribution range as

[0058]

[0059] Evenly divide the distribution range into T deletion intervals, then the lower bound a t and the upper bound b t of the t-th deletion interval are respectively

[0060]

[0061] The number of negative coefficient samples corresponding to the t-th deletion interval is

[0062]

[0063] Furthermore, S31 further includes:

[0064] If there are multiple negative coefficient Gaussian components, that is, N≥2, then for the (M + 1)-th to the (M + N)-th negative coefficient Gaussian components, the upper and lower bounds a0 and b0 of their superposition distribution range are expressed as

[0065]

[0066] The superposition distribution range is evenly divided into T deletion intervals, then the lower bound a t and the upper bound b t of the t-th deletion interval are respectively

[0067]

[0068] The number of negative coefficient samples corresponding to the t-th deletion interval is

[0069]

[0070] The advantages of the present invention are as follows: The present invention makes full use of the modeling accuracy of the generalized GMM method. According to the new energy prediction error distribution model, new energy prediction error samples are generated. Then, according to the Gaussian components with negative weight coefficients in the new energy prediction error distribution model, sample deletion is carried out on the new energy prediction error samples, so that it is not necessary to repeatedly solve the inverse function of the complex distribution function, solving the problem of repeatedly solving the inverse function of the complex distribution function in the sample generation process of the generalized GMM method, improving the sampling efficiency, and realizing the rapid generation of new energy prediction error samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a sample deletion flow chart in the new energy prediction error sample generation method based on the sample deletion technology disclosed in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0073] Embodiment 1

[0074] Embodiment 1 of the present invention provides a new energy prediction error sample generation method based on a sample reduction technique, including the following steps:

[0075] S1. Use the GMM method to establish a new energy prediction error distribution model; the specific process is as follows:

[0076] Use the generalized GMM method to establish a new energy prediction error distribution model. Let the number of Gaussian components be M + N, and the weight coefficient α of the kth Gaussian component k , mean μ k , standard deviation σ k ; among them, the weight coefficients of the first to the Mth Gaussian components are positive, and the weight coefficients of the M + 1 to the M + Nth Gaussian components are negative. Then the probability distribution function f(x) of the new energy prediction error can be expressed as a linear combination of several Gaussian components g k (x|μ k , σ k )

[0077]

[0078] And the new energy prediction error distribution model satisfies the following relationship

[0079] -∞ < α k < +∞ (2)

[0080]

[0081] S2. According to the new energy prediction error distribution model established by the generalized GMM, generate new energy prediction error samples that follow a specific distribution. Let the sample length be H. The specific process is as follows:

[0082] S21. Generate component samples that follow a Gaussian distribution according to the Gaussian components with positive weight coefficients

[0083] Gaussian sample generation. Existing programming languages generally come with Gaussian random number generation functions, such as the normrnd function in Matlab, the standard_normal function in the Python extension library NumPy, the std::normal_distribution function in C++, etc. The present invention can quickly realize the generation of Gaussian samples based on existing programs.

[0084] S22. For the kth Gaussian component, where k = 1,..., M, allocate the sample quantity L k , L k = α k H, and generate Gaussian component samples with a length of α k H according to the mean μ k , standard deviation σ k H, denoted as Sk Combine k component samples to obtain the overall sample S`, S` = {S 1 , …, S k , …, S M}, k = 1, …, M. The elements in S` are denoted as S` i ,

[0085] S3. Based on the Gaussian components with negative weight coefficients in the new - energy prediction error distribution model, conduct sample deletion on the new - energy prediction error samples to obtain the new - energy prediction error samples after deletion; the specific process is: Based on the Gaussian components with negative weight coefficients, conduct sample deletion on the overall sample to obtain the new - energy prediction error samples that follow a specific distribution.

[0086] S31. Determine the sample deletion interval and the number of samples with negative coefficients

[0087] (1) Single Gaussian component with a negative coefficient

[0088] If there is only 1 Gaussian component with a negative coefficient, that is, N = 1, then for the (M + N)-th Gaussian component, the upper and lower bounds a0 and b0 of its corresponding distribution range can be expressed as

[0089]

[0090] Divide the distribution range evenly into T deletion intervals, then the lower bound a t and the upper bound b t of the t - th deletion interval are respectively

[0091]

[0092] The number of samples with negative coefficients corresponding to the t - th deletion interval is

[0093]

[0094] (2) Multiple Gaussian components with negative coefficients

[0095] If there are multiple Gaussian components with negative coefficients, that is, N≥2, then for the (M + 1)-th to the (M + N)-th Gaussian components with negative coefficients, the upper and lower bounds a0 and b0 of their superimposed distribution range can be expressed as

[0096]

[0097] Divide the superimposed distribution range evenly into T deletion intervals, then the lower bound a t and the upper bound b t of the t - th deletion interval are respectively

[0098]

[0099] The number of negative coefficient samples corresponding to the t-th deletion interval is

[0100]

[0101] S32. Perform the sample deletion operation

[0102] First, establish a deletion counter D for each deletion interval t , where t = 1, …, T. The initial value of D t is taken as 0. Establish a set S of samples after deletion, and its elements are empty.

[0103] Next, traverse each element S` in the overall sample S` obtained from S2 i , and determine whether S` i is located in the t-th deletion interval, that is, S` i ∈[a t , b t ), and whether the value of the deletion counter for the t-th deletion interval is less than or equal to the number of negative coefficient samples, that is, D t ≤L t .

[0104] If both conditions are satisfied, the value of D t is incremented by 1, and the element S` i will not be stored in the set S of samples after deletion. If the value of D t exceeds L t , or the element S` i does not belong to any deletion interval, then the element S` i is stored in the set S of samples after deletion. In this way, the obtained set S of samples after deletion is the new energy prediction error samples that follow a specific distribution. The specific deletion steps are as Figure 1 shown.

[0105] Through the above technical solutions, the present invention proposes a method for quickly generating new energy prediction error samples for the generalized GMM method with negative coefficient Gaussian components. On the one hand, it makes full use of the modeling accuracy of the generalized GMM method. On the other hand, it solves the problem of repeatedly solving the inverse function of a complex distribution function in the sample generation process of the generalized GMM method, improving the sampling efficiency.

[0106] Embodiment 2

[0107] Based on Embodiment 1, Embodiment 2 of the present invention further provides a new energy prediction error sample generation system based on the sample deletion technology, including:

[0108] A model construction module for using the GMM method to establish a new energy prediction error distribution model;

[0109] A sample generation module for generating new energy prediction error samples according to the new energy prediction error distribution model;

[0110] A sample deletion module, configured to perform sample deletion on new - energy prediction error samples according to Gaussian components with negative weight coefficients in the new - energy prediction error distribution model, so as to obtain the new - energy prediction error samples after deletion.

[0111] Specifically, the model construction module is further configured to:

[0112] Assume that the number of Gaussian components is M + N, the weight coefficient of the k - th Gaussian component is α k , the mean value is μ k , and the standard deviation is σ k . The weight coefficients of the 1st to the M - th Gaussian components are positive, and the weight coefficients of the (M + 1) - th to the (M + N) - th Gaussian components are negative. The new - energy prediction error distribution model established using the GMM method is

[0113]

[0114] where g k (x|μ k ,σ k ) is the k - th Gaussian component and x represents the numerical value of the new - energy prediction error.

[0115] More specifically, the new - energy prediction error distribution model satisfies the following relationship:

[0116] -∞ < α k < + ∞

[0117]

[0118] More specifically, the sample generation module is further configured to:

[0119] Assume that the length of the new - energy prediction error sample is H. For the k - th Gaussian component, where k = 1,…,M, allocate the sample number L k , L k =α k H. Generate a Gaussian component sample with a length of α k , standard deviation σ k H according to the mean value μ k H, denoted as the k - th Gaussian component sample S k ; Combine the Gaussian component samples to obtain the new - energy prediction error sample S`, S`={S 1 ,…,S k ,…,S M}, k = 1,…,M. The elements in S` are denoted as S` i ,

[0120] More specifically, the method for generating the Gaussian component samples is as follows: According to the Gaussian components with positive weight coefficients in the new energy prediction error distribution model, use the normrnd function in Matlab, the standard_normal function in the Python extension library numpy, or the std::normal_distribution function in C++ to generate Gaussian component samples.

[0121] More specifically, the sample deletion module is further used for:

[0122] S31. Determine the sample deletion intervals respectively in the case of only one Gaussian component with a negative coefficient and in the case of multiple Gaussian components with negative coefficients, and calculate the number of negative coefficient samples corresponding to the sample deletion intervals;

[0123] S32. First, establish a deletion counter D for each sample deletion interval t , t = 1, …, T, the initial value of D t is taken as 0, establish a set S of samples after deletion, and its elements are empty; where T represents the total number of deletion intervals; then, traverse each element S' in the new energy prediction error sample S' obtained by the sample generation module i , and judge whether S' i is located in the t-th deletion interval and whether the value of the deletion counter of the t-th deletion interval is less than or equal to the number of negative coefficient samples; if both conditions are met, the value of D t is incremented by 1, and the element S' i will not be stored in the set S of samples after deletion. If the value of D t exceeds the number of negative coefficient samples, or the element S' i does not belong to any deletion interval, then the element S' i is stored in the set S of samples after deletion. In this way, the obtained set S of samples after deletion is the new energy prediction error sample after deletion.

[0124] More specifically, S31 includes:

[0125] If there is only one Gaussian component with a negative coefficient, that is, N = 1, then for the (M + N)-th Gaussian component, represent the upper and lower bounds a0 and b0 of its corresponding distribution range as

[0126]

[0127] Evenly divide the distribution range into T deletion intervals, then the lower bound a t and the upper bound b t of the t-th deletion interval are respectively

[0128]

[0129] The number of negative coefficient samples corresponding to the t-th deletion interval is

[0130]

[0131] More specifically, S31 further includes:

[0132] If there are multiple negative coefficient Gaussian components, i.e., N≥2, then for the (M + 1)-th to the (M + N)-th negative coefficient Gaussian components, the upper and lower bounds a0 and b0 of their superposition distribution range are expressed as

[0133]

[0134] If the superposition distribution range is evenly divided into T deletion intervals, then the lower bound a of the t-th deletion interval t and the upper bound b t are respectively

[0135]

[0136] The number of negative coefficient samples corresponding to the t-th deletion interval is

[0137]

[0138] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A new energy prediction error sample generation method based on sample deletion technology, characterized in that: include: S1. Use GMM method to establish new energy prediction error distribution model; S2. Generate new energy prediction error samples according to the new energy prediction error distribution model; S3. According to the Gaussian component with a negative weight coefficient in the new energy prediction error distribution model, sample deletion is performed on the new energy prediction error samples to obtain the deleted new energy prediction error samples.

2. The new energy prediction error sample generation method based on sample deletion technology according to claim 1 is characterized in that: S1 includes: Assume that the number of Gaussian components is M+N, and the weight coefficient of the kth Gaussian component is α k , mean μ k , standard deviation σ k , the weight coefficients of the 1st to Mth Gaussian components are positive, and the weight coefficients of the M+1th to M+Nth Gaussian components are negative. The new energy prediction error distribution model established using the GMM method is: Among them, g k (x|μ k ,σ k ) is the kth Gaussian component and x represents the numerical value of the new energy prediction error.

3. The new energy prediction error sample generation method based on sample deletion technology according to claim 2 is characterized in that: The new energy prediction error distribution model satisfies the following relationship: -∞<α k <+∞ 4. The new energy prediction error sample generation method based on sample deletion technology according to claim 2 is characterized in that S2 include: Assume that the length of the new energy prediction error sample is H, and for the kth Gaussian component, where k = 1,…,M, the number of samples allocated is L k , L k =α k H, according to the mean μ k , standard deviation σ k Generate a length of α k The Gaussian component sample of H is recorded as the kth Gaussian component sample S k ; Combine the Gaussian component samples to obtain the new energy prediction error sample S`, S` = {S 1 ,…,S k ,…,S M }, k = 1,…,M, the elements in S` are denoted as S` i , 5. The new energy prediction error sample generation method based on sample deletion technology according to claim 4 is characterized in that: The method for generating the Gaussian component samples is: according to the Gaussian components with positive weight coefficients in the new energy prediction error distribution model, the Gaussian component samples are generated using the normrnd function in matlab, the standard_normal function in the pythond extension library numpy, or the std::normal_distribution function in C++.

6. The new energy prediction error sample generation method based on sample deletion technology according to claim 4 is characterized in that S3 include: S31, determining a sample deletion interval in a case where there is only one negative coefficient Gaussian component and in a case where there are multiple negative coefficient Gaussian components, and calculating the number of negative coefficient samples corresponding to the sample deletion interval; S32, first establish a deletion counter D for each sample deletion interval t , t=1,…,T,D t The initial value of is 0, and a pruned sample set S is established, whose elements are empty; where T represents the total number of pruned intervals; then, each element S' in the new energy prediction error sample S' obtained by traversing S2 is i , judge S` i Whether it is in the tth deletion interval, and whether the deletion counter value of the tth deletion interval is less than or equal to the number of negative coefficient samples; if both conditions are met at the same time, then D t The value is increased by 1, and the element S` i Will not be stored in the pruned sample set S if D t The value exceeds the number of negative coefficient samples, or the element S` i If it does not belong to any deletion interval, then the element S` i The pruned sample set S is stored, and thus the pruned sample set S obtained is the pruned new energy prediction error sample.

7. The new energy prediction error sample generation method based on sample deletion technology according to claim 6 is characterized in that: S31 includes: If there is only one negative coefficient Gaussian component, that is, N = 1, then for the M+Nth Gaussian component, the corresponding distribution range upper and lower bounds a0 and b0 are expressed as The distribution range is evenly divided into T deletion intervals, then the lower bound a of the tth deletion interval is t With upper bound b t They are The number of negative coefficient samples corresponding to the tth deletion interval is 8. The new energy prediction error sample generation method based on sample deletion technology according to claim 6 is characterized in that: The S31 also includes: If there are multiple negative coefficient Gaussian components, that is, N ≥ 2, then for the M+1th to M+Nth negative coefficient Gaussian components, the upper and lower bounds a0 and b0 of the superposition distribution range are expressed as The superposition distribution range is evenly divided into T deletion intervals, then the lower bound a of the tth deletion interval is t With upper bound b t They are The number of negative coefficient samples corresponding to the tth deletion interval is 9. A new energy prediction error sample generation system based on sample deletion technology, characterized in that: include: Model building module, used to build new energy forecast error distribution model using GMM method; A sample generation module, used to generate new energy prediction error samples according to a new energy prediction error distribution model; The sample deletion module is used to perform sample deletion on the new energy prediction error samples according to the Gaussian components with negative weight coefficients in the new energy prediction error distribution model to obtain the deleted new energy prediction error samples.

10. The new energy prediction error sample generation system based on sample deletion technology according to claim 9, characterized in that: The model building module is also used to: Assume that the number of Gaussian components is M+N, and the weight coefficient of the kth Gaussian component is α k , mean μ k , standard deviation σ k , the weight coefficients of the 1st to Mth Gaussian components are positive, and the weight coefficients of the M+1th to M+Nth Gaussian components are negative. The new energy prediction error distribution model established using the GMM method is: Among them, g k (x|μ k ,σ k ) is the kth Gaussian component and x represents the numerical value of the new energy prediction error.

Images