A canyon wind power short-term power prediction method based on grid clustering

Abstract

The present application relates to the technical field of wind power prediction, and discloses a canyon wind power short-term power prediction method based on grid clustering, comprising the following steps: step S1, according to the characteristics of canyon wind power, the short-term power prediction method is divided into a data cleaning link, a grid decomposition link and a power prediction link; step S2, in the data cleaning link, the historical meteorological data, wind power data and wind power prediction data learned by the LSTM network are subjected to anomaly detection, data correction and smoothing processing, and complete historical data are obtained; step S3, in the grid decomposition link, the regional grid clustering of the canyon wind power station is carried out according to the latitude, and the wind power and the characteristic grid associated with the meteorological data are obtained; step S4, in the power prediction link, the power of each grid is predicted by using the LSTM network, and power superposition and error correction are carried out. The present application is used to solve the problem of low accuracy of short-term wind power prediction caused by the influence of canyon wake on canyon wind power.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power prediction, in particular to a canyon wind power short-term power prediction method based on grid clustering, which is used to solve the problem of low accuracy of wind power short-term power prediction caused by the influence of canyon wake on canyon wind power. BACKGROUND

[0002] As an effective clean energy, wind energy belongs to renewable resources and has been widely valued and utilized. A lot of research has been done on canyon wind, especially on canyon wind power short-term power prediction. The research mainly includes error correction type and multi-modal decomposition type wind power short-term power prediction methods.

[0003] In the error correction type wind power short-term power prediction method, the historical prediction data deviation is analyzed to correct the wind power short-term power prediction result. A wind power short-term power prediction method based on core density diffusion is proposed. The error analysis is converted by a Gaussian function to improve the accuracy of wind power short-term power prediction. A wind power short-term power prediction method based on long short-term memory network is proposed. The wind farm data chaos analysis and long short-term memory network historical error correction prediction are used to improve the accuracy of wind power short-term power prediction. A wind power short-term power prediction method based on random forest is proposed. The wind power short-term power prediction error is extracted by a bidirectional gate model, and the wind power short-term power prediction error is corrected by a random forest to improve the accuracy of wind power short-term power prediction. A wind power short-term power prediction method based on robust regression is proposed. The noise is eliminated and the prediction error is robustly regressed to improve the accuracy of wind power short-term power prediction. However, in the above-mentioned research methods, the wind power station is analyzed as a whole, and the power composition of the wind power station cannot be decomposed, so the prediction accuracy is insufficient.

[0004] In the multi-modal decomposition type wind power short-term power prediction method, the data of the wind power station is decomposed according to the time sequence, and then each component is predicted separately, and finally the wind power station short-term power prediction data is formed by superposition. A wind power short-term power prediction method based on singular spectrum analysis is proposed. The power time series decomposition and iterative prediction are used to improve the accuracy of wind power short-term power prediction. First, the power of the wind power station is adaptively decomposed according to the time sequence, and then the component data prediction and reconstruction are performed to form the wind power short-term power prediction result.

[0005] It can be seen that the canyon wind power short-term prediction power prediction method is various, but in the above method, the variable weather information of the canyon is not considered, the variable wind force wake of the canyon, the opposite direction of the adjacent wind turbine of the wind power station, the deviation analysis of the historical prediction data in the error correction type wind power short-term prediction power prediction method, and the correction of the wind power short-term power prediction result. In this method, the wind power station is analyzed as a whole, the power structure of the wind power station cannot be decomposed, and the prediction accuracy is insufficient. In the multi-modal decomposition type wind power short-term prediction power prediction method, the data of the wind power station is decomposed according to the time sequence, each component is predicted separately, and finally the wind power station short-term power prediction data is formed. The method does not consider the variable weather information of the canyon, the variable wind force wake of the canyon, and the opposite direction of the adjacent wind turbine of the wind power station, so that the above method is insufficient to improve the accuracy of the canyon wind power short-term power prediction.

[0006] Therefore, the above method is insufficient to improve the accuracy of the canyon wind power short-term power prediction, in order to solve the problem of low accuracy of the wind power short-term power prediction caused by the influence of the canyon wake on the canyon wind power, the present application provides a canyon wind power short-term power prediction method based on grid clustering. SUMMARY

[0007] The present application aims to provide a canyon wind power short-term power prediction method based on grid clustering, which has the effect of solving the problem of low accuracy of the wind power short-term power prediction caused by the influence of the canyon wake on the canyon wind power.

[0008] The present application is realized by the following technical scheme: a canyon wind power short-term power prediction method based on grid clustering, comprising the following steps:

[0009] Step S1. According to the characteristics of canyon wind power, the short-term power prediction method is divided into data cleaning link, grid decomposition link and power prediction link;

[0010] Step S2. In the data cleaning link, the historical weather, wind power, wind power prediction data learned by the LSTM network are subjected to anomaly detection, data correction and smoothing processing, and complete historical data are obtained;

[0011] Step S3. In the grid decomposition link, the regional grid clustering of the canyon wind power station is carried out according to the latitude, and the wind power and the characteristic grid associated with the weather are obtained;

[0012] Step S4. In the power prediction link, the LSTM network is used to predict the power of each grid, and the power superposition and error correction are carried out.

[0013] In order to better realize the present application, further, step S2 comprises:

[0014] Step S2.1. Abnormal data detection is performed on historical meteorological and wind power prediction data, historical wind power data, meteorological monitoring data and wind turbine geographical location data, and abnormal values are removed through outlier detection.

[0015] Step S2.2. A linear interpolation method is used to build a missing data correction framework to realize missing data repair.

[0016] Step S2.3. The wind power and meteorological observation data are smoothed by Kalman filtering to eliminate the useless components in the wind power data.

[0017] In order to better realize the present application, further, step S3 comprises:

[0018] Step S3.1. First, the wind farm grid is divided according to the geographical longitude and latitude;

[0019] Step S3.2. Then, the grid clustering analysis is performed according to the weather type and wind power curve characteristic index to obtain the grid wind power characteristics.

[0020] Step S3.3. Finally, the grid wind power prediction unit is formed by decomposition.

[0021] In order to better realize the present application, further, step S4 comprises:

[0022] Step S4.1. First, the short-term weather forecast information of the grid is obtained through digital weather forecast;

[0023] Step S4.2. Based on the long short-term memory network, the grid wind power fluctuation trend is analyzed to predict the short-term power of the canyon wind power in the next 10 days;

[0024] Step S4.3. The LSTM is used for short-term power prediction of the canyon wind power, and the least square method is used for error correction.

[0025] Step S4.4. Finally, the short-term power prediction result of the canyon wind power is generated.

[0026] In order to better realize the present application, further, the method of abnormal data detection in step S2.1 comprises:

[0027] According to the analysis of the distance between the local outlier factor and the data, the abnormal data with large difference from the normal attributes of the wind power data is detected.

[0028] In order to better realize the present application, further, the method of missing data correction in step S2.2 comprises:

[0029] The linear interpolation method is used to build a missing data correction framework to realize missing data repair.

[0030] In order to better realize the present application, further, the method of data smoothing processing in step S2.3 comprises:

[0031] The Kalman filter is used to optimally estimate the filtering state, and eliminate the influence of noise and interference data on the model.

[0032] In order to better realize the present application, further, the method of geographical grid division in step S3.1 comprises:

[0033] The wind farm grid is divided according to the geographical longitude and latitude by using the normalized segmentation method.

[0034] In order to better realize the present application, further, the method of grid clustering analysis in step S3.2 comprises:

[0035] The hierarchical agglomerative clustering method is used for grid clustering analysis.

[0036] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0037] (1) The present application proposes a grid clustering analysis method, which obtains the characteristics of the canyon wind power station grid through grid clustering, and obtains the grid characteristics of the canyon wind power station by clustering the grid according to the weather type, wind speed and power generation power index;

[0038] (2) The present application divides the geographical grid according to the topographic features and geographical position information of the wind turbine in the canyon area, and divides the canyon wind power station grid according to the geographical longitude and latitude coordinates, so as to improve the decomposition granularity of the canyon wind power prediction. BRIEF DESCRIPTION OF DRAWINGS

[0039] The present application is further illustrated in combination with the following drawings and examples, and all the innovative ideas of the present application should be regarded as the disclosed content and the protection scope of the present application.

[0040] Figure 1 A flow chart of a canyon wind power short-term power prediction method based on grid clustering provided by the present application.

[0041] Figure 2 An LSTM grid power prediction schematic diagram provided by the present application. DETAILED DESCRIPTION

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. It should be understood that the described embodiments are only a part of the embodiments of the present application, not all the embodiments, and therefore should not be regarded as limiting the protection scope. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0043] In the present application, the canyon wind power is a valley with steep slopes on both sides, and the width is smaller than the depth. In the canyon, along the geological structure fissure construction, the slopes on both sides of the canyon are cliffs, and the valley bottom is flat. When the air enters the canyon mouth from the flat area, the cross-sectional area of the airflow decreases, and the air pressure increases, which increases the speed of the airflow movement, forming a strong wind in the canyon. Therefore, the canyon wind is suitable for building a wind power station, but the canyon wind power station depends on the construction of the canyon terrain, and generally extends for tens of kilometers. The valley bottom topography of the canyon is tortuous, and the wind direction and wind force of the canyon are variable, making it difficult to predict the short-term power of the wind power.

[0044] Grid, in information science, is a mechanism for integrating or sharing various resources (including computer systems, storage systems, communication systems, files, databases, programs, etc.) distributed geographically, so as to become an organic whole, and to complete various required tasks together.

[0045] The wind turbine is a power device that converts wind energy into mechanical work, and the mechanical work drives the rotor to rotate, and finally outputs alternating current. The wind turbine generally has a wind wheel, a generator (including a device), a direction regulator (a tail wing), a tower, a speed limiting safety mechanism and an energy storage device. The working principle of the wind turbine is relatively simple. The wind wheel rotates under the action of the wind force, which converts the kinetic energy of the wind into the mechanical energy of the wind wheel shaft. The generator rotates under the driving of the wind wheel shaft to generate electricity.

[0046] Embodiment 1:

[0047] A canyon wind power short-term power prediction method based on grid clustering in the present embodiment, as shown in Figure 1 In order to solve the problem of low accuracy of short-term power prediction of wind power caused by the influence of canyon wake on canyon wind power, a canyon wind power short-term power prediction method based on grid clustering is proposed in the present embodiment. First, the historical meteorological, wind power and wind power prediction data learned by the LSTM network are subjected to anomaly detection, data correction and smoothing processing. Then, the regional grid clustering of the canyon wind power station is carried out according to the latitude and longitude, and the characteristic grid of the wind power and the meteorological data is obtained. On this basis, the LSTM network is used to predict the power of each grid, and the power superposition and error correction are carried out.

[0048] Embodiment 2:

[0049] This embodiment is further optimized on the basis of embodiment 1. In this embodiment, in the data cleaning link, first, the historical meteorological and wind power prediction data, historical wind power data, meteorological monitoring data and wind turbine geographical position data are detected for abnormal data, and the outliers are removed through the outlier detection method; then, a linear interpolation method is used to build a missing data correction framework to realize missing data repair; finally, Kalman filtering is used to smooth the wind power and meteorological observation data, and eliminate the useless components in the wind power data.

[0050] The other parts of this embodiment are the same as embodiment 1, and will not be described again.

[0051] Embodiment 3:

[0052] This embodiment is further optimized on the basis of embodiment 1 or 2. In the grid decomposition, first, the wind farm grid is divided according to the geographical longitude and latitude, then the grid clustering is carried out according to the weather type, wind power curve characteristics and other indicators, the grid wind power characteristics are obtained, and finally the grid wind power prediction unit is decomposed. The grid wind power characteristic is the characteristic grid related to the weather.

[0053] The other parts of this embodiment are the same as the above-mentioned embodiment 1 or 2, and will not be described again.

[0054] Embodiment 4:

[0055] This embodiment is further optimized on the basis of any one of the above embodiments 1-3. In this embodiment, in the power prediction link, first, the short-term weather forecast information of the grid is obtained through digital weather forecast, then the grid wind power fluctuation trend is analyzed based on the long short-term memory network (LSTM) to predict the short-term power of the canyon wind power in the next 10 days. Again, the wind power of all grids is superimposed to obtain the overall short-term prediction power of the wind power station. The LSTM network is used to correct the error, and finally the short-term power prediction result of the canyon wind power is generated.

[0056] As Figure 2As shown, the numerical weather prediction is the key of the grid power prediction, the numerical weather prediction (NWP) in this paper is from the Spanish analysis data, which has high accuracy, and the prediction of the numerical weather prediction is not repeated in this paper due to the limited space. The LSTM is a neural network that circulates in time, which can solve the problem of insufficient dependence on different time scales in the recurrent neural network (RNN), and the LSTM network has input, output and forgetting gate, which can process the problem of wind power prediction of different time scales of canyon wind power. Therefore, the LSTM is used for short-term power prediction of canyon wind power in this paper. Let th be the current time of the canyon wind power grid prediction, the last time be th-1, Ph-1 and Ph be the distribution of the input grid historical power and the grid predicted power of the LSTM network. The input and output numerical grid weather prediction data are Sh-1 and Sh, Zg, Zh and Zo are the distribution of the forgetting, input and output gates of the LSTM network, the current input state is Pw, the sigmoid function is β, the hyperbolic tangent function is tanh, and the LSTM prediction structure is as shown in Figure 2 On the basis of the short-term power prediction of the canyon wind power grid, all the grid short-term predicted power of the canyon wind power station is superimposed, and the overall power of the canyon wind power station is formed after superposition. In this embodiment, the LSTM network is used for short-term power prediction of canyon wind power, and the effect achieved is the most expected. It should be noted that the LSTM network selected in this embodiment is the most expected method among all methods, and if other methods appear later and achieve better prediction effect, they also fall within the protection scope of the present application.

[0057] In order to reduce the error of the short-term power prediction of the canyon wind power, the method adopts error prediction correction for model updating in this embodiment, the collection point number of the input short-term predicted power in the first 10 days (period) is nx, different wind power data is Pri, the least square method is used for different fitting coefficients of the canyon wind power, Bri, the data of the canyon wind power before short-term power prediction correction is Pk, and the corrected short-term power prediction data Po is:

[0058] .

[0059] It should be noted that the error prediction correction method selected in this embodiment is the most expected method among all methods, and if other methods appear later and achieve better or equivalent correction, they also fall within the protection scope of the present application.

[0060] The other parts of this embodiment are the same as the above-mentioned embodiments 1-3, and will not be repeated here.

[0061] Embodiment 5:

[0062] The embodiment is further optimized on the basis of any one of the above embodiments 1-4. In the embodiment, the collected data of the wind power station includes 96 points of real-time power of a single wind turbine, weather forecast data, wind power prediction data, etc., which are of multiple types and large in quantity. The wind power station monitoring database is prone to be disturbed by transmission noise and prone to have missing and out-of-range data. Therefore, before wind power prediction, the abnormal data of the historical data need to be detected first. The local outlier factor (LOF) is an abnormal value detection method based on distance analysis of wind power and meteorological data. The method can detect abnormal data with large differences from the normal attributes of wind power data through distance analysis between data. The LOF detection method has the characteristics of high speed and high efficiency, and therefore, the method is adopted as the abnormal data detection method in the application.

[0063] Let the s-th distance of the wind power collection data point j be hs(j), the number of the field of point j be Ls(j), and Ls(j) ≥ s, the reachable distance of point o to point j be Re, and the local density lrd s (j) of point j be: ;

[0064] Let the local density reachable of point o be lrds(o), and the outlier factor F s (j) of point j be:

[0065] ;

[0066] Through the LOF method, the abnormal values of 96 points of real-time power of a single wind turbine, weather forecast data, wind power prediction data, etc. can be detected and removed. It should be noted that the LOF method selected in the embodiment is the most suitable method among all methods at present, and if other methods appear in the future and achieve better or equivalent detection and removal effect, they also fall within the protection scope of the application.

[0067] The other parts of the embodiment are the same as any one of the above embodiments 1-4, and therefore, will not be described again.

[0068] Embodiment 6:

[0069] The embodiment is further optimized on the basis of any one of the above embodiments 1-5. In the embodiment, the purpose of missing data correction is to enable the short-term power prediction model of the canyon wind power to accurately learn the historical change rule when analyzing the historical data and reduce the influence of missing data on model training.

[0070] The linear interpolation method is a method for determining unknown power and meteorological data between two points by using two known power and meteorological data in historical wind power and meteorological data. The method has a small correction error for parabolic, linear and other power and meteorological curve data. Therefore, the linear interpolation method is used to build a missing data correction framework to realize missing data repair.

[0071] Suppose that the canyon wind power and meteorological data sequence is k, and the data time period for statistics is n a Suppose that the data before and after the missing point k h are k a and k b Suppose that the interpolation distribution before and after the missing point is e a and e b , the missing point interpolation value is e c , and the missing point k h satisfies:

[0072] .

[0073] The linear interpolation of the above formula can obtain the missing point prediction value of the canyon wind power and meteorological data, and eliminate the influence of missing data on the short-term prediction data of the canyon wind power.

[0074] It should be noted that the linear interpolation method selected in the embodiment is the most expected method among all methods. If other methods appear later and achieve better or equivalent missing data repair effect, they also fall within the protection scope of the present application.

[0075] The other parts of the embodiment are the same as any one of the above embodiments 1-5, and will not be described again.

[0076] Embodiment 7:

[0077] The embodiment is further optimized on the basis of any one of the above embodiments 1-6. In the embodiment, the purpose of data smoothing processing is to solve the problem that the noise data pollutes the real data due to the influence of equipment precision on wind power historical measurement data. Through data smoothing processing, the canyon wind power measurement data can be maximized to approach the real measurement value of the equipment. Kalman filtering is a linear state filtering algorithm. The wind power historical power and meteorological data of the input and output of the Kalman filter are observed, so that the filtering state is best estimated. The method can eliminate the influence of noise and interference data on the model. Therefore, Kalman filtering is selected for data smoothing processing. Suppose that the state transition matrix of the Kalman filter is Ra, the input is converted into the state matrix is Ga, the state matrix of the Kalman filter at the last time is Ba-1, the conversion process noise of the Kalman filter is Oa-1, and the control amount of the system at a-1 time is Ca-1. Then the state matrix of the Kalman filter is:

[0078] ;

[0079] Suppose the state observation matrix of the Kalman filter is Rb, the measurement noise of the Kalman filter is Za, and the noise is subject to Gaussian distribution, then the filter state quantity observation array is:

[0080] ;

[0081] Suppose the prior slant variance of the Kalman filter at the last time is Ea-1, and the noise variance of the filter conversion process is Aa, then the covariance Ea of the filter intermediate output result at the time a is:

[0082] ;

[0083] Suppose the measurement noise covariance is Zb, and the state conversion to the observation value matrix is Wa, then the Kalman gain filter matrix Ua is:

[0084] ;

[0085] Suppose the filter conversion matrix is Ha, and the filter result covariance Ve at the posterior time is:

[0086] ;

[0087] It should be noted that the Kalman filtering method selected in this embodiment is the most expected method among all methods. If other methods appear later and achieve better or equivalent results, they also fall within the protection scope of the present application.

[0088] The other parts of this embodiment are the same as any one of the above embodiments 1-6, and thus will not be described again.

[0089] Embodiment 8:

[0090] This embodiment is further optimized on the basis of any one of the above embodiments 1-7. In this embodiment, the canyon is a mountain valley with steep slopes on both sides, and the width is smaller than the depth. In the canyon, the canyon is constructed along the geological structure fissure, the slopes on both sides of the canyon are cliffs, and the valley bottom is flat. When air enters the canyon from the flat area, the cross-sectional area of the airflow becomes smaller, and the air flow speed is increased under the influence of air pressure, forming a strong wind in the canyon. Therefore, the canyon wind is suitable for building a wind power station, but the canyon wind power station is generally tens of kilometers long according to the terrain of the canyon. The valley bottom topography of the canyon is tortuous, and the wind direction and wind power of the canyon are variable, which causes great difficulty in short-term power prediction of wind power. According to the topographic features and geographical position information of the wind turbine in the canyon area, the canyon wind power station is divided into grids according to the geographical longitude and latitude coordinates, so as to improve the decomposition granularity of the canyon wind power prediction in each grid area.

[0091] Normalized Cuts (N-cut) is a kind of minimum canyon terrain segmentation method based on graph theory, by mapping the terrain map as a weighted canyon undirected graph, that is, the canyon wind power terrain map is divided according to graph theory. Let the terrain map of the whole canyon wind power station be Qall, any two sets of the canyon wind power station terrain be Qa and Qb, the weight sum of the above two sets is cut(Qa, Qb), and the weight sum of the connected edges between the coordinate points in Qa and Qb and the whole canyon wind power station terrain map is as(Qa, Qall) and as(Qb, Qall), then the minimum segmentation of the canyon wind power station is:

[0092] .

[0093] It should be noted that the normalized segmentation method selected in this embodiment is the most expected method among all methods. If other methods appear later and achieve better or equivalent division effect, they also fall within the protection scope of the present application.

[0094] The other parts of this embodiment are the same as any one of the above embodiments 1-7, and will not be described again.

[0095] Embodiment 9:

[0096] This embodiment is further optimized on the basis of any one of the above embodiments 1-8. In this embodiment, the purpose of grid clustering is to obtain the characteristics of the canyon wind power station grid. The grid is clustered by weather type, wind speed and power generation power index to obtain the grid characteristics of the canyon wind power station. Hierarchical Agglomerative Clustering (HAC) is a single-chain clustering method, which clusters a sample data in the canyon wind power grid, and then condenses and iterates two adjacent clustering sets. This method is highly adaptable in the canyon wind power grid, therefore, this method is selected for clustering. When the canyon wind power station grid is clustered by HAC, the Euclidean distance is used to calculate the similarity between two classes. Let the two different classes of objects of clustering be Yai and Ybi, the number of clustering features be nf, and the similarity of the two classes be

[0097] The similarity degree of the canyon wind power grid is calculated by iteration, that is, the classification of the canyon wind power grid can be judged. After clustering, the grid wind power prediction unit is formed.

[0098] It should be noted that the hierarchical agglomerative clustering method selected in this embodiment is the most expected method among all methods. If other methods appear later and achieve better or equivalent clustering effect, they also fall within the protection scope of the present application.

[0099] The other parts of this embodiment are the same as any one of the above-mentioned embodiments 1-8, and thus will not be described again.

[0100] The above is only the preferred embodiment of the present application, and does not limit the present application in any form. Any simple modification or equivalent change made to the above embodiment according to the technical essence of the present application falls within the protection scope of the present application.

Claims

1. A method for short-term power prediction of canyon wind power based on grid clustering, characterized in that, The method comprises the following steps: Step S1. According to the characteristics of canyon wind power, the short-term power prediction method is divided into data cleaning, grid decomposition and power prediction; Step S2. In the data cleaning step, the LSTM network learns the historical meteorological, wind power and wind power prediction data, and performs abnormal detection, data correction and smoothing processing to obtain complete historical data; Step S3. In the grid decomposition step, the geographical grid clustering of the wind power station is carried out according to the latitude, and the wind power and the characteristic grid associated with the meteorological data are obtained; Step S4. In the power prediction step, the LSTM network is used to predict the power of each grid, and the power is superimposed and the error is corrected; The step S3 comprises: step S3.

1. The wind farm grid is divided according to the geographical longitude and latitude; step S3.

2. The grid clustering analysis is carried out according to the weather type and the characteristic index of the wind power curve, and the grid wind power characteristics are obtained; step S3.

3. The grid wind power prediction unit is formed by decomposition; The method of geographical grid division in step S3.1 comprises: selecting a normalized segmentation method, and dividing the wind farm grid according to the geographical longitude and latitude; The normalized segmentation is realized by mapping the terrain map into a weighted canyon undirected graph, i.e. the canyon wind power terrain map is divided according to the graph theory, wherein the terrain map Qall of the whole canyon wind power station all , any two sets Q a and Q b of the canyon wind power station terrain, the weight sum of the two sets is cut(Q a ,Q b ), the weight sum of the connected edges between the coordinate points in Q a and Q b and the terrain map Qall of the whole canyon wind power station is as(Q a ,Qall) and as(Q b ,Q all , and the minimum segmentation of the canyon wind power station is obtained. ； The method of grid clustering analysis in step S3.2 comprises: first, using hierarchical agglomerative clustering method to carry out grid clustering analysis; In the hierarchical agglomerative clustering of the canyon wind power plant grid, the Euclidean distance is used to calculate the similarity between two categories. Let Y ai and Y bi be the objects of two different classes of clustering, and the number of features of clustering be n f , and the similarity of two classes be: By iteratively calculating the similarity of the canyon wind power grid, the classification of the canyon wind power grid can be determined, and after clustering, the grid wind power prediction unit is formed.

2. The canyon wind power short-term power prediction method based on grid clustering according to claim 1, characterized in that, The step S2 comprises: step S2.

1. The LSTM network learns the historical meteorological and wind power prediction data, historical wind power data, meteorological monitoring data and geographical position data of the wind turbine, and the abnormal data detection is carried out through the outlier detection method to eliminate the abnormal values; step S2.

2. The linear interpolation method is used to build a missing data correction framework to realize the missing data repair; step S2.

3. The Kalman filter method is used to smooth the wind power and meteorological observation data, and the useless components in the wind power data are eliminated.

3. The canyon wind power short-term power prediction method based on grid clustering according to claim 1, characterized in that, The step S4 comprises: step S4.

1. The short-term weather forecast information of the grid is obtained through digital weather forecast; step S4.

2. The long-term and short-term memory network is used to analyze the grid wind power fluctuation trend, and the canyon wind power short-term power in the expected time is predicted; step S4.

3. The LSTM network is used for canyon wind power short-term power prediction, and the least square method is used for error correction; step S4.

4. The canyon wind power short-term power prediction result is generated.

4. The canyon wind power short-term power prediction method based on grid clustering according to claim 2, characterized in that, The method of abnormal data detection in step S2.1 comprises: according to the distance analysis between the local outlier factor and the data, the abnormal data with large difference from the normal attributes of the wind power data is detected.

5. The canyon wind power short-term power prediction method based on grid clustering according to claim 2, characterized in that, The method of missing data correction in step S2.2 comprises: the linear interpolation method is used to build a missing data correction framework to realize the missing data repair.

6. The canyon wind power short-term power prediction method based on grid clustering according to claim 2, characterized in that, The method of data smoothing processing in step S2.3 comprises: the Kalman filter is used to optimally estimate the filter state, and the influence of noise and interference data on the model is eliminated.

Images