A rolling prediction method for the input wind power of downwind turbines in a wind farm

Abstract

The present invention discloses a rolling prediction method for the input wind power of downwind turbines in a wind farm. The method collects the active power of the target turbine and each reference turbine to form a wind power sequence. In each prediction period, the wind power change trend sequences of the target turbine and each reference turbine are calculated respectively, and the correlation coefficient between the two is calculated. The correlation coefficient is used as the basis for determining the time lag between each reference turbine and the target turbine. According to the time lag and the corresponding relationship of power change, the wind power prediction results of the target turbine corresponding to each reference turbine are obtained respectively. Then, according to the maximum correlation coefficient between each reference turbine and the target turbine, the weights of the prediction results corresponding to each reference turbine are determined, and through weighted summation calculation, the prediction result of the target turbine in this round of prediction is obtained. The rolling prediction method for the input wind power of wind turbines proposed by the present invention has high accuracy and can provide a basis for rolling evaluation of the frequency modulation capabilities of each turbine in the wind farm at different times.

Description

Technical Field

[0001] The present invention relates to the field of power system control, and particularly to a rolling prediction method for the input wind power of downwind units in a wind farm. Background Art

[0002] With the deepening of the degree of power electronics in the power system, the frequency regulation ability of the power system is weakened. Some accidents that cause a large drop in the grid frequency have drawn people's attention to the problem of grid frequency stability. To ensure the frequency stability of the power system under severe accidents, it is necessary to actively regulate the system power balance through emergency control, quickly suppress the rapid drop of the system frequency, and win time for the subsequent conventional frequency modulation measures to play a role. In addition to the conventional frequency regulation measures, since wind power generation has become an important power source in the grid, many countries in the world have made regulations on how wind power generation provides frequency support for the grid.

[0003] The conventional control of wind turbines cannot respond to the change of the system frequency. Wind turbines must rely on additional control measures to participate in the grid frequency regulation. The basic control methods of wind power frequency modulation include virtual inertia control, droop control, pitch control, and over-speed control. Among them, virtual inertia control and droop control belong to rotor kinetic energy control, which releases the rotor kinetic energy to provide active power support; pitch control and over-speed control belong to load shedding control, which provides standby power for wind turbines. When the wind turbine generator participates in primary frequency modulation by operating at a reduced load or leaving standby power through an energy storage device, otherwise the wind turbine can only achieve a rapid frequency response by releasing the rotor kinetic energy.

[0004] For wind power frequency modulation, the time range of rapid frequency response is generally within 10 s, while the time of primary frequency modulation is generally within 30 s. Even within the time range of rapid frequency response, the wind speed is not fixed, but most of the existing studies on the frequency regulation of wind turbines do not consider the change of wind speed. The change of wind speed has a great impact on the frequency regulation ability of wind turbines. For example, for load shedding control, if the wind speed drops, the standby power may be lost. To accurately analyze the frequency modulation support ability of wind power generation, it is necessary to predict the change of wind speed after the unit starts the frequency response. However, the anemometer of the wind turbine is installed on the nacelle behind the wind wheel, and the measured wind speed is not the true input wind speed of the wind turbine. Therefore, the present invention ignores the power loss of the wind turbine itself, and uses the accurately measured electric power (active power) value as the input wind power of the unit to characterize the change of wind speed. The rolling prediction method for the input wind power of downwind units in the wind farm proposed by the present invention can provide a relatively accurate basis for rolling evaluation of the frequency regulation ability of downwind units at different times. The wind speed of the upwind units in the wind farm can be obtained by the existing numerical prediction methods. Although the prediction accuracy is not high, it can still be used and is not involved in the present invention. Summary of the Invention

[0005] The object of the present invention is to provide a rolling prediction method for the input wind power of downwind turbines in a wind farm, which can provide a relatively accurate basis for rolling assessment of the frequency modulation capabilities of downwind turbines at different times, and plays an important role in promoting the technology of wind power participating in power grid frequency support.

[0006] To achieve the above functions, the present invention designs a rolling prediction method for the input wind power of downwind turbines in a wind farm. For the target turbine, at least one turbine is selected as the reference turbine of the target turbine within the preset range upwind of it. For the target turbine and each reference turbine, the following steps S1 - S6 are executed to realize the rolling prediction of the input wind power of the target turbine:

[0007] Step S1: The active power of the target turbine and each reference turbine is collected in real time and continuously. The active power of the target turbine and each reference turbine is used as the wind power respectively. The wind power sequences of the target turbine and each reference turbine corresponding to the sampling times at preset time intervals are formed. The preset time point is defined as the prediction starting point.

[0008] Step S2: According to the wind power sequences of the target turbine and each reference turbine, with the prediction starting point as the starting point, the total prediction period is extended for a preset duration in the future time direction. When the prediction starting point arrives, the wind power data of the target turbine and each reference turbine at each preset time point within 500 s in the historical time direction starting from the current moment are calculated respectively. The wind power data is pre - processed into wind power trend data, and based on the wind power trend data of the target turbine and each reference turbine, the wind power change trend sequences of the target turbine and each reference turbine are constructed respectively.

[0009] Step S3: Based on the Pearson correlation coefficient algorithm, the correlation coefficient C between the wind power change trend sequences of the target turbine and each reference turbine is calculated, and based on the correlation coefficient C, the time lag between the target turbine and each reference turbine is calculated.

[0010] Step S4: Calculate the corresponding relationship between the wind power changes of the target turbine and each reference turbine within a 5 - s time period starting from the prediction starting point in the future time direction. According to the time lag between the target turbine and each reference turbine and the corresponding relationship between the wind power changes of the target turbine and each reference turbine, the wind power prediction results of the target turbine corresponding to each reference turbine are calculated respectively.

[0011] Step S5: According to the correlation coefficient C between the wind power change trend sequences of the target turbine and each reference turbine, the weights of the wind power prediction results corresponding to each reference turbine are calculated, and the weighted combined prediction result of the target turbine wind power is calculated.

[0012] Step S6: When the next prediction starting point arrives, repeat Steps S2 - S5 to achieve the rolling prediction of the input wind power of the target unit.

[0013] As a preferred technical solution of the present invention: The preset time interval in Step S1 is 5 s.

[0014] As a preferred technical solution of the present invention: In Step S2, the total prediction cycle duration is 25 s, including a prediction start cycle and a single prediction distance. Taking the prediction starting point as the starting point, the period extending 5 s in the future time direction is the prediction start cycle, and the time period from the end time point of the prediction start cycle to the end time point of the total prediction cycle is the single prediction distance, and the single prediction distance duration is 20 s.

[0015] As a preferred technical solution of the present invention: The specific method of converting the wind power data into wind power trend data through preprocessing in Step S2 is as follows:

[0016] Step S21: Divide the wind power trend into the following five situations: single increase, remain unchanged, single decrease, wind strengthening, and wind weakening; respectively mark the five wind power trends with the corresponding wind power trend data -2, -1, 0, 1, 2;

[0017] Step S22: Calculate the wind power trend based on the following formula through the wind power data:

[0018]

[0019] In the formula, ΔP(t - 1) is the wind power increment of the unit from time t - 1 to time t, ΔP(t) is the wind power increment of the unit from time t to time t + 1, and sign represents matching the corresponding wind power trend data for this wind power trend.

[0020] As a preferred technical solution of the present invention: The specific steps of Step S3 are as follows:

[0021] Step S31: Based on the Pearson correlation coefficient algorithm, calculate the correlation coefficient C between the wind power change trend sequences of the target unit and each reference unit as follows:

[0022]

[0023] In the formula, N is the length of the wind power trend data of the target unit and each reference unit, X is the wind power trend data of the target unit, and Y is the wind power trend data of the reference unit;

[0024] Step S32: Based on the correlation coefficient C, calculate the time lag Δt between the target unit and each reference unit as follows:

[0025]

[0026] I = t - N, t - N + 1, …, t - 1

[0027] Step S33: Select different time lags Δt within a preset range, and take the maximum value among the correlation coefficients C corresponding to different time lags Δt as the maximum correlation coefficient C max , and take the time lag Δt corresponding to the maximum correlation coefficient C max as the time lag between the wind power of the target unit and the reference unit.

[0028] As a preferred technical solution of the present invention: The specific steps of Step S4 are as follows:

[0029] Step S41: Define the wind power change from time t to its next moment as the wind power change at this moment, and calculate the ratio k of the wind power changes of the target unit and the reference unit within the prediction start period of this round of prediction as follows, and use the ratio k of the wind power changes to represent the corresponding relationship between the wind power changes of the target unit and each reference unit:

[0030]

[0031] Step S42: According to the time lag between the target unit and each reference unit, and the corresponding relationship between the wind power changes of the target unit and each reference unit, calculate the wind power prediction results of the target unit corresponding to each reference unit as follows:

[0032] P tar (t + l) = P tar (t + l - 1) + k * ΔP ref (t - Δt + l)

[0033] l = 1, 2, …, 5

[0034] As a preferred technical solution of the present invention: The weight α i of the wind power prediction result corresponding to the reference unit i in Step S5 is calculated as follows:

[0035]

[0036] In the formula, C max_i is the maximum correlation coefficient between the wind power change trend sequences of the reference unit i and the target unit, represents the sum of the maximum correlation coefficients between the wind power change trend sequences of all reference units and the target unit;

[0037] The weighted combined prediction result P tar (t) is calculated as follows:

[0038]

[0039] Wherein, P tar_i (t) represents the predicted result of the wind power of the target unit corresponding to the reference unit i.

[0040] As a preferred technical solution of the present invention: In step S6, starting from the end time point of the current prediction period, a time point 5 s shifted in the historical time direction is used as the next prediction starting point.

[0041] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0042] Within dozens of seconds when wind power generation participates in rapid frequency response or primary frequency modulation, the wind speed is continuously changing. To evaluate the frequency modulation support ability of wind power generation, it is necessary to predict the change of the wind speed after the wind turbine starts the frequency response. For this reason, a rolling prediction method for the input wind power of the downwind units in a wind farm is proposed. The present invention utilizes the spatial correlation of the power between wind turbines, and performs weighted summation on the predicted results of the wind power of the downwind units predicted according to the wind power curves of several upwind wind turbines, so as to obtain the future wind power change curve of the downwind units. The rolling prediction method of the wind turbine power proposed by the present invention generally has high accuracy and can provide a basis for rolling evaluation of the frequency modulation ability of the wind farm at different times. Description of the Drawings

[0043] Figure 1 is a flowchart of a rolling prediction method for the input wind power of wind turbines in a wind farm provided according to an embodiment of the present invention;

[0044] Figure 2 is a topology diagram of some wind turbines in a wind farm provided according to an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of rolling prediction provided according to an embodiment of the present invention;

[0046] Figure 4(a) is a power curve of two related units within 500 s provided according to an embodiment of the present invention;

[0047] Figure 4(b) is a trend curve of two related units within 500 s provided according to an embodiment of the present invention;

[0048] Figure 5 is the correlation coefficient of the trends of two units changing with time lag provided according to an embodiment of the present invention;

[0049] Figure 6 is the rolling prediction error of the unit power under different prediction distances provided according to an embodiment of the present invention;

[0050] Figure 7 is the rolling prediction result of Unit 37 provided according to an embodiment of the present invention. Detailed Embodiments

[0051] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0052] Referring to Figure 1 , a rolling prediction method for the input wind power of a downwind unit in a wind farm provided by an embodiment of the present invention selects at least one unit as a reference unit for the target unit within a preset range upwind of the target unit. For the target unit and each reference unit, the following steps S1 - S6 are executed to realize the rolling prediction of the input wind power of the target unit:

[0053] Step S1: Real - time and continuously collect the active power of the target unit and each reference unit. Respectively use the active power of the target unit and each reference unit as the wind power, and respectively use the wind power of the target unit and each reference unit corresponding to the sampling time at a preset time interval to form a wind power sequence, where the preset time interval is 5s. Define the preset time point as the prediction starting point.

[0054] In one embodiment, three units, namely #72, #73, and #74, in a wind farm along the coast of China are used as the upwind reference units for the target unit #37. The partial fan topology diagram of this wind farm is referred to Figure 2 , and the sampling interval of the power data of each unit is 5s.

[0055] Step S2: According to the wind power sequences of the target unit and each reference unit, with the prediction starting point as the starting point, continue in the future time direction for a preset duration as the total prediction period. When the prediction starting point arrives, calculate the wind power data of the target unit and each reference unit at each preset time point within 500s in the historical time direction with the current moment as the starting point respectively. Pre - process the wind power data into wind power trend data, and respectively construct the wind power change trend sequences of the target unit and each reference unit based on the wind power trend data of the target unit and each reference unit.

[0056] In step S2, the total prediction period duration is 25s, including the prediction starting period and the single - time prediction distance. Continuing 5s in the future time direction with the prediction starting point as the starting point is the prediction starting period, and the time period from the end time point of the prediction starting period to the end time point of the total prediction period is the single - time prediction distance, and the single - time prediction distance duration is 20s. Refer to Figure 3 , the prediction result is output within 5s after the prediction starting point, that is, the power at five time points within 25s after the prediction starting point is calculated within 5s, and the four time points within the latter 20s are taken as the rolling prediction result (one prediction period).

[0057] In step S2, the specific method for converting wind power data into wind power trend data through preprocessing is as follows:

[0058] Step S21: Divide the wind power trend into the following five cases: single increase, remain unchanged, single decrease, wind strengthening, and wind weakening; mark the five wind power trends with the corresponding wind power trend data -2, -1, 0, 1, and 2 respectively;

[0059] The markings of the five wind power trends are shown in Table 1:

[0060] Table 1

[0061]

[0062]

[0063] Step S22: The specific method for judging the wind power trend at a certain time point is to judge the trend of this point by comparing the wind power increments before and after this point and combining the positive and negative of the wind power increments on both sides. Calculate the wind power trend based on the wind power data according to the following formula:

[0064]

[0065] In the formula, ΔP(t - 1) is the wind power increment of the unit from time t - 1 to time t, ΔP(t) is the wind power increment of the unit from time t to time t + 1, and sign represents matching the corresponding wind power trend data for this wind power trend.

[0066] Figure 4 shows the power curves and trend curves of two relevant units measured within 500 s. In Figure 4(a), the black curve is the power of the upwind unit WTG ref and the other gray curve is the power of the downwind unit WTG tar . According to the above formula, the power trend change curves of WTG ref and WTG tar are shown in Figure 4(b). Comparing Figure 4(a) and Figure 4(b), it can be seen that compared with the power curve, the trend curve can more intuitively reflect the time lag between the powers of the two wind turbines.

[0067] Step S3: Based on the Pearson correlation coefficient algorithm, calculate the correlation coefficient C between the wind power change trend sequences of the target unit and each reference unit respectively, and calculate the time lag between the target unit and each reference unit based on the correlation coefficient C.

[0068] The specific steps of step S3 are as follows:

[0069] Step S31: Based on the Pearson correlation coefficient algorithm, calculate the correlation coefficient C between the wind power change trend sequences of the target unit and each reference unit respectively as follows:

[0070]

[0071] Wherein, N is the length of the wind power trend data of the target unit and each reference unit, X is the wind power trend data of the target unit, and Y is the wind power trend data of the reference unit;

[0072] Step S32: Based on the correlation coefficient C, calculate the time delay Δt between the target unit and each reference unit as follows:

[0073]

[0074] I = t - N, t - N + 1, …, t - 1

[0075] In this embodiment, the sampling time interval for collecting power is 5 s. To statistically analyze the correlation coefficient of the power trend between the reference unit and the target unit within the current 500 s, the length N of the trend data of the two units is taken as 100.

[0076] Step S33: Select different time delays Δt within a preset range, and use the maximum value among the correlation coefficients C corresponding to different time delays Δt as the maximum correlation coefficient C max , and use the time delay Δt corresponding to the maximum correlation coefficient C max as the time delay between the wind power of the target unit and the reference unit.

[0077] The curve of the change of the correlation coefficient C between the wind power change trend sequences of the target unit and each reference unit obtained when t is 500 s with respect to the time delay Δt is as Figure 5 shown. When Δt = 50 s, the correlation coefficient between the two units is the largest. Therefore, it can be considered that the current power time delay between the two units is 50 s.

[0078] Step S4: Calculate the corresponding relationship between the wind power changes of the target unit and each reference unit in a 5-s time period extending from the prediction starting point to the future time direction. According to the time delay between the target unit and each reference unit, and the corresponding relationship between the wind power changes of the target unit and each reference unit, calculate the wind power prediction results of the target unit corresponding to each reference unit respectively.

[0079] Considering that the power change rates of the target unit and the reference unit are inconsistent, the differential method is used to calculate the ratio of the power change amounts of the target unit and the reference unit to find the corresponding relationship between the power changes of the target unit and the reference unit. First, the power change amount from a certain moment to its next moment is defined as the power change amount at that moment; then, the ratio k of the power change amounts of the target unit and the reference unit within 5 s before the start of this round of prediction is calculated. During this round of prediction, the power change amount of the target unit at the current moment corresponds to the power change amount of the reference unit before the time delay Δt. During prediction, k is used to correct the power change amount of the reference unit corresponding to the prediction point of the target unit, and then added to the power value of the target unit at the previous moment, which is used as the power prediction value at this moment. The wind power of the target unit predicted according to a certain reference unit can be expressed by the following formula, that is, there are differences in the power change amounts between the wind power of the target unit and the wind power of the reference unit, and there is a certain time delay in time (where Δt > 0).

[0080] The specific steps of step S4 are as follows:

[0081] Step S41: Define the wind power change amount from time t to its next moment as the wind power change amount at that moment, and calculate the ratio k of the wind power change amounts of the target unit and the reference unit within the prediction start period of this round of prediction as follows, and use the ratio k of the wind power change amounts to represent the corresponding relationship between the wind power changes of the target unit and each reference unit:

[0082]

[0083] In the formula, ΔP tar represents the wind power change amount of the target unit, and ΔP ref represents the wind power change amount of the reference unit;

[0084] Step S42: According to the time delay between the target unit and each reference unit, and the corresponding relationship between the wind power changes of the target unit and each reference unit, calculate the wind power prediction results of the target unit corresponding to each reference unit as follows:

[0085] P tar (t + l) = P tar (t + l - 1) + k * ΔP ref (t - Δt + l)

[0086] l = 1, 2, …, 5

[0087] In the formula, P tar represents the predicted wind power of the target unit;

[0088] Step S5: Calculate the weights of the wind power prediction results corresponding to each reference unit according to the correlation coefficient C between the target unit and the wind power change trend sequences of each reference unit, and calculate the weighted combined prediction result of the wind power of the target unit.

[0089] The weight α of the wind power prediction result corresponding to reference unit i in Step S5 i is calculated as follows:

[0090]

[0091] In the formula, C max_i is the maximum correlation coefficient between the wind power change trend sequences of reference unit i and the target unit, indicating the sum of the maximum correlation coefficients between the wind power change trend sequences of all reference units and the target unit;

[0092] The weighted combined prediction result P tar (t) of the wind power of the target unit is calculated as follows:

[0093]

[0094] In the formula, P tar_i (t) represents the wind power prediction result of the target unit corresponding to reference unit i.

[0095] Step S6: When the next prediction starting point arrives, repeat Steps S2 - S5 to achieve the rolling prediction of the input wind power of the target unit.

[0096] Refer to Figure 3 , in Step S6, the time point 5 s before the end time point of the current prediction cycle is used as the next prediction starting point and is shifted in the historical time direction.

[0097] In an embodiment, the rolling prediction is performed on Unit 37 using the weighted combined prediction method. The power prediction and error analysis of the target unit are carried out using the data of this wind farm from 20:41:40 to 22:05:00 on January 17, 2014. The mean absolute percentage error (MAPE) of the rolling prediction of Target Unit 37 under different prediction distances is as Figure 6 shown. Considering that the rolling prediction needs to balance the prediction accuracy when selecting as long a prediction distance as possible, in this embodiment, 5 data points with a single prediction distance interval of 5 s are selected.

[0098] Figure 7 is the result of the weighted combined prediction of Unit 37 under the conditions that the rolling prediction interval is 20 s, the single prediction distance is 5 data points with an interval of 5 s, and the total length of the rolling prediction is 1000 data points. From Figure 7It can be seen that the prediction results of the target unit are in good agreement with the measured values, and generally the prediction accuracy is relatively high. The existing deficiency is that there are prone to deviations in the prediction results near the actual power turning point, and the overall prediction error will be relatively large when the actual power fluctuates frequently.

[0099] In this embodiment, the root mean square error (RMSE) and the mean absolute percentage error (MAPE) are used to analyze the error of the power prediction results of the wind turbine. The rolling prediction errors of the power of Unit 37 are shown in Table 2. In this embodiment, it is considered that MAPE>5% indicates a large prediction error. Then the prediction results show that the number of prediction points with large errors for Unit 37 is small, accounting for only 2.20%. This proves that the wind turbine power prediction method proposed in this embodiment has high accuracy.

[0100] Table 2

[0101]

[0102] The above has described in detail the embodiments of the present invention in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the gist of the present invention within the scope of knowledge possessed by those of ordinary skill in the art.

Claims

1. A rolling prediction method for the input wind power of downwind turbines in a wind farm, characterized in that For the target unit, select at least one unit within the preset range upwind of it as the reference unit of the target unit. For the target unit and each reference unit, perform the following steps S1 - S6 to achieve the rolling prediction of the input wind power of the target unit: Step S1: Real - time and continuously collect the active power of the target unit and each reference unit. Respectively take the active power of the target unit and each reference unit as the wind power, and respectively form wind power sequences with the wind power of the target unit and each reference unit corresponding to the sampling time at preset time intervals; Define the preset time point as the prediction starting point; Step S2: According to the wind power sequences of the target unit and each reference unit, extend a preset duration in the future time direction with the prediction starting point as the starting point for the total prediction period. When the prediction starting point arrives, respectively calculate the wind power data of the target unit and each reference unit at each preset time point within 500 s in the historical time direction with the current moment as the starting point. Convert the wind power data into wind power trend data through pre - processing, and respectively construct the wind power change trend sequences of the target unit and each reference unit based on the wind power trend data of the target unit and each reference unit; Step S3: Based on the Pearson correlation coefficient algorithm, calculate the correlation coefficient C between the wind power change trend sequences of the target unit and each reference unit respectively, and based on the correlation coefficient C, calculate the time lag Δt between the target unit and each reference unit as follows: N is the length of the wind power trend data of the target unit and each reference unit. Different time lags Δt are selected within a preset range, and the maximum value among the correlation coefficients C corresponding to different time lags Δt is taken as the maximum correlation coefficient C max , and the maximum correlation coefficient C max The corresponding time lag Δt is used as the time lag between the wind power of the target unit and the reference unit; Step S4: Calculate the corresponding relationship between the wind power changes of the target unit and each reference unit within a 5 - s time period extending in the future time direction with the prediction starting point as the starting point. According to the time lag between the target unit and each reference unit and the corresponding relationship between the wind power changes of the target unit and each reference unit, respectively calculate the wind power prediction results of the target unit corresponding to each reference unit; Define the wind power change amount from moment t to its next moment as the wind power change amount at this moment, and calculate the ratio k of the wind power change amounts of the target unit and the reference unit within the prediction starting period of this round of prediction as follows, and use the ratio k of the wind power change amounts to represent the corresponding relationship between the wind power changes of the target unit and each reference unit: According to the time lag between the target unit and each reference unit and the corresponding relationship between the wind power changes of the target unit and each reference unit, respectively calculate the wind power prediction results of the target unit corresponding to each reference unit as follows: P tar (t + l)=P tar (t + l - 1)+k * ΔP ref (t - Δt + l) l＝1,2,…,5； Step S5: According to the correlation coefficient C between the wind power change trend sequences of the target unit and each reference unit, calculate the weights of the wind power prediction results corresponding to each reference unit respectively, and calculate the weighted combined prediction result of the target unit wind power; The weight α of the wind power prediction result corresponding to the reference unit i i is calculated as follows: Where C max_i is the maximum correlation coefficient between the wind power change trend sequences of the reference unit i and the target unit, represents the sum of the maximum correlation coefficients between the wind power change trend sequences of all reference units and the target unit; The weighted combined prediction result P of the target unit's wind power tar (t) is calculated as follows: Where, P tar_i (t) represents the predicted result of the wind power of the target unit corresponding to the reference unit i; Step S6: When the next prediction starting point arrives, repeat steps S2 - S5 to achieve the rolling prediction of the input wind power of the target unit.

2. The rolling prediction method for the input wind power of the downwind turbines in a wind farm according to claim 1, characterized in that The preset time interval described in step S1 is 5 s.

3. A rolling prediction method for the input wind power of downwind turbines in a wind farm according to claim 1, characterized in that, In step S2, the total prediction period duration is 25 s, including the prediction starting period and the single - prediction distance. Extend 5 s in the future time direction with the prediction starting point as the starting point as the prediction starting period, and the time period from the end time point of the prediction starting period to the end time point of the total prediction period is the single - prediction distance, and the single - prediction distance duration is 20 s.

4. A rolling prediction method for the input wind power of downwind turbines in a wind farm according to claim 1, characterized in that The specific method for converting wind power data into wind power trend data through preprocessing in step S2 is as follows: Step S21: Divide the wind power trend into the following five cases: single increase, remain unchanged, single decrease, wind strengthening, and wind weakening; mark the five wind power trends with the corresponding wind power trend data -2, -1, 0, 1, and 2 respectively; Step S22: Calculate the wind power trend from the wind power data according to the following formula: In the formula, ΔP(t - 1) is the wind power increment of the unit from time t - 1 to time t, ΔP(t) is the wind power increment of the unit from time t to time t + 1, and sign represents matching the corresponding wind power trend data for this wind power trend.

5. A rolling prediction method for the input wind power of downwind turbines in a wind farm according to claim 4, characterized in that In step S3, based on the Pearson correlation coefficient algorithm, calculate the correlation coefficient C between the wind power change trend sequences of the target unit and each reference unit as follows: In the formula, X is the wind power trend data of the target unit, and Y is the wind power trend data of the reference unit.

6. The rolling prediction method for the input wind power of the downwind turbines in a wind farm according to claim 1, characterized in that In step S6, take the end time point of the current prediction period as the starting point and push forward 5 s in the historical time direction as the next prediction starting point.

Images