Optimization method for wind-storage station power reporting in two stages: day-ahead and intraday

Abstract

The present invention discloses a method for optimizing the two-stage power reporting of a wind-storage station, namely, day-ahead and day-intraday, based on the difference between meteorological data and the prediction time step, and constructing a day-ahead and day-intraday wind power scenario set based on the binned prediction error sample set; based on the day-ahead and day-intraday wind power scenario set obtained in step 1, with the minimum expected assessment power as the optimization goal, constructing a day-ahead wind power optimization reporting model and a day-intraday wind-storage power optimization reporting model considering energy storage regulation; taking into account the multiple nonlinear characteristics of the day-ahead and day-intraday reporting optimization problem, an optimization algorithm is used to solve the above-mentioned optimization reporting model to obtain an optimized reporting strategy. While considering the uncertainty of wind power, the present invention can deeply integrate the optimization of different time scales, thereby effectively improving the accuracy of wind-storage station power reporting and reducing the assessment power.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind-storage station power reporting strategy optimization, and specifically relates to a method for optimizing wind-storage station power reporting in two stages: day-ahead and intra-day. Background Art

[0002] Driven by the goals of "carbon peak and carbon neutrality" and the policy of building a new power system dominated by new energy, my country's wind power installed capacity and power generation continue to grow. In 2021, my country's wind power installed capacity reached 328 million kilowatts, and wind power generation reached 652.6 billion kilowatt-hours, a year-on-year increase of 17% and 40.5%. The intermittent and fluctuating characteristics of large-scale wind power require the power grid to arrange more flexible adjustment resources to achieve the safe absorption of wind power, resulting in an increase in the operation and maintenance costs of the power grid. Since 2020, local energy regulatory bureaus have successively signed and issued the new "Implementation Rules for the Management of Grid-Connected Operation of Power Plants" and "Implementation Rules for the Management of Auxiliary Services of Grid-Connected Power Plants" (two rules), which punish wind farms for inaccurate power forecasts in the form of electricity assessments, thereby allocating the operation and maintenance costs caused by forecast deviations to wind farms. Therefore, reasonable optimization of the reported power of wind farms will help improve the overall operational efficiency of wind farm grid-connected regulation.

[0003] The existing wind / wind-storage station power reporting strategy optimizes the reporting time scale relatively simply and lacks the specificity of the two detailed assessment mechanisms. On the one hand, most power reporting strategies are limited to the day-ahead scale. However, the two current detailed regulations both impose assessment requirements on reported power at the day-ahead and intraday scales. On the other hand, the existing power reporting model mainly uses market-based electricity prices, ancillary service compensation, and electricity sales revenue as optimization targets, which is far from the electricity quantity assessment form in the two detailed regulations. Therefore, there is an urgent need to develop a new optimization method for wind-storage station power reporting in the two stages of day-ahead and intraday to solve the problems existing in the existing technology. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the existing technology and provide an optimization method for the two-stage power reporting of wind-storage stations before and during the day. This method can deeply integrate the optimization of different time scales while considering the uncertainty of wind power, thereby effectively improving the accuracy of wind-storage station power reporting and reducing the assessed power consumption.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A method for optimizing power reporting in two stages, day-ahead and day-intraday, for a wind-storage station includes the following steps:

[0007] Step 1: Bin the day-ahead and day-ahead forecast errors based on the difference between meteorological data and forecast time steps, and construct the day-ahead and day-ahead wind power scenario sets based on the binned forecast error sample sets;

[0008] Step 2: Based on the day-ahead and intraday wind power scenario sets obtained in Step 1, with the minimum expected assessment power as the optimization goal, a day-ahead wind power optimization reporting model and an intraday wind-storage power optimization reporting model considering energy storage regulation are constructed;

[0009] In step 3, considering the multiple nonlinear characteristics of the day-ahead and intraday reporting optimization problems, an optimization algorithm is used to solve the day-ahead wind power optimization reporting model and the intraday wind-storage power optimization reporting model to obtain the optimized reporting strategy.

[0010] Furthermore, step 1 specifically includes:

[0011] Step 1.1: Calculate the forecast errors for the day-ahead and intraday phases based on the historical actual power and forecast power of the wind farm;

[0012] Step 1.2: bin the day-ahead forecast error according to the meteorological mode; bin the intraday forecast error according to the forecast time step;

[0013] Step 1.3: Sample the samples in each forecast error bin for the day-ahead phase and the intraday phase in step 1.2 and calculate the corresponding day-ahead and intraday forecast error scenarios. Then, use the Cholesky decomposition method to reorder the position of each row element in the day-ahead and intraday forecast error scenarios taking into account the time correlation to obtain a set of day-ahead and intraday forecast error scenarios that meet the time correlation.

[0014] Step 1.4: Superimpose each prediction error scenario in the day-ahead and intraday prediction error scenario sets obtained in step 1.3 with the corresponding prediction value to obtain the corresponding wind power scenario set.

[0015] Furthermore, in step 1.2, the improved Kmeans algorithm is used to bin the forecast error of the day-ahead stage.

[0016] Furthermore, the specific steps of using the improved Kmeans algorithm to bin the intraday forecast error are as follows:

[0017] Step 1.2.1: Based on the historical meteorological dataset and the historical wind power prediction error dataset at the same sampling time, use the elbow method to determine the number of clusters W;

[0018] Step 1.2.2: Initialize the number of iterations z = 1, calculate the data density of each meteorological data sample, and select the initial cluster center candidate set W O , and randomly select Z from maxPoints are used as the candidate set W of the first initial cluster center O,1 ;

[0019] Step 1.2.3: Get W O,1 The zth point in is taken as the first initial cluster center and W is calculated. O The distance between the rest of the points and the cluster is taken as the second initial cluster center, and then W is calculated. O The remaining samples are closest to the two, and the farthest point is taken as the third initial cluster center, and so on, until K initial cluster centers are obtained;

[0020] Step 1.2.4: Based on the determined initial cluster centers, perform classic K-means clustering calculation on the historical meteorological dataset to obtain the corresponding clustering results;

[0021] Step 1.2.5: Based on the clustering results, the historical wind power forecast error dataset is binned and the kernel density estimation method is used to obtain the wind power forecast error probability density curves under different bins.

[0022] Step 1.2.6: Calculate the cumulative root mean square error (SRMSE) of the clustering target corresponding to the clustering result, compare the SRMSE of two adjacent clusterings, and save the clustering result corresponding to the larger one;

[0023] Step 1.2.7: Repeat steps 1.2.3 to 1.2.6 until the SRMSE is greater than the set threshold λ or the maximum number of iterations Z is reached. max ;

[0024] Step 1.2.8: Based on the meteorological data clustering results obtained in step 1.2.7, the corresponding historical wind power forecast error dataset is divided into different meteorological modes to implement meteorological mode-based forecast error binning processing.

[0025] Furthermore, step 2 specifically includes:

[0026] Using the day-ahead wind power scenario set obtained in step 1, by minimizing the expected assessment power under each scenario, the objective function of the day-ahead wind power optimization reporting model is constructed as follows:

[0027]

[0028] Where: x DA The variable set to be optimized is reported on the previous day; is the power reported by the wind farm at time t; T is the total number of dispatching periods, P DA,all is the total expected assessment power in the day ahead; I is the total number of scenarios; is the assessment power of the i-th scenario in the previous day;

[0029] The corresponding upper and lower limits of the reported power for the above model are:

[0030]

[0031] The corresponding constraints for the calculation of power consumption in the above model are:

[0032]

[0033]

[0034] Where: P cap Installed capacity for wind farms; is the day-ahead power reporting accuracy of the i-th scenario; is the wind power of the i-th scenario predicted at time t on the day before.

[0035] Furthermore, in step 2, when constructing the intraday wind-storage power optimization reporting model taking into account energy storage regulation, it is necessary to take into account the established day-ahead reporting curve and change the actual grid-connected power through the power correction of energy storage. That is, when the wind power for the next n hours (4n steps) is predicted and optimized at time t in the day, the wind power at time t+4n is reported, while the energy storage corrects the actual grid-connected power at time t+1. There is a time difference between the two. Therefore, the intraday wind-storage power optimization reporting model is divided into the following according to the time period: reporting model for the period t+4n≥1&t+1<1, reporting model for the period t+1≥1&t+4n≤T, and reporting model for the period t+4n>T&t+1≤T, where T is the total number of scheduling periods. After the wind-storage optimization reporting in each period is completed, the reported power at the nth hour is recorded. Energy storage charging and discharging power and Residual energy This is used for rolling scheduling calculations at subsequent times.

[0036] Furthermore, the objective function of the reporting model in the period t+4n≥1&t+1<1 is:

[0037]

[0038]

[0039]

[0040] Where: x ID is the set of variables to be optimized reported within the day, P ID,all is the total expected assessment power in a day; I is the total number of scenarios; is the daily assessment electricity in the i-th scenario; is the intraday reporting accuracy of the i-th scenario; P capInstalled capacity for wind farms; For the intraday forecast of wind power in the i-th scenario at the j-th step at time j, is the reported power at time j within the day and is the variable to be optimized. Since energy storage is not introduced, during this period, only the root mean square error (RMS) between all reported power from 0:00 onwards and the power of each scenario needs to be calculated as the accuracy rate for each scenario. This is then used in the power consumption calculation to achieve optimal power reporting for the first n hours of each day.

[0041] Furthermore, the objective function of the reporting model in the period t+1≥1&t+4n≤T is:

[0042]

[0043] The energy storage constraints are:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] The calculation formulas for the accuracy of the day-ahead and day-intraday reporting are:

[0051]

[0052]

[0053] Where: x ID Report the variable set to be optimized within the day; and are binary variables of energy storage discharge and charging status at time t+1 respectively; and The energy storage discharging and charging power at time t+1; are the maximum values ​​of energy storage charging and discharging power respectively; is the stored energy at time t+1, is the stored energy at time t, is the total energy storage; P ID,all is the total expected assessment power in a day; I is the total number of scenarios; is the daily assessment electricity in the i-th scenario; c ID is the intraday assessment weight coefficient; c DA is the day-ahead assessment weight coefficient; is the assessment power of the i-th scenario in the previous day; are the energy storage charging and discharging power at time t+1, η ch ,η dis is the charging and discharging efficiency, μ up 、μ down is the upper and lower limit coefficient of energy storage; μ deep is the discharge depth coefficient; is the discharge power at time j; is the charging power at time j; For the intraday forecast, the wind power of the i-th scenario is predicted in the first step at time t+1; is the wind-storage grid-connected power scenario value after energy storage correction in the i-th scenario at time t+1; is the day-ahead power reporting accuracy of the i-th scenario; is the intraday reporting accuracy of the i-th scenario; is the actual grid-connected power value of wind-storage at the jth moment; The power reported by the wind farm at time t+1; P is the power reported before time j; cap is the total installed capacity of the wind farm; is the power reported within the day at time j; The power reported within the day at time t+1; It is the wind power of the i-th scenario predicted at the j-th step in the intraday forecast j time.

[0054] Furthermore, the objective function of the reporting model in the period t+4n>T&t+1≤T is:

[0055]

[0056] The calculation formulas for the accuracy of the day-ahead and day-intraday reporting are:

[0057]

[0058]

[0059] Where: x ID Report the variable set to be optimized within the day; and are binary variables of energy storage discharge and charging status at time t+1 respectively; and P is the energy storage discharge and charging power at time t+1; ID,all is the total expected assessment power in a day; I is the total number of scenarios; is the daily assessment power consumption under the i-th scenario; c ID is the intraday assessment weight coefficient; c DAis the day-ahead assessment weight coefficient; is the assessment power of the i-th scenario in the previous day; are the energy storage charging and discharging power at time t+1 respectively, is the discharge power at time j; is the charging power at time j; is the wind-storage grid-connected power scenario value after energy storage correction in the i-th scenario at time t+1; is the day-ahead power reporting accuracy of the i-th scenario; is the intraday reporting accuracy of the i-th scenario; is the actual grid-connected power value of wind-storage at the jth moment; The power reported by the wind farm at time t+1; P is the power reported before time j; cap is the total installed capacity of the wind farm; is the power reported within the day at time j; The power reported within the day at time t+1; It is the wind power of the i-th scenario predicted at the j-th step in the intraday forecast j time.

[0060] Furthermore, in step 3, the day-ahead wind power optimization reporting model is solved based on the particle swarm optimization algorithm, and the intraday wind-storage power optimization reporting model is solved using a small step-size linearized iterative algorithm.

[0061] Compared with the existing technology, the present invention has the following advantages: It aims to optimize wind farm power reporting in the day-ahead and intraday phases within the context of two detailed assessment criteria. While accounting for wind power uncertainty, it can deeply integrate optimization across different timescales and fully utilize forecast data and energy storage regulation resources at different timescales. Compared with traditional power reporting methods that report based on forecast values ​​or optimize reports separately at the day-ahead and intraday scales, this method can further improve the accuracy of wind farm power reporting and enhance the pertinence of the two detailed assessment criteria, thereby increasing reporting accuracy and reducing assessed power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a schematic diagram of daily rolling optimization reporting for a solar-storage power station according to an embodiment of the present invention;

[0063] Figure 2 This is a flowchart of intraday rolling optimization reporting in an embodiment of the present invention;

[0064] Figure 3 Schematic diagram of intraday wind-storage reporting and energy storage correction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0066] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0067] The present invention will be further described below with reference to specific examples, but they are not intended to limit the present invention.

[0068] like Figure 1 As shown, the present invention discloses a method for optimizing power reporting of a wind-storage station in two stages, namely, day-ahead and day-intraday, which specifically includes the following steps:

[0069] Step 1: Bin the day-ahead and day-ahead forecast errors based on the differences between meteorological data and forecast time steps, and construct the day-ahead and day-ahead wind power scenario sets based on the binned forecast error sample sets. This step specifically includes:

[0070] Step 1.1: Calculate the prediction error;

[0071] The historical forecast errors for the day-ahead and intraday phases can be obtained by subtracting the historical actual power from the forecast power of the wind farm, as shown in equations (1) and (2):

[0072]

[0073]

[0074] Where: T and S are the total number of forecast periods and the number of intraday forecast steps, respectively. According to the current two detailed assessment mechanisms, their values ​​are 96 and 16 (corresponding to the 24-hour forecast and the intraday ultra-short-term 4-hour rolling forecast, with a time resolution of 15 minutes); D is the total number of historical days; are the day-ahead wind power forecast error, actual wind power and day-ahead forecast wind power at time t on the d-th historical day, respectively; are the wind power prediction error, actual wind power and predicted wind power for the day corresponding to the s-step prediction on the d-th historical day at time t. In this embodiment, all wind power prediction values ​​and actual values ​​are known quantities, so the corresponding and They can all be calculated using formulas (1)-(2).

[0075] Step 1.2: Binning of day-ahead forecast errors;

[0076] Since the input of the current day-ahead wind power forecast is mainly based on the four meteorological elements of numerical weather forecast (i.e., wind speed, wind direction, temperature, and air pressure), the nature of the forecast error under different meteorological conditions / modes is different. Therefore, in this embodiment, the forecast error of the day-ahead stage is binned according to the meteorological mode. In the process of binning the forecast error according to the meteorological mode, this embodiment adopts an improved Kmeans algorithm to perform adaptive clustering on the meteorological data, such as Figure 2 As shown in the figure, it divides meteorological data by maximizing the difference between the probability distributions of wind power forecast errors under different meteorological modes. This ensures that there are significant differences between the conditional probability distributions of forecast errors under different meteorological modes, thereby ensuring the application value of the binning process. The specific steps for binning wind power forecast errors based on this algorithm are as follows:

[0077] Step 1.2.1: Based on the historical meteorological dataset and the historical wind power prediction error dataset at the same sampling time, use the elbow method to determine the number of clusters W;

[0078] Step 1.2.2: Initialize the number of iterations z = 1, calculate the data density of each meteorological data sample, and select the points with the largest data density as the initial cluster center candidate set W O , randomly select Z from max Points are used as the candidate set W of the first initial cluster center O,1 ;The data density of a sample represents the number of other samples within a certain range of the sample;

[0079] Step 1.2.3: Get W O,1 The zth point in is taken as the first initial cluster center and W is calculated. O The distance between the rest of the points and the cluster is taken as the second initial cluster center, and then W is calculated. O The remaining samples are closest to the two, and the farthest point is taken as the third initial cluster center, and so on, until K initial cluster centers are obtained;

[0080] Step 1.2.4: Based on the determined initial cluster centers, perform classic K-means clustering calculation on the historical meteorological dataset to obtain the corresponding clustering results;

[0081] Step 1.2.5: Based on the clustering results, the historical wind power forecast error dataset is binned and the kernel density estimation method is used to obtain the wind power forecast error probability density curves under different bins (i.e., meteorological modes);

[0082] Step 1.2.6: Calculate the cumulative root mean square error (SRMSE) of the clustering target corresponding to the clustering result, compare the SRMSEs of two adjacent clusterings, and save the clustering result corresponding to the larger one; SRMSE represents the overall difference in the probability density curves of wind power prediction errors between meteorological models corresponding to the clustering results. The specific calculation is as shown in formula (3);

[0083]

[0084] Where: l i With l j represents the probability density curves of the wind power prediction error data set corresponding to meteorological model i and meteorological model j respectively; W represents the number of clusters; RMSE(·) represents the root mean square error between the two curves;

[0085] Step 1.2.7: Repeat steps 1.2.3 to 1.2.6 until the SRMSE is greater than the set threshold λ or the maximum number of iterations Z is reached. max ;

[0086] Step 1.2.8: Based on the meteorological data clustering results obtained in step 1.2.7, the corresponding historical wind power forecast error dataset is divided into different meteorological modes to implement meteorological mode-based forecast error binning processing.

[0087] Step 1.3: Intraday forecast error binning;

[0088] For the intraday ultra-short-term wind power forecast error, as the number of forecast steps increases, the time span between the forecast time and the current time becomes longer, so the corresponding forecast error tends to gradually increase. Therefore, the intraday forecast error is calculated according to the existing forecast time step, and all forecast error data at the same time step are counted and divided into a box, totaling S (in this embodiment, S = 16) boxes;

[0089] Step 1.4: Construct the scenario set of day-ahead wind power forecast error;

[0090] Based on step 1.2, we can obtain the day-ahead forecast error sample set for each bin (weather model) after binning. For the forecast error samples in each bin, we use the Cornish-Fisher series expansion containing fifth-order cumulants to obtain the quantile of its cumulative probability function (CDF), as shown in equations (4)-(10):

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] Where: ω=1,2,...,W is the prediction error bin number; and Represent the mean and standard deviation of the forecast error in the ωth bin on the day before; are the third, fourth, and fifth-order origin moments of the forecast error in the ωth bin the day before; is the day-ahead wind power forecast error in the ωth box; are the quantiles of the CDF probability q of the forecast error in the ωth box after standardization and de-standardization; N ω is the number of prediction error samples contained in the ωth box on the day before; ξ q is the probability q quantile of the standard normal distribution. From (4)-(10), we can see that the final CDF quantile Only samples with each prediction error bin And the given ξ q Therefore, it can be simplified as formula (11):

[0099]

[0100] Where: H(.) represents the polynomial operation contained in the Cornish-Fisher series.

[0101] Furthermore, Latin hypercube sampling (LHS) is used to perform equal probability interval sampling on the CDFs corresponding to the W prediction error bins to obtain an error scenario set containing I quantiles, which is equivalent to substituting q = i / (I+1), i = 1, 2, ..., I into equations (8)-(9) to obtain the corresponding

[0102] After obtaining the day-ahead weather forecast data, the SVM is trained based on the mapping relationship between weather data and forecast error bins in step 1.2, and the forecast error bins to which the day-ahead weather forecast data belongs are matched to obtain the preliminary day-ahead forecast error scenario F. DA , as shown in formula (12):

[0103]

[0104] Considering that the prediction errors of different prediction periods have time correlation, the F DAArranging them only according to the size of the probability value does not conform to the correlation of historical forecast error samples.

[0105] Therefore, the following describes how to use the Cholesky decomposition method to implement F DA Reordering to account for time dependencies:

[0106] Step 1.4.1: Count historical forecast error samples e DA , and calculate the correlation coefficient matrix C for each forecast period by row DA , as shown in formula (13):

[0107]

[0108] Step 1.4.2: Randomly generate a T×I-dimensional matrix S0, where each row consists of a random permutation of integers 1, 2, …, I, representing the current F DA Sorting method (sequential sorting);

[0109] Step 1.4.3: Calculate the correlation coefficients between the rows of the matrix S0 to obtain the correlation coefficient matrix C0, and perform Cholesky decomposition on C0 to obtain the lower triangular matrix L0; DA Perform Cholesky decomposition to obtain the lower triangular matrix L:

[0110]

[0111]

[0112] Step 1.4.4: Calculate the sorting matrix S1 according to formula (15), and sort F according to the size sequence position of each row element of matrix S1. DA The elements of each row are reordered to obtain the day-ahead forecast error scenario set Δp that satisfies the time correlation. DA .

[0113] Step 1.5: Construct the scenario set of intraday wind power forecast error;

[0114] Using the prediction error samples binned according to the prediction time step in step 1.3, the Cornish-Fisher series is also used to establish its CDF quantile as follows:

[0115]

[0116] Similar to step 1.4, LHS sampling is used to obtain the initial I prediction error scenarios for the s-th step prediction error as follows:

[0117]

[0118] Time correlation sorting of intraday forecast errors First, count the historical forecast error samples at each forecast time step within the day, and calculate the correlation coefficient between each time step:

[0119]

[0120] C ID With F ID Replace C respectively DA With F DA Execute steps 1.4.2-1.4.5 to obtain the scenario set Δp that takes into account the temporal correlation of the intraday forecast error. ID .

[0121] Step 1.6: Construction of day-ahead and intraday wind power scenario sets;

[0122] After obtaining the day-ahead and day-intraday forecast error scenario sets according to steps 1.4 and 1.5, superimposing them with the corresponding forecast values ​​can obtain the corresponding wind power scenario sets, as shown in equations (19) and (20).

[0123]

[0124]

[0125] Where: is the wind power forecast value at time t the day before; is the wind power forecast value at the sth step at time t within the day; is the wind power of the i-th scenario at time t the day before; is the wind power of the i-th scenario at the s-th step within the day t.

[0126] Step 2: Based on the day-ahead and intraday wind power scenario sets obtained in Step 1, with the minimum expected assessment power as the optimization goal, a day-ahead wind power optimization reporting model and an intraday wind-storage power optimization reporting model considering energy storage regulation are constructed;

[0127] Step 2.1: Construct a day-ahead wind power optimization reporting model;

[0128] Both current regulations introduce the concept of assessed power in their assessment requirements for wind farm day-ahead power reporting, penalizing inaccurate forecasts. To this end, this model utilizes the set of day-ahead wind power scenarios obtained in step 1.6 to achieve optimal day-ahead power reporting by minimizing the expected assessed power under each scenario. The objective function of the constructed day-ahead wind power optimization reporting model is shown in Equation (21):

[0129]

[0130] Where: x DA The variable set to be optimized is reported on the previous day; P is the power reported by the wind farm at time t; DA,all is the total expected assessment power on the day before; T is the total number of scheduling periods; I is the total number of scenarios; is the assessment power of the i-th scenario on the previous day. The corresponding upper and lower limit constraints of the reported power of this model are shown in formula (22), and the assessment power calculation constraints are shown in formulas (23)-(24):

[0131]

[0132]

[0133]

[0134] Where: P cap Installed capacity for wind farms; is the day-ahead power reporting accuracy of the i-th scenario, is the wind power of the i-th scenario predicted at time t on the day before.

[0135] Step 2.2: Construct a daily wind-storage power optimization reporting model that takes energy storage regulation into account;

[0136] Considering that the two current regulations have assessment requirements for both day-ahead and intra-day wind power reporting, it is also necessary to take into account the established day-ahead reporting curve when reporting intra-day power. By modifying the actual grid-connected power through energy storage power correction, the expected optimal reporting of both day-ahead and intra-day power can be achieved. On the other hand, when predicting / optimizing the wind power for the next n hours (a total of 4n steps) at time t within the day, in this embodiment, n=4 is selected, such as Figure 3 As shown, at time t, the wind power for the next 4 hours (16 steps) is predicted / optimized and reported, reporting the wind power at time t+16, while the energy storage corrects the actual grid-connected power at time t+1. There is a time difference between the two. Therefore, the daily wind-storage power optimization reporting model will be divided according to the time period. The specific division method is as follows:

[0137] Step 2.2.1: Establish a reporting model for the period t+16≥1&t+1<1;

[0138] During this period, power reporting has already begun, but energy storage has not yet been introduced for correction. Therefore, only the wind power reporting at time t+16 within the day needs to be considered. The objective function is shown in Equation (25), and the assessment of power consumption and reporting accuracy are calculated as shown in Equations (26)-(27):

[0139]

[0140]

[0141]

[0142] Where: x ID Report the variable set to be optimized within the day. is the reported power at time t+16 within the day, P ID,all is the total expected assessment power in a day; I is the total number of scenarios; is the daily assessment electricity in the i-th scenario; is the intraday reporting accuracy of the i-th scenario; P cap Installed capacity for wind farms; For the intraday forecast of wind power in the i-th scenario at the j-th step at time j, is the reported power at time j within the day and is the variable to be optimized. Since energy storage is not introduced, during this period, only the root mean square error (RMS) between all reported power from 0:00 onwards and the power of each scenario needs to be calculated as the accuracy rate for each scenario. This is also included in the power consumption calculation, thereby achieving optimal power reporting for the first four hours of each day (16:00).

[0143] Step 2.2.2: Report the model in the period t+1≥1&t+16≤T;

[0144] During this period, wind power reporting and energy storage power correction will be carried out simultaneously. Therefore, it is necessary to introduce the optimized reporting curve of the day-ahead. The day-ahead reporting curve is "finalized" by energy storage charging and discharging, so as to achieve the overall minimization of the day-ahead and intra-day assessment power. The objective function of the reporting model during this period is shown in Equation (28), and the energy storage constraints are shown in Equations (29)-(34). The calculation of the day-ahead and intra-day reporting accuracy is shown in Equations (35)-(36), and the assessment power calculation is shown in Equations (23) and (26):

[0145]

[0146]

[0147]

[0148]

[0149]

[0150]

[0151]

[0152]

[0153]

[0154] Where: x IDReport the variable set to be optimized within the day; and are binary variables of energy storage discharge and charging status at time t+1 respectively; and The energy storage discharging and charging power at time t+1; are the maximum values ​​of energy storage charging and discharging power respectively; is the stored energy at time t+1, is the stored energy at time t, is the total energy storage; P ID,all is the total expected assessment power in a day; I is the total number of scenarios; is the daily assessment power consumption under the i-th scenario; c ID is the intraday assessment weight coefficient; c DA is the day-ahead assessment weight coefficient; is the assessment power of the i-th scenario the day before; η ch ,η dis is the charging and discharging efficiency, μ up 、μ down is the upper and lower limit coefficient of energy storage; μ deep is the discharge depth coefficient; is the discharge power at time j; is the charging power at time j; For the intraday forecast, the wind power of the i-th scenario is predicted in the first step at time t+1; is the wind-storage grid-connected power scenario value after energy storage correction in the i-th scenario at time t+1; is the day-ahead power reporting accuracy of the i-th scenario; is the intraday reporting accuracy of the i-th scenario; is the actual grid-connected power value of wind-storage at the jth moment; The power reported by the wind farm at time t+1; P is the power reported before time j; cap is the total installed capacity of the wind farm; is the power reported within the day at time j; The power reported within the day at time t+1; It is the wind power of the i-th scenario predicted at the j-th step in the intraday forecast j time.

[0155] Formulas (29)-(32) are basic energy storage charge and discharge constraints. Considering the impact of discharge depth on energy storage life, the total energy storage charge and discharge in all periods of the day should not exceed a certain discharge depth, as shown in formula (33). and intraday The accuracy calculation is different from step 2.1 and step 2.2.1 respectively. The root mean square error (RMS) between the actual grid-connected power and the day-ahead reported value from 0:00 to the current time t, and the RMS error between the grid-connected power and the day-ahead reported value in each scenario at time t+1 are calculated respectively. The calculation is divided into three parts. The first part calculates the root mean square error (RMS) between the actual grid-connected power and the daily reported value from 0:00 to the current time t. The second part calculates the RMS error between each grid-connected power scenario and the corresponding daily reported value at time t+1. The third part calculates the RMS error between the wind power scenario and the corresponding reported value for the remaining 15 steps of the day, with the reported value at t+16 being the variable to be optimized. This calculation takes into account information from the actual time, the time to be corrected, and the time to be predicted, providing statistical information support for power calculation assessment and energy storage charging and discharging arrangements.

[0156] Step 2.2.3: Report the model during the period t+16>T&t+1≤T.

[0157] During this period, the intraday power reporting has ended, but the energy storage correction is still in progress. Its objective function is shown in Equation (37). The assessment power calculation is consistent with Equations (23) and (26). The energy storage model and the corrected scenario power at time t+1 are consistent with Equations (29)-(34). The day-ahead accuracy calculation is shown in Equation (35), and the intraday accuracy calculation is shown in Equation (38).

[0158]

[0159]

[0160] The same symbols in the above equation have the same meaning as in the previous equation. The only difference between equation (38) and (36) is that the accuracy of the last part of the future moment is a constant and does not include the decision variable for power reporting.

[0161] Step 2.2.4: Rolling update of intraday optimization data;

[0162] After the wind-storage optimization report for each period is completed, the power reported in the 4th hour is recorded. Energy storage charging and discharging power and With residual energy This is used for rolling scheduling calculations at subsequent times.

[0163] Step 3: Considering the multiple nonlinear characteristics of the day-ahead and intraday reporting optimization problems, an optimization algorithm is used to solve the day-ahead wind power optimization reporting model and the intraday wind-storage power optimization reporting model to obtain the optimized reporting strategy.

[0164] Step 3.1: Solve the day-ahead wind power optimization reporting model;

[0165] The Day-Ahead model is summarized as follows:

[0166] Objective function: Equation (21);

[0167] Constraints: Equations (22)-(24).

[0168] As can be seen, this optimization problem is a nonlinear one involving both logical and radical constraints. Given the small size of the problem and the relatively simple decision variables, the classic particle swarm optimization algorithm (PSO) is used directly for optimization. Given its simple principle and widespread application, the specific steps will not be elaborated on here.

[0169] Step 3.2: Solve the daily wind-storage power optimization reporting model;

[0170] The intraday stage model is summarized as follows:

[0171] Period 1: t+16≥1&t+1<1:

[0172] Objective function: Equation (25);

[0173] Constraints: Equations (26)-(27).

[0174] Period 2: t+1≥1&t+16≤T

[0175] Objective function: Equation (28);

[0176] Constraints: Equations (23), (26), (29)-(36).

[0177] Period 3: t+16>T&t+1≤T

[0178] Objective function: Equation (37);

[0179] Constraints: Equations (23), (26), (29)-(35), and (38).

[0180] Similar to the wind power optimization reporting model, the optimization models for each time period within a day are all nonlinear optimization problems. The difference is that optimization must be performed every 15 minutes within a day, and it is difficult to achieve high solution efficiency using the particle swarm algorithm. At the same time, the nonlinear part of the model only exists in the assessment power calculation (logical constraint) and accuracy calculation (radical constraint). Therefore, this embodiment designs a small step size linearized iterative algorithm to solve the intraday reporting model. The specific steps are as follows:

[0181] Step 3.2.1: Relaxation of logical constraints;

[0182] Using the Big-M method, equations (23) and (26) are transformed into equations (39) and (40), respectively, as follows:

[0183]

[0184]

[0185] Where: l1-l4 are 0 / 1 variables; M1-M 12 For a larger number.

[0186] Step 3.2.1: Initial value acquisition; Since the absolute value accuracy is positively correlated with the RMS accuracy to a large extent and is convenient for model solution, first replace equations (27), (35)-(36), and (38) with equations (41)-(44) and substitute them into the intraday reporting model of the corresponding time period to perform optimization, as follows:

[0187]

[0188]

[0189]

[0190]

[0191] The relevant absolute value constraints in the formula can also be linearized using the Big-M method in a form similar to formula (45):

[0192]

[0193] Where p1 and p2 correspond to the absolute value constraint internal sample variables, y err is an absolute value constraint example variable, is a 0 / 1 variable, is a larger number;

[0194] Optimization is completed and the initial value is recorded

[0195] Step 3.2.2: Small step linearization iteration;

[0196] The calculation of accuracy can be summarized as follows The polynomial shown in formula (46) with is a variable:

[0197]

[0198] Where: Z DA (.), Z ID (.) are polynomials for calculating the accuracy of the day and the day respectively. In the optimization model of each period, the small step constraint (47) is added, and Perform a first-order Taylor expansion of Equation (46) with centered on , and obtain the linearized RMS accuracy constraint, as shown in (48)-(49):

[0199]

[0200]

[0201]

[0202] Where Z DA (·) represents the polynomial for calculating the day-ahead accuracy, Z ID (·) represents the polynomial for calculating the intraday accuracy. Equations (48) and (49) are used to replace the RMS constraints of the day-ahead and intraday accuracy in each time period, and the intraday power reporting optimization problem is linearized. The obtained optimization result is used as the new initial value of the iteration to update

[0203] Step 3.2.3: Termination conditions;

[0204] 1) Repeat step 3.2.2 until two times The sum of the absolute value errors is less than the threshold ε;

[0205] 2) Reach the upper limit of the number of iterations r max .

[0206] The above are only preferred embodiments of the present invention and do not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of the present invention specification should be included in the protection scope of the present invention.

Claims

1. A method for optimizing power reporting in two stages, day-ahead and day-intraday, for a wind-storage station, characterized in that: The steps include: Step 1: Bin the day-ahead and day-ahead forecast errors based on the difference between meteorological data and forecast time steps, and construct the day-ahead and day-ahead wind power scenario sets based on the binned forecast error sample sets; Step 2: Based on the day-ahead and intraday wind power scenario sets obtained in Step 1, with the minimum expected assessment power as the optimization goal, a day-ahead wind power optimization reporting model and an intraday wind-storage power optimization reporting model considering energy storage regulation are constructed; Step 3: Considering the multiple nonlinear characteristics of the day-ahead and intraday reporting optimization problems, an optimization algorithm is used to solve the day-ahead wind power optimization reporting model and the intraday wind-storage power optimization reporting model to obtain the optimized reporting strategy. Among them, step 1 specifically includes: Step 1.1: Calculate the forecast errors for the day-ahead and intraday phases based on the historical actual power and forecast power of the wind farm; Step 1.2: bin the day-ahead forecast error according to the meteorological mode; bin the intraday forecast error according to the forecast time step; Step 1.3: Sample the samples in each forecast error bin for the day-ahead phase and the intraday phase in step 1.2 and calculate the corresponding day-ahead and intraday forecast error scenarios. Then, use the Cholesky decomposition method to reorder the position of each row element in the day-ahead and intraday forecast error scenarios taking into account the time correlation to obtain a set of day-ahead and intraday forecast error scenarios that meet the time correlation. Step 1.4: Superimpose each forecast error scenario in the day-ahead and intraday forecast error scenario sets obtained in step 1.3 with the corresponding forecast value to obtain the corresponding wind power scenario set; Step 2 specifically includes: Using the day-ahead wind power scenario set obtained in step 1, by minimizing the expected assessment power under each scenario, the objective function of the day-ahead wind power optimization reporting model is constructed as follows: ； Where: The variable set to be optimized is reported on the previous day; for t The wind farm reports power on a daily basis; T is the total number of scheduling periods, is the total expected assessment power on the previous day; I is the total number of scenes; For the day before i Assessment power for each scenario; The corresponding upper and lower limits of the reported power for the above model are: ； The corresponding constraints for the calculation of power consumption in the above model are: ； ； Where: Installed capacity for wind farms; For the i The accuracy of day-ahead power reporting for each scenario; Day-ahead forecast t Moment i Wind power for each scenario; In step 2, when constructing a daily wind-storage power optimization reporting model that takes energy storage regulation into account, it is necessary to take into account the established day-ahead reporting curve and modify the actual grid-connected power through energy storage power correction. That is, when the wind power forecast for the next n hours (4n steps) is optimized and reported at time t, the wind power at time t+4n is reported, while the energy storage correction is based on the actual grid-connected power at time t+1. There is a time difference between the two, so the daily wind-storage power optimization reporting model is divided into the following categories according to the time period: Reporting model under time period, Reporting model under time period and Reporting model in each time period, T is the total number of scheduling periods; after the wind-storage optimization reporting of each period is completed, the power reported in the nth hour is recorded , energy storage charging and discharging power and , residual energy , for use in rolling scheduling calculations at subsequent moments.

2. The optimization method for wind-storage station power reporting in two stages, day-ahead and day-intraday, according to claim 1, is characterized in that: In step 1.2, the improved Kmeans algorithm is used to bin the forecast error of the day-ahead stage.

3. The optimization method for wind-storage station power reporting in two stages, day-ahead and day-intraday, according to claim 2, is characterized in that: The specific steps of using the improved Kmeans algorithm to bin the intraday forecast error are as follows: Step 1.2.1: Based on the historical meteorological dataset and the historical wind power forecast error dataset at the same sampling time, use the elbow method to determine the number of clusters. ; Step 1.2.2: Initialize the number of iterations , calculate the data density of each meteorological data sample and select the initial cluster center candidate set , and randomly select Points are used as the candidate set for the first initial cluster center ; Step 1.2.3: Take Middle points as the first initial cluster center, calculate The distance between the rest of the points and the center is taken as the second initial cluster center, and then the The remaining samples in the cluster are closest to the two, and the farthest point is taken as the third initial cluster center, and so on, until the Initial cluster centers; Step 1.2.4: Based on the determined initial cluster centers, perform classic K-means clustering calculation on the historical meteorological dataset to obtain the corresponding clustering results; Step 1.2.5: Based on the clustering results, the historical wind power forecast error dataset is binned and the kernel density estimation method is used to obtain the wind power forecast error probability density curves under different bins. Step 1.2.6: Calculate the cumulative root mean square error of the clustering target corresponding to the clustering result , compare the two adjacent clusters Size, save the clustering result corresponding to the larger one; Step 1.2.7: Repeat steps 1.2.3 to 1.2.6 until Greater than the set threshold , or the maximum number of iterations is reached ; Step 1.2.8: Based on the meteorological data clustering results obtained in step 1.2.7, the corresponding historical wind power forecast error dataset is divided into different meteorological modes to implement meteorological mode-based forecast error binning processing.

4. The optimization method for wind-storage station power reporting in two stages, day-ahead and day-intraday, according to claim 1, is characterized in that: The objective function of the reporting model in the time period is: ； ； ； Where: Report the variable set to be optimized within the day. The total expected assessment power in the day; I is the total number of scenes; For the i Daily assessment of electricity consumption under different scenarios; For the i The accuracy of daily reporting for each scenario; Installed capacity for wind farms; For intraday forecasts j Moment jt Step prediction i The wind power of each scene, For the day j The reported power at the time is the variable to be optimized. Since energy storage is not introduced, during this period, it is only necessary to calculate the root mean square error between all reported powers from 0:00 on the day and the power of each scenario as the accuracy rate for each scenario, and include it in the assessment power calculation, so as to achieve the optimal power reporting in the first n hours of each day.

5. The optimization method for wind-storage station power reporting in two stages, day-ahead and day-intraday, according to claim 1, is characterized in that: The objective function of the reporting model in the time period is: ； The energy storage constraints are: ； ； ； ； ； ； The calculation formulas for the accuracy of the day-ahead and day-intraday reporting are: ； ； Where: Report the variable set to be optimized within the day; and They are t +1 moment energy storage discharge and charging state binary variable; and for t +1 moment energy storage discharge and charging power; 、 They correspond to the maximum charging and discharging power of energy storage respectively; for t+1 Store energy at all times, for t Store energy at all times, is the total energy stored; The total expected assessment power in the day; I is the total number of scenes; For the i Daily assessment of electricity consumption under different scenarios; c ID is the intraday assessment weight coefficient; c DA is the day-ahead assessment weight coefficient; For the day before i Assessment power for each scenario; 、 They are t+ 1 moment energy storage charging and discharging power, 、 is the charging and discharging efficiency, 、 are the upper and lower limit coefficients of the energy storage capacity; is the discharge depth coefficient; for j Discharge power at all times; for j Charging power at all times; For intraday forecasts t+ 1 moment 1 step prediction i Wind power for each scenario; for t+ 1st moment i The wind-storage grid-connected power scenario value after energy storage correction under each scenario; For the i The accuracy of day-ahead power reporting for each scenario; For the i The accuracy of daily reporting for each scenario; For the j Actual grid-connected power value of wind-storage at each moment; for t+1 The wind farm reports power on a daily basis; for j Report power before the time limit; P cap Installed capacity for wind farms; for j Report power within a certain time period; for t +1 moment to report power within the day; For intraday forecasts j Moment jt Step prediction i Wind power for each scenario.

6. The optimization method for wind-storage station power reporting in two stages, day-ahead and day-intraday, according to claim 1, is characterized in that: The objective function of the reporting model in the time period is: ； The calculation formulas for the accuracy of the day-ahead and day-intraday reporting are: ； Where: Report the variable set to be optimized within the day; and They are t +1 moment energy storage discharge and charging state binary variable; and for t +1 moment energy storage discharge and charging power; The total expected assessment power in the day; I is the total number of scenes; For the i Daily assessment of electricity consumption under different scenarios; c ID is the intraday assessment weight coefficient; c DA is the day-ahead assessment weight coefficient; For the day before i Assessment power for each scenario; 、 They are t+1 Energy storage charging and discharging power at all times, for j Discharge power at all times; for j Charging power at all times; for t+ 1st moment i The wind-storage grid-connected power scenario value after energy storage correction under each scenario; For the i The accuracy of day-ahead power reporting for each scenario; For the i The accuracy of daily reporting for each scenario; For the j Actual grid-connected power value of wind-storage at each moment; for t+1 The wind farm reports power on a daily basis; for j Report power before the time limit; P cap Installed capacity for wind farms; for t +1 moment to report power within the day; for t +1 time reported power before the day; For intraday forecasts j Moment jt Step prediction i Wind power for each scenario.

7. The optimization method for wind-storage station power reporting in two stages, day-ahead and day-intraday, according to claim 1, is characterized in that: In step 3, the particle swarm algorithm is used to solve the day-ahead wind power optimization reporting model, and the small-step linearization iterative algorithm is used to solve the day-ahead wind-storage power optimization reporting model.

Images