CN106570790A - Wind farm output power data restoration method considering segmental characteristics of wind speed data - Google Patents

Abstract

The invention discloses a method for repairing wind power output data considering the segmentation characteristics of wind speed data. The length of the time series is divided into two types: continuous abnormal type and local abnormal type; S2, for the local abnormal type data, the interpolation method is used to obtain the repaired wind farm output data; S3, for the continuous abnormal type data, based on the maximum posterior probability, Use the normal data before and after the abnormal data to judge whether the abnormal data contains segment points, and then obtain the wind speed characteristics of each segment from the normal data or pattern recognition method, based on the wind speed characteristics, use the ARMA model to generate the repaired wind speed, and further obtain the repaired wind power Field output data; S4, verify the validity of the repair data, and output the repair report; improve the decision-making accuracy of auxiliary services in the power grid, and reduce unnecessary system backup.

Description

A Wind Farm Output Data Restoration Method Considering the Segmentation Characteristics of Wind Speed Data

technical field

The invention belongs to the field of restoration of output data of new energy power stations, and in particular relates to a restoration method of output data of a wind farm taking into account the segmentation characteristics of wind speed data.

Background technique

Due to the huge amount of wind energy, wide distribution, cleanliness and no pollution, wind power generation has developed rapidly all over the world. However, due to the randomness, volatility and intermittent characteristics of wind energy, the large-scale connection of wind power will have a huge impact on the power system. Therefore, it is necessary to analyze the historical data of wind farm output connected to the system and extract wind farm output. It provides an important decision-making basis for the dispatch of power grid operation.

However, wind farms are generally located in remote locations and have poor communication conditions. The real-time communication between the detection data and the data center is not stable enough, and problems such as data loss, duplication, and errors often occur, which seriously affect the quality of wind farm output data and limit the data. Applications. Therefore, it is very important to repair abnormal data such as missing, repeated, and wrong data.

Existing data restoration technologies mostly extract the characteristics of wind farm data, and then use interpolation or prediction to obtain correction values for abnormal data. This kind of data repair technology can better ensure the consistency of the statistical characteristics of wind farm output data, and eliminate the influence of abnormal data items on the characteristics of wind farm output data.

Although this method can eliminate the interference of abnormal data on the statistical characteristics of wind farm output to a large extent, this kind of restoration technology can only ensure that the restored data maintains the same statistical characteristics as normal data in a long period of time. Even if some restoration techniques take into account the differences in statistical characteristics of wind power in different seasons or between day and night, they still cannot effectively describe the characteristics determined by real-time weather information such as strong wind days and light wind days. The distribution characteristics are very important for the planning and operation of the power system, and the failure to repair these characteristics will seriously affect the application of wind farm output data with a large abnormal proportion.

Therefore, a new wind farm output data repair technology is needed to avoid the above-mentioned defects.

Contents of the invention

Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a wind farm output data restoration method that takes into account the segmental characteristics of wind speed data. Taking the restoration of wind speed as the core, the determination of the segment points of the wind speed sequence and the abnormal data The determination of the corresponding statistical characteristics realizes the restoration technology of wind farm output data that takes into account the differences in short-term wind speed statistical characteristics, so that when repairing data, it not only plays a role in "removing inferiority", but also introduces valuable and effective information, which is helpful for improving The application of wind farm output data in power system planning and operation has very significant significance.

A method for repairing wind farm output data considering the segmentation characteristics of wind speed data, the repair method comprising the following steps:

S1. Screen out repetitive, lacking and unreasonable abnormal data from the obtained data, and divide them into two types: continuous abnormal type and local abnormal type according to the length of the continuous time series corresponding to the abnormal data;

S2, for local anomaly data, use the interpolation method to obtain the repaired wind farm output data;

S3. For continuous abnormal data, based on the maximum posterior probability, use the normal data before and after the abnormal data to judge whether the abnormal data contains segment points, and then based on the wind speed characteristics of each segment obtained by normal data or pattern recognition methods, based on the wind speed characteristics , use the ARMA model to generate the repaired wind speed, and further obtain the repaired wind farm output data;

S4, verifying the validity of the repair data, and outputting a repair report.

Preferably, the S1 is specifically:

Find the data records corresponding to multiple data output at the same time point, these data records are duplicate data; find the time points without output data in the data repair time window, these time points correspond to missing data, and filter the wind farm output data for continuous The data records with the same data at more than 4 time points, the data records with output data greater than the power-on capacity, and the data records with output at night, these data records are unreasonable data; if the corresponding data records at no less than 5 consecutive time points are abnormal data, the data records corresponding to these time points are continuous abnormal data records, and the remaining abnormal data are local abnormal data records.

Preferably, the S2 is specifically:

Record the number of local abnormal data as N, take the data [N/2] items before the local abnormal data, and [N/2] items after the data, and the abscissa values corresponding to these data are marked as 1, 2, ..., [ N/2], N+[N/2], N+[N/2]+1,..., 2N; the N-order polynomial fitting function is obtained by fitting the above N points; the calculation fitting function is in [N/2] +1, [N/2]+2, ..., [N/2]+N values are used as repair output data.

Preferably, said S3 specifically includes the following steps:

Select the normal data before and after the group of abnormal data, and the length of the normal data before and after the abnormal data is 1 day; based on the KS test, use the normal data before and after the abnormal data to judge whether the abnormal data contains segmentation points; Segment point, the position of the sampling segment point, the data before and after the segment point respectively use the normal data of the segment to obtain the ARMA model, and then use the ARMA model to obtain the wind speed of the time series corresponding to the abnormal data; if the group of abnormal data does not contain segment point, the ARMA model is obtained directly by using the normal data before and after the abnormal data, and then the wind speed of the time series corresponding to the abnormal data is obtained by using the ARMA model; based on the output characteristic formula of the fan, the output sequence of each type of fan in the wind farm is obtained from the wind speed; Calculate the output sequence of the wind turbines started in the wind farm, and obtain the repaired wind farm output data.

The technical solution of the present invention has the following beneficial effects:

The invention provides a wind farm output data restoration method that takes into account the segmentation characteristics of wind speed data. This method can consider the non-stationary characteristics of wind speed in a short period of time, and improve the accuracy of wind farm output data restoration by short-term segmentation. . The present invention can more effectively correct abnormal records in wind farm output data and improve the quality of wind farm output data; this is conducive to improving the decision-making level of grid planning and operation, improving the decision-making accuracy of auxiliary services in the grid, and reducing unnecessary System backup, thereby improving the economy of power grid construction and operation.

Description of drawings

The technical solution of the present invention will be described in further detail below with reference to the drawings and embodiments.

Fig. 1 is an overall flowchart of a wind farm output data restoration method that takes into account the segmentation characteristics of wind speed data according to the present invention;

Fig. 2 is a schematic diagram of the commutation voltage area when the AC side is faulty in a wind farm output data restoration method that takes into account the segmentation characteristics of wind speed data according to the present invention.

detailed description

In order to clearly understand the technical solution of the present invention, its detailed structure will be presented in the following description. Obviously, the implementation of the embodiments of the invention is not limited to specific details familiar to those skilled in the art. The preferred embodiments of the present invention are described in detail below, and there may be other implementations besides those described in detail.

The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

Referring to Fig. 1, the overall flow chart of a method for repairing wind farm output data considering the segmentation characteristics of wind speed data in the present invention includes screening and classification of abnormal data, feature extraction of weather information and photovoltaic output parameters, repair of abnormal data groups, etc. The main steps, specifically:

1) Screen out repetitive, lacking and unreasonable abnormal data in the data, and divide them into two types: continuous abnormal type and local abnormal type according to the length of the continuous time series corresponding to the abnormal data;

2) For local anomalous data, the output data of the repaired wind farm is obtained by interpolation method;

3) For continuous abnormal data, based on the maximum posterior probability, the normal data before and after the abnormal data is used to judge whether the abnormal data contains segment points, and then the wind speed characteristics of each segment are obtained by normal data or pattern recognition methods, based on the wind speed characteristics , use the ARMA model to generate the repaired wind speed, and further obtain the repaired wind farm output data;

4) Verify the validity of the repair data, and output a repair report.

The step 1) specifically includes the following steps: searching for data records corresponding to multiple data outputs at the same time point, these data records are repeated data; searching for time points without output data in the data repair time window, corresponding to these time points are For missing data, screen the output data of the wind farm for more than 4 consecutive time points with the same data records, data records with output data greater than the start-up capacity, and data records with output at night, these data records are unreasonable data; if the continuous data is not less than The data records corresponding to the five time points are all abnormal data, then the data records corresponding to these time points are continuous abnormal data records, and the remaining abnormal data are local abnormal data records.

The step 2) specifically includes the following steps: record the number of local abnormal data as N, take the data [N/2] item before the local abnormal data, and then the data [N/2] item, and the corresponding abscissa value of these data Respectively marked as 1, 2, ..., [N/2], N+[N/2], N+[N/2]+1, ..., 2N; the N-order polynomial fitting function is obtained by fitting the above N points; Calculate the value of the fitting function at [N/2]+1, [N/2]+2, ..., [N/2]+N as the repair output data.

The step 3) specifically includes the following steps: select the normal data before and after the group of abnormal data, the length of the normal data before and after the abnormal data is 1 day; based on the KS test, use the normal data before and after the abnormal data to judge whether the abnormal data contains segments point; if the group of abnormal data contains segment points, the position of the segment points is sampled, and the data before and after the segment points are respectively used to obtain the ARMA model by using the normal data of the segment to which they belong, and then the wind speed of the time series corresponding to the abnormal data is obtained by using the ARMA model; If the group of abnormal data does not contain segmentation points, the normal data before and after the abnormal data is used directly to obtain the ARMA model, and then the ARMA model is used to obtain the wind speed of the time series corresponding to the abnormal data; based on the output characteristic formula of the fan, the wind speed is obtained The output sequence of various types of wind turbines in the wind farm; calculate the output sequence of the starting wind turbines in the wind farm, and obtain the output data of the repaired wind farm.

Among them, the screening and classification of abnormal data: abnormal data mainly refers to repeated data, missing data and unreasonable data. As shown in Figure 2, repeated data refers to multiple different photovoltaic power station output data records corresponding to a certain moment; missing data refers to incomplete photovoltaic power station output data records corresponding to a certain moment, where "incomplete" means It points out that the data items stored in the power data records are not enough to be deduced from each other's physical meaning; unreasonable data refers to the output data records of photovoltaic power plants that do not conform to the physical reality.

According to the above classification, search and filter three types of abnormal data in turn: find data records corresponding to multiple data outputs at the same time point, these data records are duplicate data; find time points without output data in the data repair time window, all of these Corresponding to the missing data, screen out the output data of photovoltaic power plants with the same data records for more than 4 consecutive time points, the data records with output data greater than the start-up capacity, and the data records with output at night. These data records are unreasonable data.

When screening unreasonable data, the accuracy and error of data records should be considered. When the start-up capacity is P _on , the output of the photovoltaic power station can not exceed the start-up capacity of the photovoltaic power station when the output is in the interval [-αP _on , (1+α)P _on ], and the value of α can be selected from 0.03 to 0.1 according to the data quality.

Merge the abnormal data adjacent to the time point into an abnormal data group. If the number of elements in the abnormal data group is not less than 5, the data records in this abnormal data group are continuous abnormal data records, otherwise they are local abnormal data Record.

Repair of local anomaly data: The properties of local anomaly data and continuous anomaly data are different. The time interval corresponding to the former is shorter, and the change of meteorological data in this time interval is not obvious. At this time, the main factors affecting photovoltaic output Because of the characteristics of local areas, it does not need to be repaired with meteorological data. Therefore, the present invention adopts different repair methods for local abnormal data and continuous abnormal data.

For local abnormal data, the polynomial interpolation method is directly used for repair. If the number of local abnormal data is N, take the data [N/2] items before the local abnormal data, and the data [N/2] items after, and the abscissa values corresponding to these data are marked as 1, 2, ..., [ N/2], N+[N/2], N+[N/2]+1,..., 2N; the N-order polynomial fitting function is obtained by fitting the above N points; the calculation fitting function is in [N/2] +1, [N/2]+2, ..., [N/2]+N values are used as repair output data.

Segment point judgment based on KS test: take the wind speed data of the adjacent day before and after the abnormal data group, and record them as S1 and S2 respectively. The present invention uses KS test (Kolmogorov-Smirnov test) to come whether S1 and S2 test are identically distributed, and the steps are as follows:

1) Assume that S1 and S2 obey the same distribution;

2) Calculate the cumulative probability of two sets of wind speed data, which are respectively recorded as F1,n(x) and F1,n(x), where n is the data volume of S1 and S2, and the cumulative probability is defined as follows:

Among them, I[-∞,x](Xi) is an indicator function, that is, Xi<x is 1, otherwise it is 0;

3) Calculate KS statistics:

Among them, sup is the supremum operation;

4) Test whether to reject the hypothesis

If the following formula is satisfied, the hypothesis is rejected (at the 0.05 level), S1 and S2 obey different distributions, that is, there is a segmentation point in this group of abnormal data, otherwise there is no segmentation point.

If there are segment points in the abnormal data, use the uniform distribution to sample the location of the segment points.

4. Parameter estimation of ARMA model

The ARMA model (autoregressive moving average model) is a typical method of the wind speed model. The present invention uses the ARMA (3,3) model to simulate the wind speed, that is, both the autoregressive component and the moving average component use a 3rd order model. The time series form of ARMA(3,3) is as follows:

Where c is a constant, εt is white noise (that is, a random variable that obeys the Gaussian distribution with an expectation of 0 and a variance of δ2), and φi and θi are the parameters of the model.

The present invention utilizes an auto-regressive approximation method for parameter estimation. Note that the normal wind speed sequence used for parameter estimation is X, the length is n, and the parameters to be estimated are φ and θ, δ2, c.

1) Estimated constant c from the expectation of the wind speed sequence

2) Estimate the parameter φ of the wind speed sequence corresponding to the AR(3) model

remember

reaching the minimum value of s(φ) That is, the least squares estimate of φ. Ruoji

Then s(φ) can be written as

Then the least squares estimate of φ is

3) Calculate the wind speed sequence residual

Based on step 2) get Find the residual error of the wind speed sequence as

4) Calculate the parameters φ, θ and δ2 in ARMA(3,3)

remember

Then reach the minimum value of Q(φ,θ) with That is, the least squares estimation of φ and θ.

The least squares estimate of δ2 is

The invention provides a wind farm output data restoration method that takes into account the segmentation characteristics of wind speed data. This method can consider the non-stationary characteristics of wind speed in a short period of time, and improve the accuracy of wind farm output data restoration by short-term segmentation. . The present invention can more effectively correct abnormal records in wind farm output data and improve the quality of wind farm output data; this is conducive to improving the decision-making level of power grid planning and operation, improving the decision-making accuracy of auxiliary services in the power grid, and reducing unnecessary System backup, thereby improving the economy of power grid construction and operation.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art can still implement the present invention Any modification or equivalent replacement that does not deviate from the spirit and scope of the present invention is within the protection scope of the pending claims.

Claims (4)

1. a kind of meter and the segmented characterizing output of wind electric field data recovery method of air speed data, it is characterised in that the reparation side Method is comprised the following steps：

S1, the screening from the data for obtaining repeats, lacks and irrational abnormal data, and continuous according to corresponding to abnormal data Seasonal effect in time series length, is divided into continuous abnormal type and the class of local ectype two；

S2, for local anomaly type data, the output of wind electric field data for obtaining repairing using interpolation method；

S3, for continuous abnormal type data, based on maximum a posteriori probability, judges abnormal using the normal data before and after abnormal data Whether data contain waypoint, and the wind speed characteristics for being then based on per section are obtained by normal data or mode identification method, based on this Wind speed characteristics, the wind speed repaired is generated using arma modeling, further obtains the output of wind electric field data repaired；

S4, verifies the effectiveness of repair data, exports recovery report.

2. a kind of meter and the segmented characterizing output of wind electric field data recovery method of air speed data according to claim 1, it is special Levy and be, the S1 is specially：

The data record that many datas of same time point correspondence are exerted oneself is searched, these data records are duplicate data；Searching data Without the time point for going out force data in repair time window, corresponding to these time points for missing data, screen output of wind electric field Continuous 4 time points data above identical data record in data, go out data record of the force data more than start capacity and night Between there is the data record exerted oneself, these data records be unreasonable data；If being continuously no less than 5 time point corresponding datas Record be abnormal data, then the data record corresponding to these time points be continuous abnormal type data record, remaining abnormal number According to for local anomaly type data record.

3. a kind of meter and the segmented characterizing output of wind electric field data recovery method of air speed data according to claim 1, it is special Levy and be, the S2 is specially：

Note local anomaly type data amount check is N, takes data [N/2] item before local anomaly type data, afterwards data [N/2] item, The corresponding abscissa value of these data is denoted as 1,2 ..., [N/2], N+ [N/2], N+ [N/2]+1 ..., 2N；With above-mentioned N number of Point fitting obtains N rank multinomial fitting functions；Digital simulation function in [N/2]+1, [N/2]+2 ..., the value conduct of [N/2]+N That what is repaired goes out force data.

4. a kind of meter and the segmented characterizing output of wind electric field data recovery method of air speed data according to claim 1, it is special Levy and be, the S3 specifically includes following steps：

The normal data before and after this group of abnormal data is chosen, the normal data length before and after abnormal data is 1 day；Examined based on KS Test, judge whether abnormal data contains waypoint using the normal data before and after abnormal data；If containing in this group of abnormal data Waypoint, sampling waypoint position, data are utilized respectively the affiliated normal data being segmented and obtain arma modeling before and after waypoint, so Obtain seasonal effect in time series wind speed corresponding to abnormal data using arma modeling afterwards；If not containing waypoint in this group of abnormal data, Then directly arma modeling is obtained using the normal data before and after abnormal data, then obtain abnormal data institute using arma modeling Correspondence seasonal effect in time series wind speed；Based on the power producing characteristics formula of blower fan, the sequence of exerting oneself of all types of blower fans of wind energy turbine set is obtained by wind speed Row；The blower fan started shooting in wind energy turbine set sequence of exerting oneself is asked into each, the output of wind electric field data repaired are obtained.