A wind power cluster power ultra-short-term prediction error correction method

Abstract

The wind power cluster power ultra-short-term prediction error correction method has the characteristics that the harm of wind power prediction error to load peak-valley period is considered, the wind power ultra-short-term prediction error obtained by preliminary prediction is corrected according to load peak-valley period, simulation calculation and error analysis steps are carried out, the seasonal characteristics of wind power prediction error are considered to respectively correct load valley period I, peak period I, valley period II and peak period II in a day, and based on the correction, harmful error of load peak-valley period is effectively reduced, wind power curtailment loss in load valley period is also relatively reduced, and the system operation economy is improved; the correction model provided by the application is a model suitable for multi-model preliminary prediction and correction, has simple calculation, high prediction performance, clear physical meaning, effective prediction result and strong practicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind power prediction, and is a wind power cluster power ultra-short-term prediction error correction method. BACKGROUND

[0002] Wind power ultra-short-term prediction can predict wind power in the future 15min-4h at a time, which provides a basis for short-time scale scheduling. However, due to the influence of factors such as climate and season, wind power prediction has certain output regularity and error distribution regularity at different times. These errors show certain seasonal characteristics, and different types of wind power prediction errors have different influences on the safe and stable operation of the power system.

[0003] Load peak and valley periods are key periods of power consumption, and power fluctuations are more severe in these periods. Therefore, the prediction accuracy of wind power in these periods is particularly important. Large-scale new energy grid connection significantly reduces the system peak shaving capability, and the volatility of wind power may cause the load supply and demand balance to be broken. High-precision new energy output prediction is of great significance to power scheduling in load peak and valley periods. In the load valley period, if the wind power prediction value is low and the actual value is high, the power will be excessive in the actual operation of the valley period, which will seriously affect the power balance. At the same time, the excess wind power will be lost in the form of curtailment, and the system will consume more conventional energy, which will seriously affect the system operation economy. In the load peak period, if the prediction value is high and the actual value is low, the scheduling arrangement based on the prediction result will cause the system frequency to drop in the peak period, causing a series of secondary hazards and affecting the system power supply reliability.

[0004] Existing wind power ultra-short-term prediction mostly considers how to improve the prediction accuracy of wind power, and rarely considers the influence of negative errors (high prediction value) in the load peak period and positive errors (low prediction value) in the load valley period on load power balance. Such improvement of power prediction accuracy may not be suitable for the power scheduling needs of the load peak and valley periods, and may not meet the load demand. Harmful errors will still affect the safe and stable operation of the power system. SUMMARY

[0005] The purpose of the present application is to overcome the shortcomings of the prior art, and to provide a wind power cluster power ultra-short-term prediction error correction method that considers reducing harmful errors in the load peak and valley periods, improving the prediction accuracy of wind power, improving the economic efficiency of the power generation side, and improving the energy utilization rate of the dispatching side.

[0006] The technical scheme adopted to achieve the purpose of the present application is: a wind power ultra-short-term prediction error correction method, characterized in that it comprises the following steps:

[0007] 1) Preliminary forecast of wind power in the ultra-short term

[0008] To correct harmful errors in wind power output during peak and off-peak periods, a preliminary power prediction of wind power is first required. Let the mapping of the prediction model be F(), and the input for wind power prediction be: X=[x1,x2,...,x n ] T Where n represents the number of input features, the power prediction result of the corresponding wind power prediction model is expressed as:

[0009] P=F(X) (1)

[0010] Where P represents the predicted wind power output;

[0011] 2) Determine the peak and valley periods of load.

[0012] Since load sequences typically have two trough periods and two peak periods within a day, separate correction methods are used for peak periods and trough periods to reduce harmful errors. This method is based on the analysis of a year's load data, and the time when the load peak and trough periods occur together for each day of the year is taken as the load peak and trough time, including load trough period I, load peak period I, load trough period II, and load peak period II.

[0013] 3) Determine the correction interval

[0014] Based on the duration of load peak and valley periods, a correction interval for harmful errors in wind power prediction is established within the duration of load peak and valley periods. This corrects for negative errors during peak periods and positive errors during valley periods. Since load peak or valley values ​​are not instantaneous values ​​but rather durations, if the load is at a valley within that duration, it is considered that the load demand is low and the generator output is low; conversely, if the load is at a peak, it is considered that the load demand is high and the generator output is consistently high. The correction interval is defined as follows: with the load peak and valley times as the center of the correction interval and r as the radius, the length of the correction interval at a certain load peak or valley is 2r. r is adjusted according to actual conditions, and the unit of r is (15 min). -1 r reflects the step size of the period to be corrected, but the duration of the load peak and valley values ​​should be considered in actual correction to ensure effective correction during the load peak and valley periods. There will be four correction intervals within a day, namely, load valley correction interval I, load peak correction interval I, load valley correction interval II and load peak correction interval II.

[0015] 4) Determine the peak and valley periods of the load to be corrected.

[0016] After determining the correction interval radius, the correction interval of each peak-valley value period is obtained, and the average value of the difference between the historical wind power prediction value and the actual value in the correction interval is taken as the basis for judging the positive and negative error, and the calculation formula is:

[0017]

[0018] Wherein, N represents the length of the correction interval, and N is taken as 2r here; P i represents the actual value of the load peak or valley period wind power at the ith time point; represents the prediction value; Error represents the average prediction error of the peak-valley value period correction interval, and the error of two peak values and two valley values within a day is calculated, which reflects the positive and negative situation of the average wind power prediction error of the load peak or valley period within a day, and is used to judge the error type mainly existing in the load peak-valley period; At this time, it is defined that the number of days with positive Error is subtracted from the number of days with negative Error in a certain time period to measure the degree of harmful error more or less in the valley period, and the greater the difference value, the more harmful error in this time scale, and the more the period needs to be corrected; Similarly, the greater the difference value of the number of days with negative Error minus the number of days with positive Error, the more harmful error in the peak period, and the more the period needs to be corrected. The historical wind power prediction error is analyzed to determine the month corresponding to the load peak-valley period which needs to be corrected;

[0019] 5) Determining the correction coefficient of the load peak-valley period

[0020] In order to minimize the number of days of harmful error of ultra-short-term wind power prediction, the actual value and the prediction value of the load peak-valley period with high harmful error frequency in the previous year are selected, and the average value of the ratio in the entire selected period is taken as the correction coefficient k, and the specific expression is:

[0021]

[0022] Wherein, k represents the correction coefficient; m represents the number of days of the load peak or valley period to be corrected; r represents the radius of the correction interval;

[0023] 6) Correction of wind power prediction error in load peak-valley period

[0024] Based on the determination of the correction time period and the correction interval of the load peak-valley period, when the wind power in the to-be-corrected period needs to be corrected, the correction coefficient corresponding to the load peak-valley period in the previous year is multiplied by the wind power prediction value to be corrected to obtain the corrected wind power prediction value, and the correction result is:

[0025] P i *= P i x k i (4)

[0026] wherein, P i * represents the wind power correction result of the i th load period, k i represents the correction coefficient of the i th load period; when i = 1, 2, 3, 4, it represents load valley period I, load peak period I, load valley period II, load peak period II respectively;

[0027] 7) Simulation calculation

[0028] Simulation input: the measured data of the wind farm and the load data are analyzed to determine the total installed capacity of the farm; input data: wind farm historical power and numerical weather prediction; the data sampling interval is 15min; according to steps 1) to 5), the wind power ultra-short-term preliminary prediction result of the prediction period is obtained;

[0029] 8) Error analysis

[0030] In order to measure the effectiveness of the power prediction result and the error correction result, the mean absolute error (mean absolute error, MAE) and the root mean square error (root mean square error, RMSE) are used to evaluate the predicted and corrected wind power values, and the calculation formula is:

[0031]

[0032]

[0033] Wherein, C represents the number of point prediction of the prediction sequence; Cap represents the rated installed capacity.

[0034] The wind power cluster power ultra-short-term prediction error correction method provided by the application is based on the preliminary power prediction result of any model; according to the seasonal characteristics of wind power prediction error, the wind power correction coefficient of the current to be corrected period is calculated according to the adjacent historical load peak and valley period, the harmful error in the load peak and valley period is corrected according to the size of the correction interval, the negative harmful error in the load peak period and the positive harmful error in the load valley period are reduced, the wind power ultra-short-term power prediction accuracy is increased, the wind power loss is reduced, the positive significance of wind power prediction result to power dispatching is improved, the method has clear physical meaning, effective prediction result and strong practicability. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 Schematic diagram for dividing load peak and valley periods;

[0036] Figure 2 Fig. 2 is a schematic diagram of the proportion of the time of occurrence of the load valley;

[0037] Figure 3 Fig. 3 is a schematic diagram of the proportion of the time of occurrence of the load peak;

[0038] Figure 4 Fig. 4 is a schematic diagram of the corrected interval of the load peak-valley period;

[0039] Figure 5 Fig. 5 is a schematic diagram of the technical framework of the wind power prediction error correction method;

[0040] Figure 6a Fig. 6 is a monthly distribution characteristic diagram of the positive error difference of the load valley period I;

[0041] Figure 6b Fig. 7 is a monthly distribution characteristic diagram of the positive error difference of the load valley period II;

[0042] Figure 6c Fig. 8 is a monthly distribution characteristic diagram of the negative error difference of the load peak period I;

[0043] Figure 6d Fig. 9 is a semi-monthly distribution characteristic diagram of the negative error difference of the load peak period II;

[0044] Figure 7a Fig. 10 is a semi-monthly distribution characteristic diagram of the positive error difference of the load valley period I;

[0045] Figure 7b Fig. 11 is a semi-monthly distribution characteristic diagram of the positive error difference of the load valley period II;

[0046] Figure 7c Fig. 12 is a semi-monthly distribution characteristic diagram of the negative error difference of the load peak period I;

[0047] Figure 7d Fig. 13 is a semi-monthly distribution characteristic diagram of the negative error difference of the load peak period II;

[0048] Figure 8a Fig. 14 is a result diagram before and after error correction of the load valley period I;

[0049] Figure 8b Fig. 15 is a result diagram before and after error correction of the load valley period II;

[0050] Figure 8c Fig. 16 is a result diagram before and after error correction of the load peak period I;

[0051] Figure 8d Fig. 17 is a result diagram before and after error correction of the load peak period II;

[0052] Figure 9 Fig. 18 is a schematic diagram of the error correction amount under different correction radii from 2017 to 2019. DETAILED DESCRIPTION

[0053] The application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] With reference to Figure 1 The wind power cluster power ultra-short-term prediction error correction method of the application comprises the following steps:

[0055] 1) Ultra-short-term wind power preliminary prediction: obtaining a preliminary prediction result of wind power in a to-be-predicted period based on any wind power prediction model under given input; 2) determining load peak-valley periods: based on analysis of a large amount of historical provincial load data, statistically analyzing the time when load valley value periods I, load peak value periods I, load valley value periods II and load peak value periods II appear in a day, obtaining the period with the largest occurrence proportion of the four types of periods as the load peak-valley period, and considering correcting the wind power prediction error of the period; 3) determining the correction interval: according to the statistical results of the duration of the load peak-valley period in the historical load data, obtaining the duration of the load peak-valley period, dividing the length of time that needs to be corrected in the to-be-corrected load peak-valley period, which is determined based on the length of a single load duration, the length of the correction interval is the duration of a single load period, and the correction radius is half of the duration; 4) determining the to-be-corrected load peak-valley period: according to the seasonal distribution characteristics of a large amount of wind power prediction errors, obtaining the months in which the load peak-valley period in the actual to-be-predicted period needs to be corrected, and the to-be-corrected period is the to-be-corrected load peak-valley period; 5) determining the load peak-valley period correction coefficient: in order to correct the harmful error of the load peak-valley period that needs to be corrected in the specified to-be-predicted period, by analyzing and processing the wind power prediction value and the actual value of the same period of the previous year near the to-be-corrected period, the error correction coefficient of the load period that needs to be corrected in the current to-be-predicted period is obtained; 6) wind power prediction error correction of the load peak-valley period: based on the wind power prediction result of the corresponding period and the error correction coefficient, the corresponding wind power correction result can be obtained by multiplying the correction coefficient and the wind power preliminary prediction result.

[0056] The specific embodiments are as follows:

[0057] Step 1: In order to maximize the accuracy of wind power prediction, the present application mainly selects multiple prediction models to perform ultra-short-term prediction on the test set wind power. Since error correction is applicable to any model, in order to maximize the accuracy of power prediction, the power prediction results obtained by selecting the method with the smallest prediction error are further corrected to highlight the effectiveness of the correction method. The present application selects wind power data and load data from 2016 to 2019 in Northeast China for analysis, where the resolution of power data and load data is 15 min, and the installed capacity information is shown below. The present application mainly selects wind power in 2016 as the training set, and the data from 2017 to 2019 as the test set for error analysis and correction. Table 1 shows the error evaluation indicators of the multiple model wind power ultra-short-term prediction results in the test set, including BP neural network, extreme learning machine (ELM), bidirectional long short-term memory network (BILSTM), and weighted first-order local prediction method. It can be seen that among various prediction models, the prediction error of traditional shallow neural network model BP and ELM is relatively larger, while the prediction error of BILSTM and weighted first-order local prediction model is lower, especially the weighted first-order local prediction method, which has the lowest error. Therefore, the present application uses the prediction results with lower error for further error correction.

[0058] Table 1 Evaluation indicators of each prediction method from 2017-2019 Tab.1Evaluation indicators of each prediction method from 2017-2019

[0059]

[0060] Step 2: In order to effectively correct the load peak and valley period, first, based on historical load data, a large amount of analysis is performed to obtain the corresponding load valley period I, peak period I, valley period II and peak period II, which appear frequently. The schematic diagram of load peak and valley period is shown in Figure 2 , where Figure 3 and Figure 4 show the statistical results of the time of the four load periods. Based on the statistical results of a large amount of data, it is shown that the load valley time I mainly appears at 3:00 within a day, the load valley time II mainly appears at 12:30 within a day, the load peak time I mainly appears at 10:45, and the load peak time II mainly appears at 16:45. According to the time of frequent occurrence of load peak and valley time, it is used as the center of the correction interval, and according to the duration of load peak and valley period of historical load data, the correction interval is obtained.

[0061] Step 3: According to the specific occurrence time of the peak-valley period, the historical load peak-valley period duration is analyzed, in order to ensure that the harmful error of the wind power prediction of the load peak-valley period can be fully corrected, the wind power prediction error correction results under different correction radii are considered, in the present application, the correction radius is 3 (15 min) -1 , the correction interval of which is as shown in Figure 5 , including the load estimation correction interval I, the load estimation correction interval II, the load peak value correction interval I and the load peak value correction interval II. The wind power of the to-be-corrected period of the to-be-predicted period is corrected in different correction intervals.

[0062] Step 4: According to the analysis of the seasonal distribution characteristics of a large number of wind power prediction errors, the monthly distribution results of the wind power prediction error difference are obtained as shown in Figure 6a , Figure 6b , Figure 6c and Figure 6d It can be seen that whether the positive error of the load valley value period or the negative error of the load peak value period, there is a certain monthly distribution characteristic, and the negative error distribution regularity of the load peak value period is stronger, as shown in Figure 6a , Figure 6b In the load valley value period, from January to April and in November and December, the number of days of positive error is greater than that of negative error, and there is a large amount of harmful error in the load valley value period of these months; and for the load peak value period, as shown in Figure 6c and Figure 6d , the number of days of negative error is greater than that of positive error from May to October, and there is a large amount of harmful error in the peak value period of these months. In order to more finely describe the error difference distribution regularity of the load peak and valley value period, each month is divided into two parts for analysis, as shown in Figure 7a In the first eight months and the last four months of the load valley value period I, the error is mostly positive, which is consistent with Figure 6a . As for the load valley value period II, the number of days of positive error is high in the third to eighth months and the last four months, as shown in Figure 7b . As for Figure 7c and Figure 7d , in the load peak value period I, the number of days of negative error is mainly concentrated in the first to fifth and tenth to twentieth months, and the number of days of negative error of the load peak value period II is mainly concentrated in the tenth to twentieth months, which is consistent with Figure 6c and Figure 6dConsistent. Based on the above characteristics of the interannual distribution of errors in the load peak and valley periods, the months in which the errors in the load peak and valley periods mainly exist can be obtained. According to the distribution law of errors in the load peak and valley periods, the idea of separately correcting the load peak and valley periods is adopted, and the correction interval length of the load valley period I in the first to eighth half months and the twenty-first to twenty-fourth half months, the load valley period II in the third to eighth half months and the twenty-first to twenty-fourth half months, the load peak period I in the first to fifth and tenth to twentieth half months, and the load peak period II in the tenth to twentieth half months is determined, and the correction coefficient of each correction interval corresponding period is calculated to correct the wind power of each to be corrected load peak and valley period.

[0063] Step 5: According to the analysis results of wind power prediction error from 2016 to 2019, the error correction coefficients of the load peak and valley periods in 2017, 2018 and 2019 are calculated respectively, and the error correction coefficients of each load peak and valley period are shown in Table 2. The positive error of the load valley period and the negative error of the load peak period are corrected, so the correction coefficient of the load valley period is greater than 1, and the error correction coefficient of the load peak period is greater than 0 and less than 1, which can significantly correct the harmful error of the load peak and valley period.

[0064] Table 2 Correction coefficient for each peak and valley period from 2017 to 2019 Tab.2Correction coefficient for each peak and valley period from 2017 to 2019

[0065]

[0066] Step 6: The final corrected power value is obtained by multiplying the preliminary power prediction result corresponding to the load peak and valley period by the correction coefficient corresponding to the period. Table 3 shows the error evaluation index of the power prediction value before and after correction of the load peak and valley period from 2017 to 2019. It can be seen that in each peak and valley period, the RMSE and MAE after correction are decreased, especially in the peak period, but there is a phenomenon that the error is slightly larger after correction in individual peak periods. The main reason is that the prediction value and the actual value are high in these periods, and the difference between the prediction value and the actual value is not large. After the positive error correction, the prediction value is small, especially when the correction coefficient is small. Even so, the number of harmful errors in these periods is significantly reduced, which is significantly beneficial to the load peak period, and reduces the number of days of negative error in the load peak period.

[0067] Table 3 Error evaluation index before and after correction of each peak and valley period from 2017 to 2019

[0068] Tab.3Error evaluation indicators before and after correction for each peak and valley period from

[0069] 2017to 2019

[0070]

[0071] Figure 8a 、 Figure 8b 、 Figure 8c and Figure 8d The main reflection of the load peak and valley value period in 2018, the error after correction of the predicted value and the actual value, the valley value period I in 2018, as shown in Figure 8a , the corrected predicted value is significantly higher, and can better fit the actual value, the correction effect is remarkable, while Figure 8b the load valley value period II, Figure 8c the load peak value period I and Figure 8d the load peak value period II also shows that the corrected value of the valley value period is larger, and the corrected value of the peak value period is smaller, which ensures the load demand of the load peak and valley value period and reduces the prediction error.

[0072] Since the above embodiment is for the correction radius r = 3 (15min) -1 The length of the correction interval is determined as 6 (15min) -1 The wind power ultra-short-term prediction result is corrected, and the analysis of the historical load data can obtain that the load peak and valley value duration will not be very long. On this basis, the influence of different correction radii on the correction result is discussed when the correction interval length is controlled in the range of 0-3h, and the correction radius can be taken as 0, 1, 2, 3, 4, 5, 6 (15min) -1 , and the corresponding correction interval length can be taken as 0, 2, 4, 6, 8, 10, 12 (15min) -1Table 4 shows the trend of the correction results and the number of days with harmful errors during the peak and valley periods of total load under different correction radii. It can be seen that from 2017 to 2018, the total load RMSE and MAE after error correction showed significant improvement under different correction radii. While the error decreased, the number of days with harmful errors during each peak and valley period also decreased significantly, especially during the peak load period. In 2019, the corrected RMSE was slightly larger, but the number of days with reduced harmful errors was the highest among the three years, with a maximum reduction of 126 days. This is of great significance for improving the power supply reliability during peak load periods. The correction results also showed significant differences for different correction radii, and these differences varied from year to year. In 2017, under different correction radii, the amount of correction error tended to increase with the increase of the correction radius, i.e., with the increase of the correction interval. At a correction radius of 6 (15 min), the error increased further. -1 At this point, the corrected root mean square error reaches its maximum. This pattern is also evident in 2018; when the correction radius reaches its maximum, the maximum corrected root mean square error decreases compared to 2017, but significantly increases compared to other correction radii in the same year. This pattern is less pronounced in 2019, but it still applies when the correction radius is greater than 2, such as... Figure 9 As shown, the stacked bar chart represents the number of days harmful errors are reduced, and the line chart represents the amount of error correction. Therefore, during the period of sustained load peaks and valleys, within a reasonable correction range, a larger correction radius will increase the amount of corrected error. However, this does not mean that a larger correction radius will reduce the number of days harmful errors occur after correction. Rather, an appropriate correction radius will increase the amount of corrected error while reducing the number of days harmful errors occur. To maximize the reduction of the number of days harmful errors occur while increasing the amount of corrected error, r can be taken as 3 (15 min) in 2017. -1 In 2018, r can be taken as 0 (15min). -1 In 2019, r can be taken as 1 (15 min). -1 .

[0073] Table 4. Comparison of peak and trough values ​​before and after correction for different correction radii from 2017 to 2019

[0074] Tab.4Comparison of indicators before and after the correction oftotal time period of peak and

[0075] valley under different correction radii from 2017 to 2019

[0076]

[0077]

[0078] Considering that the negative error in the load peak period does not cause the loss of abandoned wind power, and the positive error caused by the predicted value in the load valley period makes the excess power not consumed in time, if this part of power is maintained by abandoning wind to maintain the power balance of the system, it will lead to the loss of abandoned wind in the valley period. Table 5 gives the theoretical calculation value of the wind power abandonment loss before and after correction in the load valley period in 2019. This is mainly obtained by subtracting the predicted wind power curve area from the actual wind power curve area before and after correction. It can be seen that after correction in the load valley period, the wind power loss is significantly reduced, and the reduced amount of power loss after correction increases with the increase of the correction radius. The wind power ultra-short-term prediction error correction method proposed in the present application will significantly reduce the wind power loss in the load valley period.

[0079] Table 5 Comparison of wind abandonment power losses during load valley periods before and after

[0080] error correction in 2019

[0081]

[0082] The specific embodiments of the present application are not exhaustive and do not constitute a limitation on the scope of protection of the claims. Based on the inspiration obtained from the embodiments of the present application, other substantially equivalent alternatives can be thought of without creative labor, which are within the protection scope of the present application.

Claims

1.A method for correcting wind power cluster power ultra-short-term prediction error, characterized in that: It includes the following steps: 1) wind power ultra-short-term preliminary prediction To correct the harmful error of wind power in peak and valley period of load, firstly, the wind power needs to be preliminarily predicted, and the mapping of the prediction model is set as The input of wind power prediction is represented as: Wherein The number of input features is represented, and the power prediction result corresponding to the wind power prediction model is represented as: ； wherein, represents the wind power prediction value; 2) determine the load peak valley period Since the load sequence usually has two valley periods and two peak periods in a day, for different peak periods, the method of separately correcting the peak period and the valley period is used to reduce harmful errors, where based on the analysis of one year of load data, the time of the load peak and the valley period of each day in a year is concentrated as the load peak and valley time, including load valley period I, load peak period I, load valley period II and load peak period II; 3) determine the correction interval According to the duration of load peak and valley period, the correction interval of harmful error of wind power prediction is established in the load peak and valley period, the negative error of peak period and the positive error of valley period are corrected. Since the load peak or load valley is not a time value but a duration period, in the duration period, if the load is in valley, it is considered that the load demand is small and the generator output is low in the duration period, on the contrary, if the load is in peak, it is considered that the load demand is large and the generator output is high in the duration period. The definition of correction interval is that the load peak or valley time is the center of correction interval, and the correction interval length is 2 r As the correction interval radius, the correction interval length at a certain load peak or valley is 2 r , r According to the actual situation, the adjustment is made, wherein r The unit of is (15min) -1 , r Reflects the step of the period to be corrected, but in actual correction, the duration of load peak and valley should be considered to ensure effective correction in the load peak and valley period. There will be four correction intervals in a day, i.e. load valley correction interval I, load peak correction interval I, load valley correction interval II and load peak correction interval II. 4) determine the to-be-corrected load peak valley period After determining the correction interval radius, the correction interval of each peak and valley period is obtained, and the average value of the difference between the historical wind power prediction value and the actual value in the correction interval is used as the basis for judging the positive and negative error, and the calculation formula is: ； wherein, denotes the length of the correction interval, here is taken ; denotes the load peak or valley period wind power actual value at the th time point; denotes the predicted value; denotes the predicted error mean of the peak-valley value period correction interval, the error of two peak values and two valley values within a day is calculated, reflecting the positive and negative situation of the load peak or valley period wind power prediction error mean within a day, used to judge the error type mainly existing in the load peak-valley period; at this time, define: in a certain time period, the difference between the number of days with positive value and the number of days with negative value is used to measure the degree of harmful error more or less in the valley value period, the greater the difference, the more harmful errors in this time scale, the more periods need to be corrected; similarly, the difference between the number of days with negative value and the number of days with positive value is greater, the more harmful errors in the peak period, the more periods need to be corrected, so as to analyze the historical wind power prediction error to determine the month corresponding to the load peak-valley period which needs to be corrected; 5) determine the load peak valley period correction coefficient In order to minimize the days of harmful error of ultra-short-term wind power prediction, the actual value and the predicted value of wind power in the load peak-valley period with high harmful error frequency in the previous year are selected, and the average value of the ratio in the whole selected period is taken as the correction coefficient The specific expression is: ； wherein, represents a correction coefficient; represents the number of days of the period of the peak or valley of the load to be corrected; represents a correction interval radius; 6) wind power prediction error correction of load peak valley period Based on the determination of the correction time period and the correction interval of the load peak valley period, when the wind power in the to-be-corrected period needs to be corrected, based on the correction coefficient corresponding to the load peak valley period calculated in the previous year and the wind power prediction value that needs to be corrected at present, the corrected wind power prediction value is obtained by multiplication, and the correction result is: ； wherein, represents the wind power correction result of the first load period, represents the correction coefficient of the first load period; when , respectively represents the load valley period I, the load peak period I, the load valley period II, and the load peak period II. 7) simulation calculation Simulation input: Analyze the measured data and load data of the wind farm to determine the total installed capacity of the farm; input data: historical power of the wind farm and numerical weather prediction; data sampling interval is 15 minutes; according to steps 1) ~ 5), the wind power ultra-short-term preliminary prediction result of the prediction period is obtained; 8) error analysis In order to measure the effectiveness of the power prediction result and the error correction result, the mean absolute error MAE and the root mean square error RMSE are used to evaluate the predicted and corrected wind power values, and the calculation formula is: ； wherein, represents the number of point predictions of the prediction sequence; represents the rated installed capacity.

Images