A wind power prediction method

Abstract

The present invention provides a wind power prediction method, comprising: for any wind turbine in a wind farm, obtaining the error between the real-time wind power and the predicted wind power predicted by a neural network model; selecting the current moment, and determining the distribution characteristics according to the real-time wind power error in the historical period before the current moment; establishing a Bayesian constant mean dynamic prediction model for error prediction according to the distribution characteristics and the real-time wind power error in the backward period after the current moment; and compensating the prediction result of the Bayesian constant mean dynamic prediction model to the predicted wind power predicted by the neural network model to obtain the compensated power. The wind power prediction method of the present invention has the characteristics of high accuracy and fast response, and can be widely used in the field of wind farm power error compensation.

Description

Technical Field

[0001] The invention relates to a wind power detection technology, and in particular to a wind power prediction method. Background Art

[0002] At present, the main forms of renewable energy power generation include wind power, photovoltaic power, nuclear power, and hydropower. Compared with other forms of energy generation, wind power generation has developed rapidly in renewable energy with its own unique advantages. In the process of wind power development, the uncertainty of wind energy has brought great troubles to power grid dispatching. How to improve the accuracy of wind power prediction is extremely necessary and urgent.

[0003] Traditional wind power forecasting methods generally use mathematical models based on historical data from wind power plants to make forecasts. However, due to inaccurate data and complex mathematical models, these methods lead to problems such as low forecast accuracy and slow response.

[0004] It can be seen that in the prior art, there is no wind power prediction method with high prediction accuracy and fast response speed. Summary of the invention

[0005] In view of this, the main purpose of the present invention is to provide a wind power prediction method with high prediction accuracy and fast response speed.

[0006] In order to achieve the above object, the technical solution proposed by the present invention is:

[0007] A wind power prediction method comprises the following steps:

[0008] Step 1: According to the real-time wind speed, real-time wind direction, and real-time atmospheric humidity of the wind farm, obtain the real-time wind power of any wind turbine in the wind farm at any time t. ; Among them, real-time wind power is a real number and t is a natural number.

[0009] Step 2: According to the real-time wind speed, real-time wind direction, and real-time atmospheric humidity of the wind farm, the neural network model is used to predict the wind power and obtain the predicted wind power of any wind turbine in the wind farm at any time t. ; Among them, the predicted wind power is a real number.

[0010] Step 3: Select the current time m: calculate the real-time wind power corresponding to each historical time within the time length L before the current time m Predicted wind power By comparison, the real-time wind power error E in the historical period can be obtained. h =[e m , e m-1 ,…,e m-i ,…,e m-L]; where the real-time error of historical wind power at any time within the time length L before the current time m is m, i, and L are all natural numbers, and 0≤i≤L; e m 、e m-1 , …, e m-i ,…,e m-L All are real numbers.

[0011] Step 4: Calculate the real-time error E of wind power during historical periods h The distribution characteristics of the wind power in the historical period are obtained by h The statistical mean is μ e , the statistical variance is σ e The normal distribution of wind power and the real-time error E of the historical period h Statistical histogram of .

[0012] Step 5: Obtain the real-time error E of wind power in the backward period within the time length S after the current time m p =[e m+1 ,…,e m+j ,…,e m+S ]; where the real-time error of the backward wind power at any time within the time length S after the current time m is j and S are both natural numbers, and 1≤j≤S; e m+1 ,…,e m+j ,…,e m+S All are real numbers.

[0013] Step 6: The real-time error E of wind power in the historical period determined in step 4 h The statistical mean is μ e , the statistical variance is σ e The normal distribution of the wind power in the backward period is used to determine the real-time error E of wind power. p It also follows that the statistical mean is μ e , the statistical variance is σ e The normal distribution of .

[0014] Step 7: The real-time error E of wind power in the backward period determined in step 6 p The distribution characteristics of wind power in the backward period are p , establish a Bayesian constant mean dynamic prediction model for error prediction, and obtain the wind power prediction error at the current moment

[0015] Step 8: The wind power prediction error at the current moment obtained in step 7 is Compensate for the current predicted wind power obtained in step 2 , get the current moment compensation wind power of any wind turbine:

[0016]

[0017] In step 7, the Bayesian constant mean dynamic prediction model is as follows:

[0018] The observation equation is:

[0019] The state equation is: E p =E p-1 +W m ;

[0020] Among them, the observation error V m =[v m , v m+1 , …, v m+j , …, v m+S ]、State error W m =[w m , w m+1 ,…,w m+j ,…,w m+S ] are all statistically mean 0 and statistically variance σ e The normal distribution of .

[0021] The real-time error E of wind power in the backward period p The mean value of m at the current time The real-time error E of wind power in the backward period p The mean value at the next moment m+1 The real-time error E of wind power in the backward period p The variance of m at the current time The real-time error E of wind power in the backward period p The variance at the next moment m+1 The observation error v at the current time m m , the state error w of m at the current moment m , the observation error v at the next moment m+1 m+1 、The next moment m+1 state error w m+1 The relationship between them is as follows:

[0022]

[0023] In summary, in the wind power prediction method of the present invention, firstly, the real-time wind power at each moment is obtained. The predicted wind power obtained by the neural network model Determine a current time m, and calculate the real-time wind power within a certain period of time before the current time. The predicted wind power obtained by the neural network model The real-time error between h, determine its statistical distribution characteristics. Real-time error of wind power in historical period E h The determination of the distribution characteristics means that the real-time wind power within a certain period of time after the current time m is The predicted wind power obtained by the neural network model The real-time error of wind power in the backward period E p The distribution characteristics of the wind power in the historical period are also determined. h Distribution characteristics and real-time error E of wind power in the backward period p This is determined by the statistical properties of probability and mathematical statistics that are well known to the public. Then, according to the determined distribution characteristics and the real-time error E of wind power in the backward period, p , establish a Bayesian constant mean dynamic prediction model, and perform error prediction to obtain the wind power prediction error Finally, the wind power prediction error Compensate for the predicted wind power predicted by the neural network model Thus, the wind power prediction method of the present invention improves the prediction accuracy of wind power by further compensating the wind power predicted by the neural network model. In addition, since the wind power prediction method of the present invention can select the time length and the algorithm is simple, its prediction response speed is also relatively fast. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The figure is a schematic diagram of the overall flow of the wind power prediction method of the present invention.

[0025] Figure 2 Part of the predicted wind power predicted by the radial basis function neural network Partial real-time wind power Schematic diagram of the difference.

[0026] Figure 3 for Figure 2 The corresponding backward period partial forecast wind power Partial real-time wind power A comparison diagram of .

[0027] Figure 4 is the real-time error E of wind power in 3000 historical periods in this embodiment. h Schematic diagram of statistical data.

[0028] Figure 5 is the real-time error of wind power in the backward period within 5 minutes, E p Schematic diagram of .

[0029] Figure 6 The wind power at each moment predicted by the Bayesian constant mean dynamic prediction model within 5 minutes Partial real-time wind power A comparison diagram of .

[0030] Figure 7 Compensate wind power for 5 minutes at the current moment Partially predicted wind power predicted by radial basis function neural network Partial real-time wind power Schematic diagram for comparison of the results. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] Figure 1 FIG. 1 is a flow chart of the wind power prediction method of the present invention. Figure 1 As shown, the wind power prediction method of the present invention comprises the following steps:

[0033] Step 1: According to the real-time wind speed, real-time wind direction, and real-time atmospheric humidity of the wind farm, obtain the real-time wind power of any wind turbine in the wind farm at any time t. Among them, real-time wind power is a real number and t is a natural number.

[0034] In step 1, the real-time wind direction includes: a real-time wind direction sine value and a real-time wind direction cosine value.

[0035] Step 2: According to the real-time wind speed, real-time wind direction, and real-time atmospheric humidity of the wind farm, the neural network model is used to predict the wind power and obtain the predicted wind power of any wind turbine in the wind farm at any time t. Among them, the predicted wind power is a real number.

[0036] In the method of the present invention, the neural network model is a prior art and will not be described in detail here. In practical applications, since there are many neural network models, a specific neural network model can be used according to actual needs.

[0037] Step 3: Select the current time m: calculate the real-time wind power corresponding to each historical time within the time length L before the current time m Predicted wind power By comparison, the real-time wind power error E in the historical period can be obtained. h =[e m , e m-1 ,…,e m-i ,…,e m-L ]; where the real-time error of historical wind power at any time within the time length L before the current time m is m, i, and L are all natural numbers, and 0≤i≤L; em 、e m-1 , …, e m-i ,…,e m-L All are real numbers.

[0038] Step 4: Calculate the real-time error E of wind power during historical periods h The distribution characteristics of the wind power in the historical period are obtained by h The statistical mean is μ e , the statistical variance is σ e The normal distribution of wind power and the real-time error E of the historical period h Statistical histogram of .

[0039] In step 4, the statistical mean μ e =19.2, statistical variance σ e =32.

[0040] Step 5: Obtain the real-time error E of wind power in the backward period within the time length S after the current time m p =[e m+1 ,…,e m+j ,…,e m+S ]; where the real-time error of the backward wind power at any time within the time length S after the current time m is j is a natural number, and S is a natural number, and 1≤j≤S; e m+1 ,…,e m+j ,…,e m+S All are real numbers.

[0041] Step 6: The real-time error E of wind power in the historical period determined in step 4 h The statistical mean is μ e , the statistical variance is σ e The normal distribution of the wind power in the backward period is used to determine the real-time error E of wind power. p It also follows that the statistical mean is μ e , the statistical variance is σ e The normal distribution of .

[0042] Step 7: The real-time error E of wind power in the backward period determined in step 6 p The distribution characteristics of wind power in the backward period are p , establish a Bayesian constant mean dynamic prediction model for error prediction, and obtain the wind power prediction error at the current moment

[0043] Step 8: The wind power prediction error at the current moment obtained in step 7 is Compensate for the current predicted wind power obtained in step 2 Get the current compensated wind power of any wind turbine:

[0044]

[0045] In summary, in the wind power prediction method of the present invention, the real-time wind power at each moment is first obtained. The predicted wind power obtained by the neural network model Determine a current time m, and calculate the real-time wind power within a certain period of time before the current time. The predicted wind power obtained by the neural network model The real-time error between h , determine its statistical distribution characteristics. Real-time error of wind power in historical period E h The determination of the distribution characteristics means that the real-time wind power within a certain period of time after the current time m is The predicted wind power obtained by the neural network model The real-time error of wind power in the backward period E p The distribution characteristics of the wind power in the historical period are also determined. h Distribution characteristics and real-time error E of wind power in the backward period p This is determined by the statistical properties of probability and mathematical statistics that are well known to the public. Then, according to the determined distribution characteristics and the real-time error E of wind power in the backward period, p , establish a Bayesian constant mean dynamic prediction model, and perform error prediction to obtain the wind power prediction error Finally, the wind power prediction error Compensate for the predicted wind power predicted by the neural network model Thus, the wind power prediction method of the present invention improves the prediction accuracy of wind power by further compensating the wind power predicted by the neural network model. In addition, since the wind power prediction method of the present invention can select the time length and the algorithm is simple, its prediction response speed is also relatively fast.

[0046] The corresponding values ​​of each historical moment in the time length L before the current moment m are used to obtain the real-time error E of wind power in the historical period. h =[e m , e m-1 ,…,e m-i ,…,e m-L ]; where the real-time error of historical wind power at any time within the time length L before the current time m is m, i, and L are all natural numbers, and 0≤i≤L; e m 、e m-1 , …, e m-i ,…,e m-L All are real numbers.

[0047] In step 7 of the method of the present invention, the Bayesian constant mean dynamic prediction model is specifically as follows:

[0048] The observation equation is:

[0049] The state equation is: E p =E p-1 +W m

[0050] Among them, the observation error V m =[v m , v m+1 , …, v m+j , …, v m+S ]、State error W m =[w m , w m+1 ,…,w m+j ,…,w m+S ] are all statistically mean 0 and statistically variance σ e The normal distribution of .

[0051] In the method of the present invention, the real-time error E of wind power in the backward period p The mean value of m at the current time Real-time error of wind power in the backward period E p The mean value at the next moment m+1 Real-time error of wind power in the backward period E p The variance of m at the current time Real-time error of wind power in the backward period E p The variance at the next moment m+1 The observation error v at the current time m m , the state error w of m at the current moment m , the observation error v at the next moment m+1 m+1 、The next moment m+1 state error w m+1 The relationship between them is as follows:

[0052]

[0053] In practical applications, the real-time error of wind power in the backward period is E p The mean value of m at the current time Real-time error of wind power in the backward period E p The mean value at the next moment m+1 Around the statistical mean μ e Fluctuation up and down; real-time error of wind power in the backward period E p The variance of m at the current time Real-time error of wind power in the backward period E p The variance at the next moment m+1 Around the statistical variance σ e Fluctuates up and down.

[0054] Example

[0055] Take a wind farm in Inner Mongolia as an example. In this wind farm, for any wind turbine, the specific operation steps in the wind power prediction method are as follows.

[0056] In the above step 2, the wind power is predicted by using a neural network model based on the real-time wind speed, real-time wind direction, and real-time atmospheric humidity of the wind farm to obtain the predicted wind power of any wind turbine in the wind farm at any time t. Specifically, the following steps are included: In the wind farm, 7900 data sets corresponding to real-time wind speed, real-time wind direction sine value, real-time wind direction cosine value, and real-time atmospheric humidity are selected according to the time series, and the radial basis function neural network model in the MATLAB experiment box is used to predict the wind power, and the predicted wind power of any wind turbine at time t is obtained. Figure 2 Part of the predicted wind power predicted by the radial basis function neural network Partial real-time wind power Schematic diagram of the difference. Figure 3 for Figure 2 The corresponding backward period partial forecast wind power Partial real-time wind power A comparison diagram of .

[0057] The above step 4 specifically includes the following steps:

[0058] Step 41: Using the above method, obtain 3000 sets of wind power real-time errors E in historical periods h =[e m , e m-1 ,…,e m-3000 ] and its statistical histogram. Figure 4 is the real-time error E of wind power in 3000 historical periods in this embodiment. h Schematic diagram of statistical data.

[0059] Step 42: Use the Distribution Fitter toolbox in MATLAB to Figure 4 The real-time error E of wind power in 3000 historical periods is shown h The statistical data is fitted to obtain the real-time error E of wind power in the historical period. h Satisfy the mean μ e =19.2, variance σ e =32 normal distribution.

[0060] Figure 5is the real-time error of wind power in the backward period within 5 minutes, E p Schematic diagram of . Figure 6 The wind power at each moment predicted by the Bayesian constant mean dynamic prediction model within 5 minutes Partial real-time wind power Using the above method, the real-time error E of wind power in the backward period within 5 minutes is p , Wind power prediction error at each moment predicted by Bayesian constant mean dynamic prediction model within 5 minutes like Figure 5 , Figure 6 As shown, it can be seen that the wind power prediction error at the moment predicted by the Bayesian constant mean dynamic prediction model is Significantly reduced.

[0061] The wind power prediction error at the current moment in this embodiment is Compensate for the wind power forecast at the current moment Get the current compensated wind power of any wind turbine:

[0062]

[0063] Figure 7 Compensate wind power for 5 minutes at the current moment Partially predicted wind power predicted by radial basis function neural network Partial real-time wind power Schematic diagram of the comparison of the results between Figure 7 It can be seen that the wind power compensation at time t = m + j The corresponding real-time wind power The error between them is generally smaller than that of the radial basis function neural network prediction of wind power. The corresponding real-time wind power The error value between indicates that the compensation wind power obtained by the wind power prediction method of the present invention has a relatively high accuracy.

[0064] In the present invention, the verification indexes of the wind power prediction method include prediction hit rate, average relative error, root mean square error, and maximum absolute error. For the above indicators, in this embodiment, the effect comparison between the prediction results of the radial basis function neural network model and the compensated wind power results is shown in Table 1:

[0065] Table 1 Comparison of the prediction results of the radial basis function neural network model and the compensated wind power results

[0066]

[0067] In Table 1, the prediction hit rate Among them, Gn Indicates whether it hits or not, and: in, Represents the neural network predicting wind power Real-time wind power The maximum error between .

[0068] In Table 1, the average relative error of the prediction results of the radial basis function neural network model is:

[0069]

[0070] The average relative error of the wind power compensation result of the present invention is:

[0071]

[0072] Where C represents the rated installed capacity of any wind turbine in the wind farm; N represents the predicted wind power of the radial basis function neural network model involved in the verification. The number of samples or the compensation wind power P m The number of samples. Prediction of wind power using radial basis function neural network model The sampling number and the compensation wind power P m The number of samples is equal.

[0073] In Table 1, the root mean square error of the prediction results of the radial basis function neural network model is:

[0074]

[0075] The average relative error of the wind power compensation result of the present invention is:

[0076]

[0077] In Table 1, the maximum absolute error of the prediction results of the radial basis function neural network model is:

[0078]

[0079] The maximum absolute error of the wind power compensation result of the present invention is:

[0080]

[0081] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A wind power prediction method, characterized in that: The wind power prediction method comprises the following steps: Step 1: According to the real-time wind speed, real-time wind direction, and real-time atmospheric humidity of the wind farm, obtain the real-time wind power of any wind turbine in the wind farm at any time t. Among them, real-time wind power is a real number, t is a natural number; Step 2: According to the real-time wind speed, real-time wind direction, and real-time atmospheric humidity of the wind farm, the neural network model is used to predict the wind power and obtain the predicted wind power of any wind turbine in the wind farm at any time t. Among them, the predicted wind power is a real number; Step 3: Select the current time m: calculate the real-time wind power corresponding to each historical time within the time length L before the current time m Predicted wind power By comparison, the real-time wind power error E in the historical period can be obtained. h =[e m , e m-1 ,…,e m-i ,…,e m-L ]; where the real-time error of historical wind power at any time within the time length L before the current time m is m, i, and L are all natural numbers, and 0≤i≤L; e m 、e m-1 , …, e m-i ,…,e m-L are all real numbers; Step 4: Calculate the real-time error E of wind power during historical periods h The distribution characteristics of the wind power in the historical period are obtained by h The statistical mean is μ e , the statistical variance is σ e The normal distribution of wind power and the real-time error E of the historical period h Statistical histogram of ; Step 5: Obtain the real-time error E of wind power in the backward period within the time length S after the current time m p =[e m+1 ,…,e m+j ,…,e m+S ]; where the real-time error of the backward wind power at any time within the time length S after the current time m is j and S are both natural numbers, and 1≤j≤S; e m+1 ,…,e m+j ,…,e m+S are all real numbers; Step 6: The real-time error E of wind power in the historical period determined in step 4 h The statistical mean is μ e , the statistical variance is σ e The normal distribution of the wind power in the backward period is used to determine the real-time error E of wind power. p It also follows that the statistical mean is μ e , the statistical variance is σ e Normal distribution of Step 7: The real-time error E of wind power in the backward period determined in step 6 p The distribution characteristics of wind power in the backward period are p , establish a Bayesian constant mean dynamic prediction model for error prediction, and obtain the wind power prediction error at the current moment Step 8: The wind power prediction error at the current moment obtained in step 7 is Compensate for the current predicted wind power obtained in step 2 Get the current compensated wind power of any wind turbine: In step 7, the Bayesian constant mean dynamic prediction model is as follows: The observation equation is: The state equation is: E p =E p-1 +W m Among them, the observation error V m =[v m , v m+1 , …, v m+j , …, v m+S ]、State error W m =[w m , w m+1 ,…,w m+j ,…,w m+S ] have a statistical mean of 0 and a statistical variance of σ e Normal distribution of The real-time error E of wind power in the backward period p The mean value of m at the current time The real-time error E of wind power in the backward period p The mean value at the next moment m+1 The real-time error E of wind power in the backward period p The variance of m at the current time The real-time error E of wind power in the backward period p The variance at the next moment m+1 The observation error v at the current time m m , the state error w of m at the current moment m , the observation error v at the next moment m+1 m+1 、The next moment m+1 state error w m+1 The relationship between them is as follows:

2. A wind power prediction method according to claim 1, characterized in that: In step 4, the statistical mean μ e =19.2, the statistical variance σ e =32.

3. A wind power prediction method according to claim 1, characterized in that: In step 1, the real-time wind direction includes: a real-time wind direction sine value and a real-time wind direction cosine value.

Images