A wind power generation forecasting method combining long-term and short-term

Abstract

The present invention discloses a wind power generation prediction method combining long-term and short-term. The method combines wind power generation data of three different time scales, and through progressive optimization and updating of the latest predicted power generation, it can not only improve the prediction accuracy, but also use the wind power generation prediction demand of different scenarios, such as daytime and hourly grid-connected power prediction. The wind power generation prediction method of the present invention has high prediction accuracy, and can provide a basis for the power grid to formulate effective absorption, thereby alleviating the problem of wind abandonment and ensuring the safe operation of the power grid.

Description

Technical field:

[0001] The invention relates to a long-term and short-term combined wind power generation prediction method, belonging to the field of power system power generation. Background technology:

[0002] Wind energy is a renewable energy source other than hydropower that is large in scale, has good commercial development prospects, and has relatively mature development technology. Actively developing and utilizing wind power resources is of great significance to adjusting the energy structure, reducing environmental pollution, and responding to climate change.

[0003] However, wind power output is volatile and intermittent, which results in inaccurate forecasts of wind power generation and difficulty in scheduling and management, making it very unfavorable for wind power consumption. It will also lead to a large number of wind abandonment problems and have an adverse impact on the safe operation of the power grid.

[0004] Currently, the common wind power generation prediction method uses a single time scale data source to conduct prediction using a neural network or based on a power grid model. However, the neural network prediction method has high computational complexity and unclear mechanism, while the accuracy of the power grid model-based method is heavily dependent on the accuracy of the power grid model. Although there are individual prediction methods that combine data sources of multiple time scales, these methods do not consider the mutual influence of data at each time scale, treat data at different time scales independently, and then perform weighted averaging of the prediction values ​​at different time scales. Therefore, they do not truly utilize the data characteristics of different time scales and it is difficult to improve the prediction accuracy. Summary of the invention:

[0005] In order to solve the problem of wind blowing, the present invention provides a wind power generation prediction method combining long-term and short-term. The method combines wind power generation data of three different time scales as a data source, which can accurately determine the accuracy of wind power generation and lay the foundation for the safe operation of the power grid.

[0006] The technical solution of the present invention is as follows:

[0007] A wind power generation forecasting method combining long-term and short-term, comprising:

[0008] Based on the wind power generation sequence and weather type data of the previous year, typical daily wind power generation sequences of different weather types in each season are constructed;

[0009] Determine the typical daily wind power generation sequence corresponding to the forecast day according to the season of the forecast day and the forecast weather type of the forecast day;

[0010] According to the correlation between the wind power generation sequence of the typical day corresponding to the forecast day and the wind power generation sequence of each day in the week before the forecast day, the wind power generation sequence of the forecast day is predicted;

[0011] The predicted wind power generation sequence for the forecast day is corrected according to the actual real-time power generation sequence for the forecast day.

[0012] Furthermore, the method for constructing a typical daily wind power generation sequence for different weather types in each season includes:

[0013] Obtain the daily wind power generation sequence and daily weather type data of the previous year; divide the daily wind power generation sequence of the previous year by season to form a data set of four seasons; then divide the daily wind power generation sequence in each season data set according to the weather type of the day according to the daily weather type data of the previous year to form multiple subsets; the elements in each subset are recorded as is the wind power generation sequence of the i-th day of the j-th weather type in the sea-th season; sea is the season number, sea∈[1,4]; j is the weather type number, j∈[1,J], J is the total number of weather types; i is the day number in the j-th weather type; t is the sequence number of the wind power generation sequence, t is [1,T], T is the total number of wind power generation sequences;

[0014] The DBSCAN algorithm is used to analyze the wind power generation sequences with the same weather type in each subset. Perform clustering and merging, and determine the typical day of each subset based on the clustering results.

[0015] Furthermore, the weather types include cloudy, sunny, overcast, sunny to overcast, sleet, thunderstorm, light rain, heavy rain, moderate rain, rainstorm, heavy snow, light snow, moderate snow and hail, a total of 14 weather types, and the total number of weather types is 14.

[0016] Furthermore, the DBSCAN algorithm is used to find all dense areas of sample points, and each dense area is used as a cluster. Points with a number of sample points greater than or equal to minpoints within a neighborhood radius R are recorded as core points, points that are not core points but within the neighborhood of a core point are recorded as boundary points; points that are neither core points nor boundary points are recorded as noise points. The DBSCAN algorithm includes the following two steps:

[0017] Step 1: Find core points to form temporary clusters;

[0018] Scan all sample points. If the number of points within the radius R of a sample point is greater than or equal to MinPoints, it will be included in the core point list, and the points directly accessible by its density will form a corresponding temporary cluster.

[0019] Step 2: Merge temporary clusters to obtain clusters;

[0020] For each temporary cluster, check whether the point in it is a core point. If so, merge the temporary cluster corresponding to the point with the current temporary cluster to obtain a new temporary cluster.

[0021] Repeat this operation until every point in the current temporary cluster is either not in the core point list or all points directly accessible by its density are already in the temporary cluster, and the temporary cluster is upgraded to a cluster.

[0022] Continue to perform the same merging operation on the remaining temporary clusters until all temporary clusters are processed;

[0023] Furthermore, the neighborhood radius R is 2% of the rated power of the wind turbine, and the minimum number of points minpoints is 10 or adjusted according to the power sequence.

[0024] Furthermore, the specific steps of determining the typical day of each subset according to the clustering result include:

[0025] Read the clustering result with the largest number of days, assuming it contains Z days, and calculate the cumulative power of each day X(z) = [X1...X Z ], and then sort the calculated cumulative power X(z) from small to large, and select the wind power generation sequence corresponding to the median as the typical daily power sequence P with season sea and weather type j sea,j typ (t), i.e., the typical daily wind power generation sequence; wherein, the calculation formula of the accumulated power X(z) on the zth day is as follows:

[0026]

[0027] Where P z (t) is the wind power generation sequence on the zth day, and z is [1, Z].

[0028] Furthermore, the specific steps of predicting the wind power generation sequence on the forecast day include:

[0029] The similarity between the typical day wind power generation sequence corresponding to the forecast day and the wind power generation sequence of each day in the last week is calculated as follows:

[0030]

[0031] Where Pcc(n) is the similarity between the typical day wind power generation sequence corresponding to the forecast day and the wind power generation sequence of the nth day in the most recent week, and the value of n ranges from 1 to 7; P wek,n (t) is the wind power generation sequence of the nth day in the latest week; P sea,j typ (t) is the typical daily wind power generation sequence corresponding to the forecast day;

[0032] Select the wind power generation sequence corresponding to the three largest values ​​in Pcc(n) and record it as P pcc,r (t), r = 1, 2, 3, the wind power generation sequence of the forecast day is predicted as follows:

[0033]

[0034]

[0035]

[0036]

[0037] Where P new1 (t) is the predicted wind power generation sequence of the forecast day; is the correction factor for the upcoming week, The value range of P is [-1,1]; ad (t) is based on P pcc,r (t) The calculated wind power generation sequence for the forecast day; w r P pcc,r (t) The corresponding weight, d r The wind power generation sequences corresponding to the three largest values ​​in Pcc(n) are respectively related to the typical day wind power generation sequence P corresponding to the forecast day. sea,j typ (t) distance;

[0038] Furthermore, the specific steps of correcting the predicted wind power generation sequence of the predicted day are as follows:

[0039] The actual real-time power generation sequence of the forecast day is obtained, and the cumulative deviation Δ between the predicted wind power generation sequence of the forecast day and the actual real-time power generation sequence is calculated. The calculation formula of the cumulative deviation Δ is as follows:

[0040]

[0041] Where Δ is the cumulative deviation between the predicted wind power generation sequence on the forecast day and the actual real-time power generation sequence; P real (t) is the actual real-time power generation sequence on the forecast day, the value range of t is [1, L], L is the maximum time series of actual power generation on the forecast day, L <T;P new1 (t) is the predicted wind power generation sequence of the forecast day;

[0042] If the cumulative deviation Δ is less than the set threshold τ, the predicted wind power generation sequence for the forecast day remains unchanged;

[0043] If the cumulative deviation Δ exceeds the set threshold τ, the wind power generation sequence of the predicted day is corrected based on the calculated cumulative deviation to obtain the corrected wind power generation sequence P of the predicted day. new2 (t), the modified formula is as follows:

[0044]

[0045] Where P new2 (t) is the revised wind power generation sequence of the forecast day, the value range of t is [L,T]; τ is 1% of the rated power of the wind turbine.

[0046] Compared with the prior art, the present invention has the following beneficial effects

[0047] The long-term and short-term combined wind power generation prediction method of the present invention combines wind power generation data of three different time scales. By progressively optimizing and updating the latest predicted power generation, it can not only improve the prediction accuracy, but also use the wind power generation prediction demand of different scenarios, such as daytime and hourly grid-connected power prediction.

[0048] The wind power generation prediction method of the present invention has high prediction accuracy and can provide a basis for the grid to formulate effective absorption, thereby alleviating the problem of wind abandonment and ensuring the safe operation of the grid. Description of the drawings:

[0049] The above and other features of the present invention will become more apparent from the following detailed description of preferred non-limiting embodiments of the present invention with reference to the accompanying drawings, in which:

[0050] Figure 1 A flow chart of the wind power generation forecasting method combining long-term and short-term;

[0051] Figure 2 Construct a flow chart for a typical day's wind generation sequence;

[0052] Figure 3 The forecast flow chart of wind power generation sequence for the forecast day;

[0053] Figure 4 This is the correction flow chart of the wind power generation sequence for the forecast day. Specific implementation method:

[0054] The present invention will be further described below in conjunction with specific embodiments and corresponding drawings.

[0055] Embodiment 1:

[0056] A wind power generation prediction method combining long-term and short-term of the present invention comprises the following specific steps:

[0057] 1) Based on the wind power generation sequence and weather type data of the previous year, typical daily wind power generation sequences of different weather types in each season are constructed respectively; the specific steps include:

[0058] First, obtain the daily wind power generation data and the corresponding weather type data of the previous year, among which the daily wind power generation data adopts the daily wind power generation sequence; divide the wind power generation sequence of the previous year into four seasons: spring, summer, autumn and winter, to form data sets for four seasons; the seasons can be divided by various division methods such as astronomy, solar terms, lunar calendar, temperature, and meteorological division. This example adopts the meteorological division method, that is, March to May is spring, June to August is summer, September to November is autumn, and December to February is winter. Then, based on the daily wind power generation sequence of the previous year, the wind power generation data in the data set of each season is divided according to the weather type of the day to form multiple subsets; the elements in each subset are recorded as is the wind power generation sequence of the i-th day of the j-th weather type in the sea-th season; sea is the season number, sea∈[1,4], corresponding to spring, summer, autumn and winter respectively; j is the weather type serial number, j∈[1,J], and J is the total number of weather types; in this example, the weather types include cloudy, sunny, cloudy, sunny to cloudy, sleet, thunderstorm, light rain, heavy rain, moderate rain, rainstorm, heavy snow, light snow, moderate snow and hail, a total of 14 types, that is, the total number of weather types J is 14; i is the day serial number in the j-th weather type, and the value of i depends on the number of days of type j that occurred last year. For example, if there are 25 days of type j weather type in total, then the value range of i is from 1 to 25; t is the serial number of the wind power generation sequence, t is [1,T], and T is the total number of wind power generation sequences. For example, if one power is generated at an interval of 15 minutes every day, then T=96, and the value range of t is [1,96], and there are 96 power sequences every day;

[0059] The DBSCAN algorithm is used to analyze the wind power generation sequences with the same weather type in each subset. Perform clustering and merging to obtain k categories; the DBSCAN algorithm can find all dense areas of sample points and use the dense areas as a cluster respectively. Points with a number of sample points greater than or equal to minpoints within the neighborhood radius R are recorded as core points, points that are not core points but within the neighborhood of a core point are recorded as boundary points; points that are neither core points nor boundary points are recorded as noise points; the DBSCAN algorithm includes the following two steps:

[0060] Step 1: Find core points to form temporary clusters;

[0061] Scan all sample points. If the number of points within the radius R of a sample point is ≥ MinPoints, it will be included in the core point list, and the points directly accessible by its density will form a corresponding temporary cluster.

[0062] Step 2: Merge temporary clusters to obtain clusters;

[0063] For each temporary cluster, check whether the point in it is a core point. If so, merge the temporary cluster corresponding to the point with the current temporary cluster to obtain a new temporary cluster.

[0064] Repeat this operation until every point in the current temporary cluster is either not in the core point list or all points directly accessible by its density are already in the temporary cluster, and the temporary cluster is upgraded to a cluster.

[0065] Continue to perform the same merging operation on the remaining temporary clusters until all temporary clusters are processed;

[0066] In this example, the neighborhood radius R is 2% of the rated power of the wind turbine, and the minimum number of points minpoints is 10, which can also be adjusted according to the wind power generation sequence;

[0067] In the above method, the specific steps of determining the typical day of each subset according to the clustering results include:

[0068] Read the clustering result with the largest number of days, assuming it contains Z days, and calculate the cumulative power of each day X(z) = [X1...X Z ], and then sort the calculated cumulative power X(z) from small to large, and select the wind power generation sequence corresponding to the median as the typical daily power sequence P with season sea and weather type j sea,j typ (t), i.e., the typical daily wind power generation sequence; wherein, the calculation formula of the accumulated power X(z) on the zth day is as follows:

[0069]

[0070] Where P z (t) is the wind power generation sequence on the zth day, and z is [1, Z];

[0071] 2) According to the season of the forecast day and the forecast weather type of the forecast day, determine the typical day wind power generation sequence corresponding to the forecast day; for example, the forecast day is May 1, the weather forecast predicts that the forecast day is cloudy, and the weather type sequence corresponding to cloudy is 1, then the typical day wind power generation sequence corresponding to the forecast day is expressed as That is, the typical daily wind power generation sequence of summer weather type 1

[0072] 3) According to the correlation between the typical day wind power generation sequence corresponding to the forecast day and the wind power generation sequence of each day in the previous week (7 days before the forecast day), the wind power generation sequence of the forecast day is predicted; the specific steps include:

[0073] The similarity between the typical day wind power generation sequence corresponding to the forecast day and the wind power generation sequence of each day in the last week is calculated as follows:

[0074]

[0075] Where Pcc(n) is the similarity between the typical day wind power generation sequence corresponding to the forecast day and the wind power generation sequence of the nth day in the most recent week, and the value of n ranges from 1 to 7; P wek,n (t) is the wind power generation sequence of the nth day in the latest week; P sea,j typ (t) is the typical daily wind power generation sequence corresponding to the forecast day;

[0076] Select the wind power generation sequence corresponding to the three largest values ​​in Pcc(n) and record it as P pcc,r (t), r = 1, 2, 3, the wind power generation sequence of the forecast day is predicted as follows:

[0077]

[0078]

[0079]

[0080]

[0081] Where P new1 (t) is the predicted wind power generation sequence of the forecast day; is the correction factor for the upcoming week, The value range of P is [-1,1]; ad (t) is based on P pcc,r (t) The wind power generation sequence of the predicted day is calculated; w r P pcc,r (t) The corresponding weight, d r The wind power generation sequences corresponding to the three largest values ​​in Pcc(n) are respectively related to the typical day wind power generation sequence P corresponding to the forecast day. sea,j typ (t) distance;

[0082] 4) Correct the predicted wind power generation sequence on the predicted day according to the actual real-time power generation sequence on the predicted day. The specific steps are as follows:

[0083] The actual real-time power generation sequence of the forecast day is obtained, and the cumulative deviation Δ between the predicted wind power generation sequence of the forecast day and the actual real-time power generation sequence is calculated. The calculation formula of the cumulative deviation Δ is as follows:

[0084]

[0085] Where Δ is the cumulative deviation between the predicted wind power generation sequence on the forecast day and the actual real-time power generation sequence; P real (t) is the actual real-time power generation sequence on the forecast day, the value range of t is [1, L], L is the maximum time series of actual power generation on the forecast day, L <T;

[0086] If the cumulative deviation Δ is less than the set threshold τ, the predicted wind power generation sequence for the forecast day remains unchanged;

[0087] If the cumulative deviation Δ exceeds the set threshold τ, the wind power generation sequence of the predicted day is corrected based on the calculated cumulative deviation to obtain the corrected wind power generation sequence P of the predicted day. new2 (t), the modified formula is as follows:

[0088]

[0089] Where P new2 (t) is the revised wind power generation sequence of the forecast day, the value range of t is [L,T]; τ is 1% of the rated power of the wind turbine.

Claims

1. A wind power generation prediction method combining long-term and short-term, characterized by: include: Based on the wind power generation sequence and weather type data of the previous year, typical daily wind power generation sequences of different weather types in each season are constructed; The method for constructing a typical daily wind power generation sequence of different weather types in each season includes: Obtain the daily wind power generation sequence and daily weather type data of the previous year; divide the daily wind power generation sequence of the previous year by season to form a data set of four seasons; then divide the daily wind power generation sequence in the data set of each season according to the weather type of the day according to the daily weather type data of the previous year to form multiple subsets; the elements in each subset are recorded as is the wind power generation sequence of the i-th day of the j-th weather type in the sea-th season; sea is the season number, sea∈[1,4]; j is the weather type number, j∈[1,J], J is the total number of weather types; i is the day number in the j-th weather type; t is the sequence number of the wind power generation sequence, t is [1,T], T is the total number of wind power generation sequences; The DBSCAN algorithm is used to analyze the wind power generation sequences with the same weather type in each subset. Perform clustering and merging, and determine the typical day of each subset based on the clustering results; The specific steps of determining the typical day of each subset according to the clustering result include: Read the clustering result with the largest number of days, assuming it contains Z days, and calculate the cumulative power of each day X(z) = [X1...X Z ], and then sort the calculated cumulative power X(z) from small to large, and select the wind power generation sequence corresponding to the median as the typical daily power sequence P with season sea and weather type j sea,j typ (t), i.e., the typical daily wind power generation sequence; wherein, the calculation formula of the accumulated power X(z) on the zth day is as follows: Where P z (t) is the wind power generation sequence on the zth day, and z is [1, Z]; The weather types include cloudy, sunny, overcast, sunny to overcast, sleet, thunderstorm, light rain, heavy rain, moderate rain, rainstorm, heavy snow, light snow, moderate snow and hail, a total of 14 weather types, and the total number of weather types is 14; Determine the typical daily wind power generation sequence corresponding to the forecast day according to the season of the forecast day and the forecast weather type of the forecast day; According to the correlation between the wind power generation sequence of the typical day corresponding to the forecast day and the wind power generation sequence of each day in the week before the forecast day, the wind power generation sequence of the forecast day is predicted; Correct the predicted wind power generation sequence of the forecast day according to the actual real-time power generation sequence of the forecast day; The specific steps of predicting the wind power generation sequence on the prediction day include: The similarity between the typical day wind power generation sequence corresponding to the forecast day and the wind power generation sequence of each day in the last week is calculated as follows: Where Pcc(n) is the similarity between the typical day wind power generation sequence corresponding to the forecast day and the wind power generation sequence of the nth day in the most recent week, and the value of n ranges from 1 to 7; P wek,n (t) is the wind power generation sequence of the nth day in the latest week; P sea,j typ (t) is the typical daily wind power generation sequence corresponding to the forecast day; Select the wind power generation sequence corresponding to the three largest values ​​in Pcc(n) and record it as P pcc,r (t), r = 1, 2, 3, the wind power generation sequence of the forecast day is predicted as follows: Where P new1 (t) is the predicted wind power generation sequence of the forecast day; is the correction factor for the upcoming week, The value range of P is [-1,1]; ad (t) is based on P pcc,r (t) The calculated wind power generation sequence for the forecast day; w r P pcc,r (t ) The corresponding weight, d r The wind power generation sequences corresponding to the three largest values ​​in Pcc(n) are respectively related to the typical day wind power generation sequence P corresponding to the forecast day. sea,j typ (t) distance; The specific steps of correcting the predicted wind power generation sequence of the predicted day are as follows: The actual real-time power generation sequence of the forecast day is obtained, and the cumulative deviation Δ between the predicted wind power generation sequence of the forecast day and the actual real-time power generation sequence is calculated. The calculation formula of the cumulative deviation Δ is as follows: Where Δ is the cumulative deviation between the predicted wind power generation sequence on the forecast day and the actual real-time power generation sequence; P real (t) is the actual real-time power generation sequence on the forecast day, the value range of t is [1, L], L is the maximum time series of actual power generation on the forecast day, L <T;P new1 (t) is the predicted wind power generation sequence of the forecast day; If the cumulative deviation Δ is less than the set threshold τ, the predicted wind power generation sequence for the forecast day remains unchanged; If the cumulative deviation Δ exceeds the set threshold τ, the wind power generation sequence of the predicted day is corrected based on the calculated cumulative deviation to obtain the corrected wind power generation sequence P of the predicted day. new2 (t), the modified formula is as follows: Where P new2 (t) is the wind power generation sequence of the revised forecast day, the value range of t is [L, T]; τ is 1% of the rated power of the wind turbine; The DBSCAN algorithm is used to find all dense areas of sample points, and each dense area is used as a cluster. Points with a number of sample points greater than or equal to minpoints within a neighborhood radius R are recorded as core points, points that are not core points but within the neighborhood of a core point are recorded as boundary points; points that are neither core points nor boundary points are recorded as noise points. The DBSCAN algorithm includes the following two steps: Step 1: Find core points to form temporary clusters; Scan all sample points. If the number of points within the radius R of a sample point is greater than or equal to MinPoints, it will be included in the core point list, and the points directly accessible by its density will form a corresponding temporary cluster. Step 2: Merge temporary clusters to obtain clusters; For each temporary cluster, check whether the point in it is a core point. If so, merge the temporary cluster corresponding to the point with the current temporary cluster to obtain a new temporary cluster. Repeat this operation until every point in the current temporary cluster is either not in the core point list or all points directly accessible by its density are already in the temporary cluster, and the temporary cluster is upgraded to a cluster. Continue to perform the same merging operation on the remaining temporary clusters until all temporary clusters are processed.

2. The long-term and short-term combined wind power generation prediction method according to claim 1 is characterized by: The neighborhood radius R is 2% of the rated power of the wind turbine, and the minimum number of points minpoints is 10.

Images