CN113468811A - Power grid reserve capacity probabilistic dynamic evaluation method, system, terminal and readable storage medium containing new energy unit - Google Patents

Abstract

The invention discloses a probabilistic dynamic evaluation method, a probabilistic dynamic evaluation system, a probabilistic dynamic evaluation terminal and a readable storage medium for the reserve capacity of a power grid containing a new energy unit. The method comprises the steps of firstly analyzing probability characteristics of new energy power generation prediction errors and load prediction errors, clustering the new energy power generation prediction errors and the load prediction errors respectively based on an improved fuzzy C-means clustering algorithm, respectively carrying out kernel density estimation on the new energy power generation prediction errors of each category, fitting the normal distribution of the load prediction error probability density of each category, and establishing a probability density model of the new energy power generation prediction errors and the load prediction errors; then, performing convolution integral on the probability densities of the two prediction errors to establish an equivalent standby probability density model; and integrating the equivalent backup probability density to obtain a probability distribution model, solving a confidence optimization model based on the optimal cost to obtain the optimal confidence corresponding to each equivalent backup category, and determining the optimal backup capacity.

Description

Power grid reserve capacity probabilistic dynamic evaluation method, system, terminal and readable storage medium containing new energy unit

Technical Field

The invention belongs to the technical field of power system reserve capacity evaluation, and particularly relates to a probabilistic dynamic evaluation method, system and terminal for reserve capacity of a power grid containing a new energy unit and a readable storage medium.

Background

With the rapid increase of new energy grid connection, the load shedding and the event of the generator tripping caused by the uncertainty of new energy power generation threaten the stable operation of the power system. Because the electric energy can not be stored in a large amount, before the high-proportion new energy power generation is combined into a grid, a power system can be additionally provided with a certain load for standby to meet the power market demand in consideration of the power demand of a user and the uncertainty of new energy power generation prediction. The traditional power system adopts a deterministic method, namely, the maximum value of daily power demand of 2% -5% or the maximum unit capacity is used as the lower limit of load reserve, and the load reserve of the power system is determined without considering the uncertainty factor of new energy power generation, which is clearly specified in the technical guide of the power system. Deterministic methods, although easy to implement, do not have theoretical support and cannot determine reasonable backup values.

On one hand, with the rapid development of the installed capacity of the new energy power generation, the deterministic method cannot solve the problem of electric energy shortage caused by the fact that the actual output of the new energy in the electric power system is less than a predicted value, so that the safety and the stability of the electric power system are lost; on the other hand, the deterministic method cannot solve the problem that the existing reserve capacity standard is larger than the actually required reserve capacity because the actual output of the new energy is larger than the predicted value, thereby causing economic loss. Therefore, in a new energy high-ratio power grid, it is important to reserve reasonable and effective reserve capacity, and the requirement is that unbalanced power caused by new energy power generation grid connection can be solved, and economic and environment-friendly effects can be realized.

Most scholars determine spare capacity based on statistical methods, data driven and risk controlled methods. The literature, "Holttinen Handle, Milligan Michael, Kirby Brendian, et al. Using Standard development as a Measure of available Operational Reserve for Wired Power [ J ]. Wired Engineering,2008,32(4): 355-377" adopts a statistical method based on Sigma (Standard Deviation) to evaluate the influence of Wind Power generation on the Reserve capacity of a Power system, and is applied to different time scales and time sequences. The literature, "Ortega-Vazquez M.A., Kirschen D.S. estimating the Spinning Reserve Requirements in Systems With Significant Wind Power Generation networking [ J ]. IEEE Transactions on Power Systems,2009,24(1):114 ″" uses mixed integer programming methods to determine the Reserve capacity and uses Monte Carlo simulations. This approach predicts the spare capacity based on deterministic criteria, taking into account random errors. The literature "Liu J.T., Feng S.H., Wang K., et al, method to determined residual requirement for a grid with large-scale with power requirement [ J ]. The Journal of Engineering,2017 (13): 1686-. The documents "Muzhikyan Aramazd, Farid Amo M., Youcef-Toumi Kamal.an a prior analytical method for the determination of operational redundancy [ J ]. International Journal of electric Power & Energy Systems,2017,86: 1-17" propose a formal mathematical framework incorporating probability distributions for a priori determining the three types of operational redundancy requirements, i.e., load tracking, ramping and regulation. Document "Mousavi Agah S. Mohammad, Flynn damian. Impact of modeling non-normal and stored dependency of variable on operating redundancy of Power Systems with high dependency of Power [ J ]. International Journal of electric Power and Energy Systems,2018,97:146 ″" considering wind Power uncertainty and non-normality of variability and its random dependency on wind Power prediction, proposes a method for dynamically determining wind Power system operational backup requirements, quantifying reserve uncertainty requirements by considering load and risk of wind Power variability.

In summary, the existing reserve capacity standard cannot meet the requirement of a new energy high-occupancy power grid, and the existing research directly aiming at the reserve capacity of the new energy high-occupancy power grid cannot be well adapted to the consideration of fluctuation and lack of economy of new energy power generation.

Disclosure of Invention

The invention aims to provide a probabilistic dynamic evaluation method, a probabilistic dynamic evaluation system, a probabilistic dynamic evaluation terminal and a readable storage medium for the spare capacity of a power grid containing a new energy unit, which are used for solving the problem that the existing spare capacity standard is not suitable in a new energy high-occupancy-ratio scene. The method makes up the defect that the deterministic prediction can only give the deterministic prediction information by utilizing the probabilistic prediction, can reflect the characteristic of the prediction error, simultaneously considers the influence of the new energy power generation prediction error and the load prediction error, and improves the accuracy of determining the reserve capacity; and the corresponding new energy power generation prediction errors and load prediction errors in the historical period are classified, so that different types have different equivalent standby probability distributions, the predicted standby capacity is more reliable, the prediction results of the different types have differences, the dynamic evaluation of the standby capacity is realized, the randomness and the volatility of new energy power generation are adapted, and the economic trend requirements are met.

On one hand, the invention provides a probabilistic dynamic evaluation method for the reserve capacity of a power grid containing a new energy unit, which comprises the following steps:

s1: acquiring power generation historical data and load historical data in a historical period in a power grid containing a new energy unit, and taking a new energy power generation prediction error and a new energy load prediction error calculated based on the power generation historical data and the load historical data as samples;

s2: classifying the new energy power generation prediction error and the load prediction error corresponding to the historical period;

s3: respectively constructing a new energy power generation prediction error probability density model and a load prediction error probability density model of corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories;

s4: calculating to obtain the equivalent standby probability distribution of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model;

the method comprises the following steps that an equivalent standby is the difference between a load prediction error and a new energy power generation prediction error, and the type of the equivalent standby is determined according to the type of a new energy power generation prediction error probability density model and the type of a load prediction error probability density model;

s5: and selecting the confidence coefficient, and obtaining the spare capacity based on the selected confidence coefficient and the probability distribution of the equivalent spare of the category.

The power grid reserve capacity probabilistic dynamic evaluation method containing the new energy unit can be applied to dynamic evaluation of the reserve capacity of the power grid containing new energy, and has a particularly remarkable effect in the power grid with high new energy occupation ratio. The method comprises the steps of analyzing probability characteristics of new energy power generation prediction errors and load prediction errors, making up the defect that deterministic prediction can only give out determined prediction information by adopting a probabilistic method, and simultaneously considering the influences of the new energy power generation prediction errors and the load prediction errors to more accurately determine the spare capacity. Particularly, the method classifies the new energy power generation prediction error and the load prediction error corresponding to the historical period, classifies similar data into one class by classification, classifies data with different characteristics into different classes, is more suitable for prediction timeliness and uncertainty, and achieves dynamic assessment of the reserve capacity.

Optionally, when classifying the new energy power generation prediction errors corresponding to the historical period in step S2, the classifying includes: classifying respectively according to two or more than two time segment lengths as units; or simultaneously classifying the data volume by taking the predicted power data volume as a standard and classifying the data volume respectively according to two or more time segment lengths as units;

the process of classifying by taking the predicted power data amount as a standard is as follows:

sequencing the new energy power generation prediction error samples according to the power generation prediction power values in the power generation historical data in the historical period;

equally dividing the new energy power generation prediction error sample according to the sample data amount;

and when the classification is respectively carried out according to two or more than two time segment lengths, the classification is carried out by adopting a clustering algorithm.

According to the method, when the new energy power generation prediction errors corresponding to the historical period are classified, time intervals are divided by taking the time intervals as units, so that the classification result has timeliness, different time intervals correspond to different types of models, the reserve capacity of the prediction time intervals is obtained by utilizing the probability distribution of equivalent reserve of the corresponding category, the difference of the prediction results of the different time intervals is promoted, and dynamic evaluation is realized on a time axis.

In addition, the method considers the correlation between the new energy power generation prediction error and the predicted power, segments the new energy power generation prediction error according to the data volume, and can embody the predicted power characteristics of different prediction errors.

Optionally, when the classification is performed respectively according to at least two or more time period length units, the selected time period lengths are tmin and 1 month respectively, and tmin is a sampling interval of the power generation historical data and the load historical data;

the method comprises the following steps of classifying by fuzzy C-means clustering, taking an average absolute error, a standard deviation, an average relative error value and an average error value as clustering indexes, and classifying the corresponding new energy power generation prediction errors in a historical period by taking a certain period length as a unit:

s2.2.1: extracting clustering indexes of data in the same time interval, and constructing a clustering index sample set E-E by using the clustering indexes₁,e₂,...，e_j,...，e_n1N1 is the number of epochs, any jth epoch sample e_jA one-dimensional vector consisting of p clustering indexes;

s2.2.2: setting an upper limit value S of iteration times, a category number c (c is more than or equal to 2 and less than or equal to n1), a minimum value epsilon of iteration termination and a weighting index m;

s2.2.3: initializing cluster center C ═ C₁,c₂,...,c_i,...,c_c}，c_iIs the cluster center of the ith group;

s2.2.4: updating a membership degree matrix U-U according to the following membership degree formula and a clustering center updating formula_ij)_c×n1And clustering the center C until the termination condition of iteration is met;

in the formula，u_ijIs the jth index e_jThe membership of the cluster center corresponding to the ith group,

and taking the class with the maximum membership degree in the membership degrees corresponding to the samples in the membership degree matrix after iteration termination as the class to which the samples belong.

The invention adopts fuzzy C-means clustering to process prediction errors in a segmented manner, compared with classification, clustering analysis is divided by the idea that the difference between classes is as large as possible and the difference in the classes is as small as possible, no preset class exists, similar data can be divided into one class according to the characteristics of the data, the data with different characteristics are divided into different classes, and the method is more suitable for the segmentation of the seasonal and time-interval prediction errors. Compared with general hard clustering, the fuzzy C-means clustering (FCM) provides the membership degree of each sample to a clustering center, and a more flexible clustering result can be obtained.

Optionally, the power generation history data includes a predicted power generation value P in each tmin of the history period_WFt(MW), actual Power Generation measurement P_WMt(MW) and installed Capacity P_it(MW); the load historical data comprises load prediction power value P per tmin in a historical period_LFt(MW), actual load measurement P_LMt(MW)；

The new energy power generation prediction error and the load prediction error are expressed by the following formula:

e_Lt＝P_LMt-P_LFt

in the formula, e_Wtpu、e_LtAnd respectively representing a new energy power generation prediction error and a load prediction error of tmin.

In the invention, the increase speed of the installed capacity of the new energy power generation is considered to be rapid, the research of using original data as a sample to make spare capacity cannot meet the requirement of change of the installed capacity, the concept of 'output ratio' is introduced into the calculation of prediction errors, and the power value of the new energy power generation is normalized to be in the interval of [0, 1 ].

Optionally, in step S3, the new energy power generation prediction error probability density model is established based on kernel density estimation in a non-parametric model, and the load prediction error probability density model is established based on normal distribution fitting.

If a Gaussian kernel function is selected, the probability density of the new energy power generation prediction error is obtained

Expressed as:

the load prediction error probability density f (e)_Lt) Expressed as:

in the formula (I), the compound is shown in the specification,

nuclear density estimation for new energy power generation prediction error yield ratio, P_iIndicating installed capacity of wind power, e_WPredicting errors of new energy power generation, wherein n is the number of samples of the new energy power generation prediction errors, and each sample of the new energy power generation prediction errors corresponds to one e_WtpuValue, h is window width, e_WpuError output ratio is predicted for new energy, e is natural constant, sigma_LwPredicting standard deviation, μ, of error samples for the corresponding intra-class load_LwThe average of the error samples is predicted for the load within the corresponding class.

The method comprises the steps of determining load probability density distribution by respectively adopting a normal distribution fitting method; the probability density distribution of the new energy power generation prediction error is determined by adopting a kernel density estimation method, the method can be more suitable for the spiking property, the multimodality property, the thick tailed property and the biased property of the new energy power generation prediction error, and more accurate probability distribution can be obtained.

The window width h is preferably selected by Mean Integrated Squared Error (MISE), which is not affected by individual samples and better reflects the overall properties than other methods.

Optionally, the process of obtaining the probability distribution of any type of equivalent spares in step S4 is:

convolving the new energy power generation prediction error probability density and the load prediction error probability density of the corresponding category to obtain equivalent standby probability density;

integrating the probability density of the equivalent standby to obtain the probability distribution of the equivalent standby;

wherein the probability density of the equivalent spares is expressed as:

in the formula, f (R)_t) For variable being equivalent reserve R_tThe probability density of the equivalent sparing.

Optionally, in step S5, determining an optimal confidence based on a confidence optimization model with the minimum expected cost of the system backup, and obtaining a backup capacity of the prediction period based on the selected optimal confidence and a probability distribution of an equivalent backup of a category to which the prediction period belongs;

if the expected cost of the system standby is selected to be the sum of the standby purchase cost and the standby penalty cost, the standby purchase cost comprises up-regulation standby purchase cost UC and down-regulation standby purchase cost DC, and the standby penalty cost comprises invalid up-regulation standby penalty cost IURP, invalid down-regulation standby penalty cost IDRP, load shedding penalty cost CLPC and load shedding penalty cost WAPC;

the objective function of the confidence optimization model is expressed as:

min(CLPC×(R_epα-R_mpα)+WAPC×(R_enα-R_mnα)+UC×R_euα+DC×R_edα+IURP×(R_eiuα-R_miuα)+IDRP×(R_midα-R_eidα))

wherein CLPC (R) is_epα-R_mpα) For the load cut with confidence 1-2 alphaPenalty cost, R_epαFor load shedding corresponding up-regulation (MW), R_mpαAlpha is a parameter set for describing confidence for the load shedding corresponding to the actual required standby;

WAPC×(R_enα-R_mnα) The penalty cost of cutting machine under the confidence coefficient of 1-2 alpha, R_enαFor standby under corresponding cutting machine, R_mnαThe cutter is needed for standby correspondingly and actually;

UC×R_euαand DC × R_edαRespectively, the standby purchase cost R under the confidence coefficient 1-2 alpha_euαAnd R_edαRespectively for up-regulation and down-regulation;

IURP×(R_eiuα-R_miuα) For invalid upshifts with confidence 1-2 alpha, reserve penalty cost, R_eiuαAnd R_miuαRespectively corresponding to invalid up-regulation under the confidence coefficient of 1-2 alpha for up-regulation standby and actually required standby;

IDRP×(R_midα-R_eidα) Reserve penalty cost, R, for invalid down-regulation at confidence 1-2 alpha_eidαAnd R_midαRespectively corresponding to invalid down regulation for standby and actual required standby;

the selected confidence coefficient also meets the following constraint conditions:

1-2α≥C_min

in the formula, C_minIs a preset minimum confidence level.

The method determines the optimal confidence coefficient based on the confidence coefficient optimization model with the minimum cost, comprehensively considers the influence of the standby purchase cost and the standby punishment cost on the standby expected cost, gives consideration to the reliability and the economy, and can obtain a more reasonable and feasible result. The method is applied to the evaluation of the reserve capacity of the power grid containing the new energy unit, is beneficial to reducing the probability of load shedding and generator tripping events under the power generation fluctuation of new energy, and ensures the safe and economic operation of the power system.

In a second aspect, the present invention provides an evaluation system based on the above method, which includes:

the historical data acquisition module is used for acquiring power generation historical data and load historical data in a historical period in a power grid containing the new energy unit;

the preprocessing module is used for taking a new energy power generation prediction error and a load prediction error which are calculated based on the power generation historical data and the load historical data as samples;

the classification module is used for classifying the new energy power generation prediction error and the load prediction error corresponding to the historical period;

the probability density model building module is used for respectively building a new energy power generation prediction error probability density model and a load prediction error probability density model of corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories;

the equivalent standby probability distribution building module is used for calculating the probability distribution of the equivalent standby of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model;

the method comprises the following steps that an equivalent standby is the difference between a load prediction error and a new energy power generation prediction error, and the type of the equivalent standby is determined according to the type of a new energy power generation prediction error probability density model and the type of a load prediction error probability density model;

the confidence coefficient selecting module is used for selecting the confidence coefficient;

and the spare capacity acquisition module is used for acquiring the spare capacity of the prediction time period based on the selected confidence coefficient and the probability distribution of the equivalent spare of the class to which the prediction time period belongs.

In a third aspect, the present invention provides a terminal comprising a processor and a memory, the memory storing a computer program, the processor calling the computer program to implement:

a probabilistic dynamic evaluation method for the reserve capacity of a power grid containing a new energy unit comprises the following steps.

In a fourth aspect, the present invention provides a readable storage medium storing a computer program for invocation by a processor to implement:

step of power grid reserve capacity probabilistic dynamic evaluation method containing new energy unit

Advantageous effects

1. The method realizes the probabilistic dynamic evaluation of the reserve capacity of the power grid containing the new energy unit based on the equivalent reserve model, makes up the defect that the deterministic prediction can only give out the deterministic prediction information by a probabilistic method, considers the influence of the power generation prediction error and the load prediction error of the new energy, and improves the accuracy of the reserve capacity determination.

Wherein, the probability density distribution is further determined by adopting a normal distribution fitting method for the load prediction error; for the new energy power generation prediction error, the probability density distribution is determined by adopting a kernel density estimation method, so that the method disclosed by the invention can be more suitable for the spiking property, the multimodality property, the thick tailed property and the biased property of the prediction error, and can obtain more accurate probability distribution.

2. The method classifies the new energy power generation prediction error and the load prediction error corresponding to the historical period, realizes classification of similar data into one class according to data characteristics, and classifies different characteristic data into different classes, so that the different classes have different equivalent standby probability distributions, the predicted standby capacity is more reliable, the prediction results of the different classes have differences, and dynamic evaluation of the standby capacity is realized.

The improved fuzzy C-means clustering is further preferably adopted to establish a prediction error segmentation model, time intervals are used as units for classification, and two or more time intervals are preferably used as units for classification, so that the method is more suitable for seasonal and time interval correlation research of errors. Compared with general hard clustering, the fuzzy C-means clustering provides the membership degree of each sample to a clustering center, and a more flexible clustering result can be obtained. The global optimal clustering initial value can be determined by means of a genetic algorithm and a simulated annealing algorithm, and the accuracy of a clustering result can be effectively improved.

3. In a further preferred scheme of the invention, the optimal confidence is determined by a confidence optimization model with the minimum expected cost of system standby, the influence of standby purchase cost and standby punishment cost on the standby expected cost is comprehensively considered, wherein the punishment cost can be used as an evaluation index of reliability, and meanwhile, in order to ensure certain operational reliability, reliability constraint conditions are set, reliability and economy are considered, and more reasonable and feasible results can be obtained.

4. The method can be applied to the evaluation of the reserve capacity of the power grid containing the new energy unit, is beneficial to reducing the probability of load shedding and generator tripping events under the power generation fluctuation of new energy, and ensures the safe and economic operation of the power system.

Drawings

FIG. 1 is a flowchart of an algorithm for determining an optimal initial cluster center;

FIG. 2 is a wind power prediction error seasonal clustering result;

FIG. 3 shows seasonal clustering results of load prediction errors;

FIG. 4 shows a probability density distribution of wind power prediction errors (1, 5, 5);

FIG. 5 is a class 1 probability density distribution of load prediction error;

FIG. 6 shows (1, 5, 5, 1) equivalent spare probability densities

FIG. 7 is a (1, 5, 5, 1) equivalent standby probability distribution

FIG. 8 is the equivalent spare (1, 5, 5, 1) different confidence spare total cost;

FIG. 9 is a comparison of spare capacity at 1/month 3, 12, 14, and 28/year 2021;

FIG. 10 is a comparison of spare capacity at 1/5/17/2021;

fig. 11 is a schematic flow chart of a probabilistic dynamic evaluation method for the reserve capacity of a power grid including a new energy unit according to the present invention.

Detailed Description

The method for probabilistic dynamic evaluation of the standby capacity of the power grid comprising the new energy unit is suitable for the power grid system comprising new energy, particularly has a remarkable effect on the high-occupancy system of the new energy, and is used for realizing dynamic evaluation of the standby capacity of the high-occupancy system of the new energy, wherein the high occupancy of the new energy is set by common general knowledge or experience in the field, and meanwhile, the method is also suitable for other high-occupancy new energy systems. In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflicting with each other.

In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention, and to illustrate a new energy high duty system, however, the present invention may be practiced in other ways than those specifically described and thus the scope of the present invention is not limited by the specific embodiments disclosed below.

Example 1:

the embodiment takes wind power as a new energy as an example for explanation, and it should be understood that the invention is not limited to a power grid of wind power new energy, and the embodiment provides a probabilistic dynamic evaluation method for standby capacity of a power grid including a new energy unit, which includes the following steps:

s1: the method comprises the steps of obtaining power generation historical data and load historical data in a historical period of a new energy high-occupancy-ratio power grid, and using new energy power generation prediction errors and load prediction errors calculated based on the power generation historical data and the load historical data as samples.

The 2020 wind power data and load data disclosed in the official network by the belgian transmission system operator Elia are adopted in the embodiment, and comprise a day-ahead wind power prediction power value (MW), an actual wind power measurement value (MW) and installed capacity (MW), a load prediction power value (MW) and an actual load measurement value (MW). The data are sampled at intervals of t-15 min, and the data comprise 96 groups of data per day and 35136 groups of data all year round. In other feasible embodiments, t is not the only value, and the value t may be adaptively adjusted according to the actual working condition requirement, which is not specifically limited in the present invention.

It should be noted that, in this embodiment, in consideration of the preliminary detection of load data, it is found that the original data has a data missing phenomenon, and lagrangian interpolation processing is performed on the missing data before formal calculation errors are performed, so as to obtain a complete data set for calculation and analysis. In other possible embodiments, other interpolation algorithms may be selected for processing, and the present invention is not limited in this respect.

The wind power prediction error corresponding to the t time period is as follows:

e_Wt＝P_WMt-P_WFt

in the formula, P_WMtRepresenting the actual wind power measured value (MW) in the period t; p_WFtAnd represents the predicted power value (MW) of the wind power day ahead in the t period.

Considering that the increase speed of the installed capacity of the wind power is rapid, the research of using original data as a sample to make spare capacity cannot meet the requirement of change of the installed capacity, the concept of 'output ratio' is introduced, the power value of the wind power is normalized to be in the interval of [0, 1], and meanwhile, the calculation is simplified.

the wind power prediction error value actually used for calculation in the period t is as follows:

the wind power prediction power value actually used for calculation in the period t is as follows:

in the formula, P_itRepresenting the installed wind capacity (MW) over a period of t.

the load prediction error corresponding to the t period is as follows:

e_Lt＝P_LMt-P_LFt

in the formula, P_LMtRepresenting the actual load Measurement (MW); p_LFtRepresenting the load predicted power value (MW).

S2: and classifying the corresponding new energy power generation prediction error and load prediction error in the historical period. The purpose of this step is to classify similar data into one class according to the characteristics of the data, and to classify data with different characteristics into different classes according to the probability characteristics of error data.

In this embodiment, the new energy power generation prediction errors are classified three times, which are respectively: the classification is carried out by taking the predicted power data amount as a standard, the classification is carried out by taking the t period length as a unit, and the classification is carried out by taking a month as a unit.

When the predicted power data amount is used as a standard for classification, the predicted power characteristic of the wind power prediction error is considered, and new energy power generation prediction error samples are sequenced according to the power generation prediction power value in power generation historical data in a historical period; and classifying the new energy power generation prediction error samples according to the sample data size. In this embodiment, the wind power prediction error sample is divided into 5 segments, so that the number of each segment is equal as much as possible, and the number of each segment is:

wherein i is the number of stages, and i is 1, 2, 3, 4, 5; n is the number of samples, where n is 35136. In other possible embodiments, the number of segments may be adjusted. The predicted power output ratio segmentation points are obtained in this example: 0.0682,0.2036,0.4047,0.5952.

The classification with the month as the unit is 2, considering the seasonality of wind power prediction errors, respectively calculating the average absolute error, the standard deviation, the average relative error value and the average error value of 12 months of wind power predictions as clustering indexes, taking the number of clustering groups as 4, taking a weighting index as 2, and clustering the wind power prediction errors and the load prediction errors by adopting an improved fuzzy C mean clustering method to obtain respective categories.

And when the unit of the t time interval length is used for classification, average absolute errors, standard deviations, average relative error values and average error values of wind power prediction in 96 time intervals are respectively calculated to serve as clustering indexes.

In this embodiment, the load prediction error is also calculated by taking a month as a unit, and an average absolute error, a standard deviation, an average relative error value, and an average error value of the load prediction errors of 12 months are respectively calculated as a clustering index.

FIG. 1 is a flow chart of an algorithm for determining an optimal initial cluster center based on a genetic algorithm and a simulated annealing algorithm. The classification process by using the improved fuzzy C-means clustering method is as follows:

s2.2.1: extracting clustering indexes of data in the same time interval, and constructing a clustering index sample set E-E by using the clustering indexes₁,e₂,...，e_j,...，e_n1N1 is the number of epochs, any jth epoch sample e_jA one-dimensional vector consisting of p clustering indexes;

s2.2.2: setting an upper limit value S of iteration times, a category number c (c is more than or equal to 2 and less than or equal to n1), a minimum value epsilon of iteration termination and a weighting index m to control the accuracy degree of a result, wherein (m > 1);

s2.2.3: initializing cluster center C ═ C₁,c₂,...,c_i,...,c_c}，c_iIs the cluster center of the ith group;

s2.2.4: updating a membership degree matrix U-U according to the following membership degree formula and a clustering center updating formula_ij)_c×n1And clustering the center C until the termination condition of iteration is met;

in the formula u_ijIs the jth index e_jThe membership of the cluster center corresponding to the ith group,

and taking the class with the maximum membership degree in the membership degrees corresponding to the samples in the membership degree matrix after iteration termination as the class to which the samples belong.

Classifying by adopting the improved fuzzy C-means clustering method to obtain clustering results of tables 1-3, wherein the table 1 shows a seasonal clustering result of wind power prediction errors, and FIG. 2 shows a clustering result which describes month wind power prediction errors by taking average absolute errors, standard deviations and relative error average values as coordinate components; and table 2 shows the wind power prediction error time interval clustering result. Table 3 shows the seasonal clustering results of the load prediction errors, and fig. 3 shows the clustering results of the load prediction errors in months described by using the average absolute error, the standard deviation, and the average value of the relative errors as coordinate components.

TABLE 1 wind power prediction error seasonal clustering results

TABLE 2 wind power prediction error time interval clustering results

TABLE 3 seasonal clustering results of load prediction errors

In the embodiment, 4 types of wind power prediction error seasonal categories, 12 types of wind power prediction error time interval categories, 5 types of prediction power categories and 4 types of load prediction error seasonal categories are obtained based on the classification; and (1, 5, 5, 1) representing error data belonging to the 1 st class of wind power prediction error seasons, the 5 th class of time interval correlation, the 5 th class of predicted power and the 1 st class of load prediction error.

It is noted that in other possible embodiments, the classification may be adjusted, such as the selection of time periods, such as irrespective of the predicted power data amount, etc., without departing from the inventive concept.

S3: and respectively constructing a new energy power generation prediction error probability density model and a load prediction error probability density model of the corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories.

In order to fully explain the implementation of the present invention, the following will take the category (1, 5, 5, 1) as an example. Wherein, the probability density of wind power prediction errors (1, 5, 5) and the probability density of the load prediction error class 1 are respectively calculated.

Setting the wind power prediction error output ratio as e_Wpu(set as independent variable), then the kernel density is estimated as:

in the formula, n is the number of samples of the wind power prediction error, and is determined according to the number of samples of each category, namely the wind power prediction error sample data in the category of the wind power prediction error (1, 5, 5), e_WtpuIs determined by samples of the corresponding category, each sample i corresponding to an e_WtpuThe value is obtained.

In this embodiment, a gaussian kernel function is selected, and kernel density estimation is performed on the wind power prediction error output ratio. Other kernel functions may be selected in other possible embodiments. The gaussian kernel function selected in this embodiment is:

thus, the kernel density estimate of the wind-power prediction error output ratio under the gaussian kernel function is:

the wind power prediction error kernel density is estimated as:

in the formula, P_iRepresenting the installed wind capacity (MW).

In this embodiment, the optimal window width is preferably determined by Mean Integrated Squared Error (MISE).

S31: MISE is defined as:

s32: solving the above equation to obtain:

in the formula (d)₂＝∫e_d ²K(e_d)de_d，e_dIs an independent variable in the Gaussian kernel function, e is used for calculating the wind power prediction error_d＝e_W，

Is the kernel density estimate of error e, and f (e) is the probability density actual of error e.

S33: the last term of the above equation is removed to obtain a progressive integral mean square error:

s34: calculating partial derivative of the above formula to make the first derivative zero to obtain the optimum window width h_AMISEComprises the following steps:

the normal distribution model of the load prediction error probability density is fit as follows:

in the formula, σ_LwStandard deviation (MW) for load prediction error within the w class; mu.s_LwThe Mean (MW) of the w class load prediction errors. In this exampleThe w category is load prediction error category 1.

Fig. 4 shows the probability density distribution of the wind power prediction error (1, 5, 5) kernel density estimation, and fig. 5 shows the normal distribution fitting of the class 1 probability density of the load prediction error. It can be seen that the kernel density estimation can well reflect the spiking property, the multimodality property, the thick tailed property and the biased property of the wind power prediction error probability density distribution.

It should be appreciated that other types of wind power prediction error probability densities and load prediction error probability densities may be calculated according to the above concepts.

S4: and calculating to obtain the equivalent standby probability distribution of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model.

Wherein, the equivalent backup is defined as the difference between the wind power prediction error and the load prediction error:

R_t＝(P_LMt-P_LFt)-(P_WMt-P_WFt)＝e_Lt-e_Wt

convolving the probability densities of the two prediction errors to obtain the probability density of the equivalent backup:

fig. 6 shows (1, 5, 5, 1) equivalent spare probability density distributions.

Integrating the probability density of the equivalent backup to obtain the probability distribution of the equivalent backup, wherein the equivalent backup meeting the confidence coefficient of 1-2 alpha is as follows:

R_α＝{R_x|P(e_rt)≤1-2α,R_x∈R}

s5: and selecting the confidence coefficient, and obtaining the spare capacity based on the selected confidence coefficient and the probability distribution of the equivalent spare of the category.

It should be noted that, when the spare capacity is predicted, the probability distribution of the equivalent spare is determined according to the category to which the prediction time period belongs, and then the spare capacity is obtained by using the probability distribution of the equivalent spare of the corresponding category. That is, in the present embodiment, for the prediction period, the spare capacity of 2021.06.30 am 8 am is predicted, first according to the prediction period: the classification of the 4-class wind power prediction error seasonal classification, the 12-class wind power prediction error time interval classification, the 5-class prediction power classification and the 4-class load prediction error seasonal classification is determined, then the probability distribution of equivalent standby of the corresponding classification is further determined, and finally the confidence coefficient is selected to obtain the standby capacity. And when the category of the 5 types of predicted power is judged, determining the category according to the power generation predicted power value of the predicted time period.

It should be further noted that, in this embodiment, in order to ensure the reliability of the model, the clustering analysis is based on historical data of the whole year of the previous year, and the historical data of the month corresponding to the previous year needs to be replaced after one month has passed for re-clustering, so as to have stronger timeliness. Meanwhile, due to the game between the data limitation and the timeliness, the situation that the prediction sample exceeds the clustering range may occur, and the traditional standby method is adopted to ensure the safe operation of the power system. In other possible embodiments, the period of data replacement may be adaptively adjusted.

Fig. 7 shows (1, 5, 5, 1) equivalent spare probability distributions, with equivalent spare points with 95%, 90%, 85%, 80% confidence levels being chosen. It can be seen that when the confidence is 95%, the system requires at least 989.3MW of up-reserve capacity, corresponding to at least 411MW of down-reserve capacity. For the same confidence increment, when the confidence level is close to 1, the system needs more up-conversion spare capacity to meet the demand for confidence increase.

In order to determine the reasonable effective spare capacity, the sum of the spare purchase cost and the penalty cost is minimum to be used as a standard for selecting the optimal confidence level, and the confidence corresponding to the minimum expected cost is selected as the optimal confidence level, so that the load spare of each period in the future is evaluated.

Setting CLPC as load shedding punishment cost and WAPC as wind abandonment punishment cost; UC is the purchase cost for up-regulation, and DC is the purchase cost for down-regulation; if IURP is the invalid up-call penalty and IDRP is the invalid down-call penalty, the objective function of the confidence optimization model can be expressed as:

min(CLPC×(R_epα-R_mpα)+WAPC×(R_enα-R_mnα)+UC×R_euα+DC×R_edα+IURP×(R_eiuα-R_miuα)+IDRP×(R_midα-R_eidα))

wherein CLPC (R) is_epα-R_mpα) Penalty cost for load shedding at confidence 1-2 alpha, R_epαFor load shedding corresponding up-regulation (MW), R_mpαCorresponding to the actual required standby (MW) for load shedding;

WAPC×(R_enα-R_mnα) Penalty cost for wind curtailment at confidence 1-2 alpha, R_enαFor setting-up (MW), R, in response to wind curtailment_mnαCorresponding to the actual needed standby (MW) for the abandoned wind;

UC×R_euαand DC × R_edαRespectively, the standby purchase cost R under the confidence coefficient 1-2 alpha_euαAnd R_edαUp-regulation for (MW) and down-regulation for (MW) respectively;

IURP×(R_eiuα-R_miuα) For invalid upshifts with confidence 1-2 alpha, reserve penalty cost, R_eiuαAnd R_miuαUp-regulation standby (MW) corresponding to invalid up-regulation under the confidence coefficient of 1-2 alpha and actually required standby (MW) respectively;

IDRP×(R_midα-R_eidα) Reserve penalty cost, R, for invalid down-regulation at confidence 1-2 alpha_eidαAnd R_midαCorresponding down-regulated standby (MW) and actual required standby (MW) are respectively ineffective down-regulated.

In order to ensure certain operational reliability, reliability constraint conditions are set as follows:

1-2α≥C_min

in the formula, C_minAnd the minimum confidence is determined by the actual requirement of the power plant for the spare capacity.

It should be understood that other possible embodiments may select other ways to determine the confidence, such as an empirical value, and may also select other cost functions or add other costs based on the spare purchase cost and the spare penalty cost for the expected cost of the system spare in the confidence optimization model, which may be adjusted according to the actual requirement, and the invention is not limited in this respect.

Adopts the literature' Zhang Danning, xu Jian, Sunyuan chapter, etc.. the dynamic evaluation and optimization of the day-ahead standby of the wind power system [ J]The power grid technology, 2019,43(09), 3252-; the purchase cost for up-regulation is 15USD/MW, and the purchase cost for down-regulation is 5 USD/MW; the penalty cost of invalid up-regulation backup is 425USD/MW, the penalty cost of invalid down-regulation backup is 104USD/MW, and the expected cost of the backup under different confidence degrees of equivalent backup (1, 5, 5, 1) is calculated. FIG. 8 shows the expected cost trend of the equivalent spares (1, 5, 5, 1) at different confidence levels, taking the minimum confidence level of 85%, the optimal confidence level of the equivalent spares (1, 5, 5, 1) is 86%, and the corresponding expected cost is 4.542 × 10⁸And (5) USD. At this confidence level, the optimal upturn for the system is 780.287MW and the optimal downturn is-256.464 MW. The calculation is carried out on each type of equivalent backup, and the optimal confidence coefficient of each type and the corresponding optimal equivalent backup can be obtained. When the price standard is different from the lowest confidence coefficient, the optimal confidence coefficient is also different, and the corresponding optimal equivalent backup is different.

The method comprises the steps of taking Belgian wind power data of 2976 time periods in 1 month in 2021 as test data, determining equivalent spares of 2976 time periods in 1 month in 2021 based on the obtained optimal equivalent spares, calculating actually required spares as verification, and calculating the spares determined by a traditional method and documents' Sinkiang, Huangviui, Liudwei, and the like.

According to the technical guidelines of the power system, a conventional method is chosen here which uses 5% of the daily maximum load as reserve capacity.

The literature is based on wind power and load prediction error independence, and applies the root mean square principle, and the specific spare capacity calculation mode is as follows:

in the formula, P_LMIs the maximum load (MW), P, during the day_WTTotal capacity (MW), P, for wind installations_IMThe maximum installed capacity (MW) of a single unit.

Fig. 9 shows the evaluation results of the reserve capacity of 3, 12, 14, and 28 days in 1 month in 2021, the actual wind power output of the four days is smaller than the predicted value, and the prediction error is larger, it can be seen that the conventional method cannot meet the reserve capacity requirement, and the literature method also has a situation that the capacity requirement cannot be met, but compared with the conventional method and the literature method, the reserve capacity determined by the present invention is closer to the actually required reserve, and the load shedding risk is lower;

fig. 10 shows the evaluation results of the reserve capacity of 5 days and 17 days in 1 month in 2021, the actual output of the wind power is greater than the predicted value on the two days, and the prediction error is larger, so that it can be seen that the reserve capacity determined by the method of the present invention is closer to the actually required reserve, the invalid up-regulation is less, the risk of wind abandoning is lower, and the demand of the reserve capacity of the power system can be better adapted.

In order to further evaluate the effectiveness and the reasonability of the model, the expected cost, the coverage rate of the prediction interval and the average width of the prediction interval are selected as indexes for evaluating the economy, the reliability and the acuity of the model. The coverage rate of the prediction interval refers to the coverage condition of the prediction interval to the actually required backup, the higher the coverage rate is, the higher the reliability of the model is, and the interval coverage rate p is_cComprises the following steps:

in the formula, #_iTo determine if the actual value is within the Boolean of the prediction interval, psi is applied if the actual value is not within the prediction interval_iTake 0, if the actual value is within the prediction interval,. psi_iTaking 1;

is the actual value of the i-th period,

in order to be the lower boundary of the interval,

at the upper boundary of the interval, n is the number of samples, here taken as 2976.

The prediction section average width is an average value of the prediction section widths, and the prediction accuracy is higher as the prediction average width is narrower. The calculation expression is:

table 4 shows evaluation indexes of the backup evaluation at 2021 year and 1 month under the three methods, that is, the expected cost, the prediction section coverage, and the prediction section average width.

TABLE 4 evaluation index comparison

It can be seen that although the average width of the prediction interval is the minimum in the conventional method, the coverage rate of the prediction interval is small, the reliable operation of the power system cannot be guaranteed, the expected cost of the backup is high, the economical efficiency of the operation of the power system cannot be guaranteed, and the method cannot meet the backup requirement of a high-proportion wind power system. The coverage rate of the equivalent spare interval determined by the equivalent spare model and the literature method provided by the invention is about 95%, but compared with the literature method, the equivalent spare model provided by the invention has smaller expected spare cost and average width of the prediction interval, and has higher reliability and economy. The equivalent standby model provided by the invention is reasonable and effective, and can meet the standby requirement of a high-proportion wind power system.

Example 2:

the embodiment of the invention provides an evaluation system based on a power grid spare capacity probabilistic dynamic evaluation method comprising a new energy unit, which comprises the following steps: the system comprises a historical data acquisition module, a preprocessing module, a classification module, a probability density model construction module, an equivalent standby probability distribution construction module, a confidence coefficient selection module and a standby capacity acquisition module.

The historical data acquisition module is used for acquiring power generation historical data and load historical data in a historical period of a power grid containing the new energy unit.

And the preprocessing module takes the new energy power generation prediction error and the load prediction error calculated based on the power generation historical data and the load historical data as samples.

The classification module is used for classifying the new energy power generation prediction error and the load prediction error corresponding to the historical period.

And the probability density model building module is used for respectively building a new energy power generation prediction error probability density model and a load prediction error probability density model of corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories.

And the equivalent standby probability distribution building module calculates the probability distribution of the equivalent standby of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model.

The method comprises the following steps that an equivalent standby is the difference between a load prediction error and a new energy power generation prediction error, and the type of the equivalent standby is determined according to the type of a new energy power generation prediction error probability density model and the type of a load prediction error probability density model;

the confidence coefficient selecting module is used for selecting the confidence coefficient;

and the spare capacity acquisition module is used for acquiring the spare capacity of the prediction time period based on the selected confidence coefficient and the probability distribution of the equivalent spare of the class to which the prediction time period belongs.

In some feasible manners, when the classification module classifies the new energy power generation prediction error and the load prediction error corresponding to the historical period, the method includes: classifying respectively according to at least two or more than two time period lengths as units; or simultaneously classifying the data according to the predicted power data amount and classifying the data according to at least two or more time segment lengths. Such as selecting t 15min and sorting in units of months.

In some feasible modes, fuzzy C-means clustering is adopted for classification, the average absolute error, the standard deviation, the average relative error value and the average error value are used as clustering indexes, and the corresponding new energy power generation prediction error in a historical period is classified by taking a certain period length as a unit.

In some possible ways, the new energy power generation prediction error probability density model is established based on kernel density estimation in a non-parametric model, and the load prediction error probability density model is established based on normal distribution fitting.

In some feasible modes, the optimal confidence coefficient is determined based on a confidence coefficient optimization model with the minimum expected cost of the system backup, and the backup capacity of the prediction period is obtained based on the selected optimal confidence coefficient and the probability distribution of the equivalent backup of the category to which the prediction period belongs.

For the specific implementation process of each unit module, refer to the corresponding process of the foregoing method. It should be understood that, the specific implementation process of the above unit module refers to the method content, and the present invention is not described herein in detail, and the division of the above functional module unit is only a division of a logic function, and there may be another division manner in the actual implementation, for example, multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or may not be executed. Meanwhile, the integrated unit can be realized in a hardware form, and can also be realized in a software functional unit form.

Example 3:

the embodiment of the invention provides a terminal, which comprises a processor and a memory, wherein the memory stores a computer program, and the processor calls the computer program to realize that:

s1: acquiring power generation historical data and load historical data in a historical period in a power grid containing a new energy unit, and taking a new energy power generation prediction error and a new energy load prediction error calculated based on the power generation historical data and the load historical data as samples;

s2: classifying the new energy power generation prediction error and the load prediction error corresponding to the historical period;

s3: respectively constructing a new energy power generation prediction error probability density model and a load prediction error probability density model of corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories;

s4: calculating to obtain the equivalent standby probability distribution of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model;

the method comprises the following steps that an equivalent standby is the difference between a load prediction error and a new energy power generation prediction error, and the type of the equivalent standby is determined according to the type of a new energy power generation prediction error probability density model and the type of a load prediction error probability density model;

s5: and selecting the confidence coefficient, and obtaining the spare capacity based on the selected confidence coefficient and the probability distribution of the equivalent spare of the category.

In some implementations, the processor calls the computer program to further implement: when classifying the corresponding new energy power generation prediction errors in the historical period, the method comprises the following steps: classifying respectively according to at least two or more than two time period lengths as units; or simultaneously classifying the data according to the predicted power data amount and classifying the data according to at least two or more time segment lengths. In some implementations, the processor calls the computer program to further implement: the process of classifying by taking the predicted power data amount as a standard is as follows:

sequencing the new energy power generation prediction error samples according to the power generation prediction power values in the power generation historical data in the historical period;

classifying the new energy power generation prediction error samples according to the sample data size;

and when the classification is respectively carried out according to at least two or more than two time period lengths as units, the classification is carried out by adopting a clustering algorithm.

In some implementations, the processor calls the computer program to further implement: and determining the optimal confidence coefficient based on a confidence coefficient optimization model with the minimum expected cost of the system standby, and obtaining the standby capacity of the prediction period based on the selected optimal confidence coefficient and the probability distribution of the equivalent standby of the category to which the prediction period belongs.

The specific implementation process of each step refers to the explanation of the foregoing method.

It should be understood that in the embodiments of the present invention, the Processor may be a Central Processing Unit (CPU), and the Processor may also be other general purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The memory may include both read-only memory and random access memory, and provides instructions and data to the processor. The portion of memory may also include non-volatile random access memory. For example, the memory may also store device type information.

Example 4:

an embodiment of the present invention provides a readable storage medium, which stores a computer program, where the computer program is called by a processor to implement:

s1: acquiring power generation historical data and load historical data in a historical period in a power grid containing a new energy unit, and taking a new energy power generation prediction error and a new energy load prediction error calculated based on the power generation historical data and the load historical data as samples;

s2: classifying the new energy power generation prediction error and the load prediction error corresponding to the historical period;

s3: respectively constructing a new energy power generation prediction error probability density model and a load prediction error probability density model of corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories;

s4: calculating to obtain the equivalent standby probability distribution of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model;

the method comprises the following steps that an equivalent standby is the difference between a load prediction error and a new energy power generation prediction error, and the type of the equivalent standby is determined according to the type of a new energy power generation prediction error probability density model and the type of a load prediction error probability density model;

s5: and selecting the confidence coefficient, and obtaining the spare capacity based on the selected confidence coefficient and the probability distribution of the equivalent spare of the category.

In some implementations, the computer program is invoked by a processor to implement: when classifying the corresponding new energy power generation prediction errors in the historical period, the method comprises the following steps: classifying respectively according to at least two or more than two time period lengths as units; or simultaneously classifying the data according to the predicted power data amount and classifying the data according to at least two or more time segment lengths.

In some implementations, the computer program is invoked by a processor to implement: the process of classifying by taking the predicted power data amount as a standard is as follows:

sequencing the new energy power generation prediction error samples according to the power generation prediction power values in the power generation historical data in the historical period;

classifying the new energy power generation prediction error samples according to the sample data size;

and when the classification is respectively carried out according to at least two or more than two time period lengths as units, the classification is carried out by adopting a clustering algorithm.

In some implementations, the computer program is invoked by a processor to implement: and determining the optimal confidence coefficient based on a confidence coefficient optimization model with the minimum expected cost of the system standby, and obtaining the standby capacity of the prediction period based on the selected optimal confidence coefficient and the probability distribution of the equivalent standby of the category to which the prediction period belongs.

The readable storage medium is a computer readable storage medium, which may be an internal storage unit of the controller according to any of the foregoing embodiments, for example, a hard disk or a memory of the controller. The readable storage medium may also be an external storage device of the controller, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like provided on the controller. Further, the readable storage medium may also include both an internal storage unit of the controller and an external storage device. The readable storage medium is used for storing the computer program and other programs and data required by the controller. The readable storage medium may also be used to temporarily store data that has been output or is to be output.

Based on such understanding, the technical solution of the present invention essentially or partially contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned readable storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

The probabilistic dynamic evaluation method for the reserve capacity of the power grid comprising the new energy unit can be applied to dynamic evaluation of the reserve capacity of the power grid comprising the new energy unit. The method adopts a probabilistic method, makes up the defect that the deterministic prediction can only give the deterministic prediction information, considers the influence of the new energy power generation prediction error and the load prediction error, and can more accurately determine the reserve capacity. The method comprises the steps of determining load probability density distribution by respectively adopting a normal distribution fitting method; the probability density distribution of the new energy power generation prediction error is determined by adopting a kernel density estimation method, the method can be more suitable for the spiking property, the multimodality property, the thick tailed property and the biased property of the new energy power generation prediction error, and more accurate probability distribution can be obtained. The method establishes the prediction error segmentation model based on the improved fuzzy C-means clustering, is more suitable for the research of seasonality and time-interval correlation of errors, can obtain more flexible clustering results, and can effectively improve the accuracy of the clustering results by determining the overall optimal clustering initial value by means of a genetic algorithm and a simulated annealing algorithm. The method determines the optimal confidence coefficient based on the confidence coefficient optimization model with the minimum cost, comprehensively considers the influence of the standby purchase cost and the standby punishment cost on the standby expected cost, gives consideration to the reliability and the economy, and can obtain a more reasonable and feasible result. The method is applied to the evaluation of the reserve capacity of the power grid containing the new energy unit, is beneficial to reducing the probability of load shedding and generator tripping events under the power generation fluctuation of new energy, and ensures the safe and economic operation of the power system.

It should be emphasized that the examples described herein are illustrative and not restrictive, and thus the invention is not to be limited to the examples described herein, but rather to other embodiments that may be devised by those skilled in the art based on the teachings herein, and that various modifications, alterations, and substitutions are possible without departing from the spirit and scope of the present invention.

Claims (10)

1. A probabilistic dynamic evaluation method for the reserve capacity of a power grid comprising a new energy unit is characterized by comprising the following steps: the method comprises the following steps:

s1: acquiring power generation historical data and load historical data in a historical period in a power grid containing a new energy unit, and taking a new energy power generation prediction error and a new energy load prediction error calculated based on the power generation historical data and the load historical data as samples;

s2: classifying the new energy power generation prediction error and the load prediction error corresponding to the historical period;

s3: respectively constructing a new energy power generation prediction error probability density model and a load prediction error probability density model of corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories;

s4: calculating to obtain the equivalent standby probability distribution of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model;

the method comprises the following steps that an equivalent standby is the difference between a load prediction error and a new energy power generation prediction error, and the type of the equivalent standby is determined according to the type of a new energy power generation prediction error probability density model and the type of a load prediction error probability density model;

s5: and selecting the confidence coefficient, and obtaining the spare capacity based on the selected confidence coefficient and the probability distribution of the equivalent spare of the category.

2. The method of claim 1, wherein: when classifying the new energy power generation prediction errors corresponding to the historical period in step S2, the method includes: classifying respectively according to two or more than two time segment lengths as units; or simultaneously classifying the data volume by taking the predicted power data volume as a standard and classifying the data volume respectively according to two or more time segment lengths as units:

the process of classifying by taking the predicted power data amount as a standard is as follows:

sequencing the new energy power generation prediction error samples according to the power generation prediction power values in the power generation historical data in the historical period;

equally dividing the new energy power generation prediction error sample according to the sample data amount;

and when the classification is respectively carried out according to two or more than two time segment lengths, the classification is carried out by adopting a clustering algorithm.

3. The method of claim 2, wherein: when the classification is respectively carried out according to two or more than two time interval length units, the selected time interval lengths are tmin and 1 month respectively, and tmin is a sampling interval of the power generation historical data and the load historical data;

the method comprises the following steps of classifying by fuzzy C-means clustering, taking an average absolute error, a standard deviation, an average relative error value and an average error value as clustering indexes, and classifying the corresponding new energy power generation prediction errors in a historical period by taking a certain period length as a unit:

s2.2.1: extracting clustering indexes of data in the same time interval, and constructing a clustering index sample set E-E by using the clustering indexes₁,e₂,...，e_j,...，e_n1N1 is the number of epochs, any jth epoch sample e_jA one-dimensional vector consisting of p clustering indexes;

s2.2.2: setting an upper limit value S of iteration times, a category number c (c is more than or equal to 2 and less than or equal to n1), a minimum value epsilon of iteration termination and a weighting index m;

s2.2.3: initializing cluster center C ═ C₁,c₂,...,c_i,...,c_c}，c_iIs the cluster center of the ith group;

s2.2.4: updating a membership degree matrix U-U according to the following membership degree formula and a clustering center updating formula_ij)_c×n1And clustering the center C until the termination condition of iteration is met;

in the formula u_ijIs the jth index e_jCluster center c corresponding to ith group_iDegree of membership of;

and taking the class with the maximum membership degree in the membership degrees corresponding to the samples in the membership degree matrix after iteration termination as the class to which the samples belong.

4. The method of claim 1, wherein: the power generation history data includes a predicted power generation value P of day ahead in each tmin of a history period_WFt(MW), actual Power Generation measurement P_WMt(MW) and installed Capacity P_it(MW); the load historical data comprises load prediction power value P per tmin in a historical period_LFt(MW), actual load measurement value P_LMt(MW)；

The new energy power generation prediction error and the load prediction error are expressed by the following formula:

e_Lt＝P_LMt-P_LFt

in the formula, e_Wtpu、e_LtAnd respectively representing a new energy power generation prediction error and a load prediction error of tmin.

5. The method of claim 1, wherein: in the step S3, the new energy power generation prediction error probability density model is established based on kernel density estimation in a non-parametric model, and the load prediction error probability density model is established based on normal distribution fitting.

If a Gaussian kernel function is selected, the probability density of the new energy power generation prediction error is obtained

Expressed as:

the load prediction error probability density f (e)_Lt) Expressed as:

in the formula (I), the compound is shown in the specification,

nuclear density estimation for new energy power generation prediction error yield ratio, P_iIndicating installed capacity of wind power, e_WFor new energy power generation prediction error, n is new energyThe number of samples of the power generation prediction error is equal to one, and each new energy power generation prediction error sample corresponds to one e_WtpuValue, h is window width, e_WpuError output ratio is predicted for new energy, e is natural constant, sigma_LwPredicting standard deviation, μ, of error samples for the corresponding intra-class load_LwThe average of the error samples is predicted for the load within the corresponding class.

6. The method of claim 1, wherein: the process of obtaining the probability distribution of any kind of equivalent spares in step S4 is:

convolving the new energy power generation prediction error probability density and the load prediction error probability density of the corresponding category to obtain equivalent standby probability density;

integrating the probability density of the equivalent standby to obtain the probability distribution of the equivalent standby;

wherein the probability density of the equivalent spares is expressed as:

R_t＝e_Lt-e_Wt

in the formula, f (R)_t) For variable being equivalent reserve R_tThe probability density of the equivalent sparing.

7. The method of claim 1, wherein: in step S5, determining an optimal confidence based on the confidence optimization model with the minimum expected cost of the system backup, and obtaining the backup capacity of the prediction period based on the selected optimal confidence and the probability distribution of the equivalent backup of the category to which the prediction period belongs;

if the expected cost of the system standby is selected to be the sum of the standby purchase cost and the standby penalty cost, the standby purchase cost comprises up-regulation standby purchase cost UC and down-regulation standby purchase cost DC, and the standby penalty cost comprises invalid up-regulation standby penalty cost IURP, invalid down-regulation standby penalty cost IDRP, load shedding penalty cost CLPC and load shedding penalty cost WAPC;

the objective function of the confidence optimization model is expressed as:

min(CLPC×(R_epα-R_mpα)+WAPC×(R_enα-R_mnα)+UC×R_euα+DC×R_edα+IURP×(R_eiuα-R_miuα)+IDRP×(R_midα-R_eidα))

wherein CLPC (R) is_epα-R_mpα) Penalty cost for load shedding at confidence 1-2 alpha, R_epαFor load shedding corresponding up-regulation (MW), R_mpαThe load is cut to correspond to actual needs for standby;

WAPC×(R_enα-R_mnα) The penalty cost of cutting machine under the confidence coefficient of 1-2 alpha, R_enαFor standby under corresponding cutting machine, R_mnαThe cutter is needed for standby correspondingly and actually;

UC×R_euαand DC × R_edαRespectively, the standby purchase cost R under the confidence coefficient 1-2 alpha_euαAnd R_edαRespectively for up-regulation and down-regulation;

IURP×(R_eiuα-R_miuα) For invalid upshifts with confidence 1-2 alpha, reserve penalty cost, R_eiuαAnd R_miuαRespectively corresponding to invalid up-regulation under the confidence coefficient of 1-2 alpha for up-regulation standby and actually required standby;

IDRP×(R_midα-R_eidα) Reserve penalty cost, R, for invalid down-regulation at confidence 1-2 alpha_eidαAnd R_midαRespectively corresponding to invalid down regulation for standby and actual required standby;

the selected confidence coefficient also meets the following constraint conditions:

1-2α≥C_min

in the formula, C_minIs a preset minimum confidence level.

8. A system based on the method of any one of claims 1-7, characterized by: the method comprises the following steps:

the historical data acquisition module is used for acquiring power generation historical data and load historical data in a historical period in a power grid containing the new energy unit;

the preprocessing module is used for taking a new energy power generation prediction error and a load prediction error which are calculated based on the power generation historical data and the load historical data as samples;

the classification module is used for classifying the new energy power generation prediction error and the load prediction error corresponding to the historical period;

the probability density model building module is used for respectively building a new energy power generation prediction error probability density model and a load prediction error probability density model of corresponding categories based on the new energy power generation prediction error samples and the load prediction error samples of the corresponding categories;

the equivalent standby probability distribution building module is used for calculating the probability distribution of the equivalent standby of each category based on the new energy power generation prediction error probability density model and the load prediction error probability density model;

the method comprises the following steps that an equivalent standby is the difference between a load prediction error and a new energy power generation prediction error, and the type of the equivalent standby is determined according to the type of a new energy power generation prediction error probability density model and the type of a load prediction error probability density model;

the confidence coefficient selecting module is used for selecting the confidence coefficient;

and the spare capacity acquisition module is used for acquiring the spare capacity of the prediction time period based on the selected confidence coefficient and the probability distribution of the equivalent spare of the class to which the prediction time period belongs.

9. A terminal, characterized by: comprising a processor and a memory, the memory storing a computer program that the processor calls to implement:

the process steps of any one of claims 1 to 7.

10. A readable storage medium, characterized by: a computer program is stored, which is invoked by a processor to implement:

the process steps of any one of claims 1 to 7.

Images