A wind power prediction method and system based on time-space fusion

Through the spatial and temporal fusion wind power power prediction method, local and global models are used to explore the spatial and temporal correlation in the wind farm, the problem of neglected spatiotemporal correlation in the traditional wind power prediction mode is solved, and the accuracy of wind power power prediction and grid stability are improved. Suitable for large-scale wind power clusters.

CN118589494BInactive Publication Date: 2024-11-19ANHUI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411059986.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The traditional single-field wind power prediction model ignores the complex spatiotemporal correlation of atmospheric systems on wind farm groups in the region, limiting the accuracy of wind power power prediction, especially in large-scale wind power clusters, which affects the stability and control of the power grid. reliability.

Method used

Using the wind power power prediction method based on space-time fusion, by selecting the historical data of wind power power of multiple adjacent sites, a local model is built to mine the temporal correlation of the site itself and the global model is used to mine the temporal correlation between the sites, and to use the decision model to carry out space-time Fusion to generate final prediction results.

Benefits of technology

Improve the accuracy of wind power prediction, enhance the stability and control reliability of power grid operation, is suitable for wind farms of different sizes and geographical locations, optimizes the operating strategies of wind farms, and reduces the power system caused by inaccurate predictions. Adjust costs to improve the reliability and economicality of power supply.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118589494B_ABST
    Figure CN118589494B_ABST
Patent Text Reader

Abstract

The present invention discloses a wind power prediction method and system based on time-space fusion, and builds a local-global intelligent decision-making model to predict wind power; the local model mines the time correlation of the site itself, and the global model mines the time-space correlation between the site and adjacent sites. Furthermore, the decision model performs time-space fusion on the time correlation of the predicted site itself and the time-space correlation between adjacent sites to better characterize the time-space correlation between adjacent sites in the region, which can not only improve the accuracy of wind power prediction, but also help to enhance the stability of power grid operation and control reliability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to a wind power prediction method and system based on time-space fusion. Background Art

[0002] Due to its high efficiency, low cost and abundance, wind power generation has continued to grow in installed capacity around the world, and the distribution scale of wind farms has also evolved from early dispersed and small-scale to centralized and large-scale distribution. Compared with single machines or single wind farms, large-scale wind power clusters can more effectively utilize wind resources and improve power generation quality. However, due to the randomness, intermittency and volatility of wind speed, wind power generation is extremely unstable. With the large-scale grid connection of wind power to the power system, accurate prediction of wind power is an important measure to improve the stability of grid operation and control reliability.

[0003] With the continuous expansion of wind power installed capacity, the construction of large-scale wind farms has become the mainstream. In this case, the accuracy and precision of wind power forecasting are more important. The traditional single-field forecasting model only considers the data information of a single wind farm, ignoring the complex spatiotemporal correlation brought by the inertia of the atmospheric system to the wind farm group in the region, which limits the improvement of forecasting accuracy. The forecasting method based on spatiotemporal fusion can better cope with the forecasting needs of large-scale wind farms.

[0004] Therefore, it is necessary to propose a wind power prediction method based on time-space fusion to improve the accuracy of wind power prediction. Summary of the invention

[0005] The purpose of the present invention is to provide a wind power prediction method and system based on time-space fusion, so as to improve the accuracy of wind power prediction and enhance the operation stability and control reliability of the power grid.

[0006] In order to achieve the above technical objectives, the present invention provides a wind power prediction method based on time-space fusion, comprising:

[0007] Selecting historical wind power data of n adjacent sites within a predetermined area within the same time period;

[0008] Preprocess the historical wind power data of the selected sites;

[0009] Build a local model to mine the time correlation of the prediction site itself and obtain local prediction results;

[0010] Build a global model to mine the spatiotemporal correlation between the predicted site and its adjacent sites to obtain the global prediction results;

[0011] A decision model is built to integrate the temporal correlation of the prediction site itself with the temporal and spatial correlation between adjacent sites to obtain the final prediction result.

[0012] Optionally, the predetermined area is within 80-120 km of the prediction site, and the historical wind power data of n adjacent sites in the same time period are: X=[X1,X2,X3,...,X n ];

[0013] Where X is the set of historical wind power data of all sites, and the data length of each site is p; X1 represents the historical wind power data of site 1 within a given time period; X2 represents the historical wind power data of site 2 within a given time period; and so on, X n Represents the historical wind power data of site n within a given time period.

[0014] Optionally, the preprocessing includes data cleaning, and the data cleaning includes filling missing values, deleting duplicate data or abnormal data processing;

[0015] The missing value filling includes mean filling, median filling, mode filling, interpolation and filling based on regression model; the abnormal data processing includes deletion, mean replacement, median replacement, mode replacement, binning processing and identification and replacement based on deep learning model.

[0016] Optionally, the process of building the local model includes:

[0017] The data X of site i will be predicted i Normalize to between 0 and 1 and divide the training set X i-train and the test set X i-test ;

[0018] When selecting the deep neural network model, the time window length is set to a when predicting site i, and the training set input data and actual data are divided;

[0019] The input local_X is:

[0020] ;

[0021] Among them, the time window shape is (a, 1, p / 2), a represents the time window size, 1 is the number of input features, p / 2 is the input data length, and the actual value y is: .

[0022] Optionally, use data from 1 to a of site i to predict the (a+1)th data, data from 2 to (a+1)th data to predict the (a+2)th data, data from j to (j+(a-1))th data to predict the (j+a)th data, and so on, and set the training batch and training cycle to train the local model;

[0023] Use the trained model to train the test set data X i-test Make a prediction and get the prediction result y local .

[0024] Optionally, the process of building the global model includes:

[0025] Analyze the lead / delay relationship between sites and keep the data X of the predicted site i i The rest of the site data are moved within ±T steps, and the Pearson correlation coefficient between each site and site i after the movement is calculated. The site data ahead / behind when the correlation coefficient is the largest are selected as input features.

[0026] Optionally, normalize all site data to between 0 and 1 and divide the training set X train and the test set X test ;

[0027] When using the deep neural network model to predict site i, the time window length is set to b, dividing the training set input data and actual data;

[0028] Among them, the input global_X is:

[0029] ;

[0030] Among them, the time window shape is (b, n, p / 2), b represents the time window size, n is the number of input features, p / 2 is the input data length, and the actual value y is: .

[0031] Optionally, use 1 to b data of all sites to predict the (b+1)th data of site i, 2 to (b+1)th data to predict the (b+2)th data, j to (j+(b-1))th data to predict the (j+b)th data, and so on, setting the training batch and training cycle to train the global model;

[0032] Use the trained model to train the test set data X test Make a prediction and get the prediction result y global .

[0033] Optionally, a decision model is built to perform spatiotemporal fusion of the temporal correlation of the prediction site itself and the spatiotemporal correlation between adjacent sites;

[0034] A two-layer decision tree model is used to make intelligent decisions on local and global prediction results, and the final prediction result Y of the target site is obtained after denormalizing the decision results.

[0035] The present invention also provides a wind power prediction system based on time-space fusion, which is used to implement the above-mentioned wind power prediction method based on time-space fusion, including: a data processing module, a model building module and a final result prediction module;

[0036] The data processing module is used to perform preprocessing based on historical wind power data within a predetermined area;

[0037] The model building module is used to build a local model and obtain a local prediction result and to build a global model and obtain a global prediction result;

[0038] The final result prediction module is used to build a decision model to perform spatiotemporal fusion of the temporal correlation of the prediction site itself and the spatiotemporal correlation between adjacent sites to obtain the final prediction result.

[0039] Compared with the prior art, the present invention has at least the following beneficial effects: the present invention builds a local-global intelligent decision-making model, mines the time correlation of the prediction site itself and the time-space correlation between adjacent sites in the region and performs time-space fusion, thereby further improving the accuracy of wind power prediction, improving the operation stability and control reliability of the power grid. The prediction method provided by the present invention has important practical significance and application value for the operation management of wind farms and the optimal scheduling of power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the steps of a wind power prediction method based on time-space fusion in an embodiment of the present invention;

[0041] Figure 2 It is a structural diagram of the local-global intelligent decision-making model in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] A wind power prediction method and system based on time-space fusion of the present invention will be described in more detail below in conjunction with the accompanying drawings, wherein preferred embodiments of the present invention are shown, and it should be understood that those skilled in the art can modify the present invention described herein while still achieving the beneficial effects of the present invention. Therefore, the following description should be understood as being widely known to those skilled in the art, and not as a limitation of the present invention.

[0043] The present invention is described in more detail in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the present invention will become more apparent from the following description. It should be noted that the accompanying drawings are in very simplified form and are not in exact proportions, and are only used to facilitate and clearly assist in illustrating the purpose of the embodiments of the present invention.

[0044] The present invention provides a wind power prediction method based on time-space fusion, comprising the following steps:

[0045] S1. Selecting historical wind power data of n adjacent sites within a predetermined area within the same time period;

[0046] S2. Preprocessing the historical wind power data of the selected site;

[0047] S3, build a local model to mine the time correlation of the prediction site itself and obtain the local prediction result;

[0048] S4, build a global model to mine the spatiotemporal correlation between the predicted site and its adjacent sites to obtain the global prediction results;

[0049] S5. Build a decision model to integrate the temporal correlation of the prediction site itself with the temporal and spatial correlation between adjacent sites to obtain the final prediction result.

[0050] In this embodiment, local and global information are fully utilized to improve the accuracy of the prediction results. Among them, the local model can mine the time correlation of a single site, while the global model takes into account the time and space correlation between multiple sites, which helps to predict wind power more accurately; because time and space fusion takes into account more site information, the prediction results are more stable, reducing the uncertainty caused by the fluctuation of single site data; this method can adapt to the wind power prediction needs of wind farms of different sizes and different geographical locations, and has strong versatility and flexibility; by accurately predicting wind power, the operation strategy of the wind farm can be optimized, and the adjustment cost of the power system caused by inaccurate power prediction can be reduced, thereby saving operating costs; in addition, accurate prediction results can provide decision support for the dispatch and operation of the power system, help optimize the operating efficiency of the power system, and improve the reliability and economy of power supply.

[0051] Specifically, in step S1, the predetermined area is within 80-120km centered on the prediction site i, and the historical wind power data of n adjacent sites in the same time period are: X=[X1,X2,X3....X n ].

[0052] Where X is the set of historical wind power data of all sites, and the data length of each site is p; X1 represents the historical wind power data of site 1 within a given time period; X2 represents the historical wind power data of site 2 within a given time period; and so on, X n Represents the historical wind power data of site n within a given time period.

[0053] Preferably, in this embodiment, the regional scope selects the historical wind power data of n adjacent sites within 100 km centered on the prediction site i in the same time period.

[0054] Where X is the set of historical wind power data of all sites, and the data length of each site is p; X1 represents the historical wind power data of site 1 within a given time period; X2 represents the historical wind power data of site 2 within a given time period; and so on, X n Represents the historical wind power data of site n within a given time period.

[0055] Furthermore, in step S2, the data of the selected site is cleaned and other preprocessing steps are performed, including filling missing values, deleting duplicate data, processing abnormal data, etc. The purpose of preprocessing the data is to ensure data quality and improve the effectiveness of model training and the accuracy of prediction.

[0056] Among them, the missing value filling includes mean filling, median filling, mode filling, interpolation and filling based on regression model; the abnormal data processing includes deletion, mean replacement, median replacement, mode replacement, binning processing and identification and replacement based on deep learning model.

[0057] In step S3, the process of building a local model includes:

[0058] The data X of site i will be predicted i Normalize to between 0 and 1 and divide the training set X i-train and the test set X i-test , preferably, X i The first 50% is used as the training set, and the last 50% is used as the test set.

[0059] When using the deep neural network model to predict site i, the time window length is set to a, and the training set input data and actual data are divided; among them, the input local_X is:

[0060] ;

[0061] Among them, the time window shape is (a, 1, p / 2), a represents the time window size, 1 is the number of input features, p / 2 is the input data length, and the actual value y is: .

[0062] Furthermore, use 1 to a data of site i to predict the (a+1)th data, 2 to (a+1)th data to predict the (a+2)th data, j to (j+(a-1))th data to predict the (j+a)th data, and so on. Set the training batch and training cycle to train the local model.

[0063] Use the trained model to train the test set data X i-test Make a prediction and get the prediction result y local .

[0064] For a specific example, see Figure 2 ,The local model uses a long short-term memory network (LSTM).,When predicting site i, the time window length is set to 50. In other examples, the time window length can be adjusted according to the ,prediction results, dividing the training set input data and the actual data.

[0065] The input local_X is:

[0066] ;

[0067] The time window shape is (50, 1, p / 2), which means the time window size is 50, the number of input features is 1, and the input data length is p / 2; the actual value y is:

[0068] .

[0069] Furthermore, use the 1st to 50th data of site i to predict the 51st data, the 2nd to 51st data to predict the 52nd data, the jth to (j+49)th data to predict the (j+50)th data, and so on. The training batch and training cycle are set to train the local model.

[0070] Use the trained local model to train the test set data X i-test Make a prediction and get the prediction result y local .

[0071] Furthermore, in step S4, the process of building a global model includes:

[0072] In this embodiment, the time lag cross-correlation function is used to analyze the lead / delay relationship between sites, and the data X of the predicted site i is kept. i The rest of the site data are moved within ±T steps, and the Pearson correlation coefficient between each site and site i after the movement is calculated. The site data ahead / behind when the correlation coefficient is the largest are selected as input features.

[0073] In this embodiment, the data X of the prediction site i is maintained. i The data of other stations remain unchanged, and the data of other stations move within the range of ±10 steps. The step of the moving range can be adjusted according to the accuracy of the data set. If the data set is taken once every 1 minute, the time of plus or minus 10 steps is 20 minutes, and the step of the value is slightly smaller. The 10 steps in this embodiment are based on taking data every 0.5 hours as an example.

[0074] Furthermore, all site data are normalized to between 0 and 1, and the training set X is divided into train and the test set X test .

[0075] Furthermore, a deep neural network model is selected, and the time window length is set to b to divide the training set input data and actual data.

[0076] Enter global_X as:

[0077] ;

[0078] Among them, the time window shape is (b, n, p / 2), b represents the time window size, n is the number of input features, p / 2 is the input data length, and the actual value y is:

[0079] .

[0080] Furthermore, use 1 to b data from all sites to predict the (b+1)th data of site i, 2 to (b+1)th data to predict the (b+2)th data, j to (j+(b-1))th data to predict the (j+b)th data, and so on. Set the training batches and training cycles to train the global model.

[0081] Use the trained model to train the test set data X test Make a prediction and get the prediction result y global .

[0082] For a specific example, please refer to Figure 2 ,The global model also uses LSTM. When predicting site i, the time window length is set to 50. The time window length can be adjusted according to the ,prediction results, dividing the training set input data and the ,actual data.

[0083] Among them, the input global_X is:

[0084] ;

[0085] Among them, the time window shape is (50, n, p / 2), which respectively means that the time window size is 50, the number of input features is n, and the input data length is p / 2; the actual value y is: .

[0086] In this embodiment, data 1 to 50 of all sites are used to predict the 51st data of site i, data 2 to 51 are used to predict the 52nd data, data j to (j+49) are used to predict the (j+50)th data, and so on. The training batches and training cycles are set to train the global model.

[0087] Use the trained model to train the test set data X test Make a prediction and get the prediction result y global .

[0088] In step S5, a decision model is built to perform spatiotemporal fusion of the temporal correlation of the prediction site itself and the spatiotemporal correlation between adjacent sites; a two-layer decision tree model is used to predict the temporal correlation of the site itself and the spatiotemporal correlation between adjacent sites. local and global Make intelligent decisions and denormalize the decision results to obtain the final prediction result Y of the target site.

[0089] The present invention also provides a wind power prediction system based on time-space fusion, which is used to implement the above-mentioned wind power prediction method based on time-space fusion, including: a data processing module, a model building module and a final result prediction module;

[0090] The data processing module is used to perform preprocessing based on historical wind power data within a predetermined area;

[0091] The model building module is used to build a local model and obtain a local prediction result and to build a global model and obtain a global prediction result;

[0092] The final result prediction module is used to build a decision model to perform spatiotemporal fusion of the temporal correlation of the prediction site itself and the spatiotemporal correlation between adjacent sites to obtain the final prediction result.

[0093] The present invention builds a local-global intelligent decision-making model to mine the spatiotemporal correlation between adjacent sites and perform spatiotemporal fusion. The local model mines the temporal correlation of the site itself, and the global model mines the spatiotemporal correlation between the site and adjacent sites. Furthermore, the decision model performs spatiotemporal fusion of the temporal correlation of the predicted site itself and the spatiotemporal correlation between adjacent sites to better characterize the spatiotemporal correlation between adjacent sites in the region. This not only improves the accuracy of wind power prediction, but also helps to reduce operation and maintenance costs, improve decision-making efficiency and improve data utilization efficiency. It has important practical significance for promoting the development and consumption of wind power energy.

[0094] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A wind power prediction method based on time-space fusion, characterized in that: include: Selecting historical wind power data of n adjacent sites within a predetermined area within the same time period; Preprocess the historical wind power data of the selected sites; Build a local model to mine the time correlation of the prediction site itself and obtain local prediction results; Build a global model to mine the spatiotemporal correlation between the predicted site and its adjacent sites to obtain the global prediction results; A decision model is built to integrate the temporal correlation of the prediction site itself with the temporal correlation between adjacent sites. A two-layer decision tree model is used to make intelligent decisions on local and global prediction results. The final prediction results of the target site are obtained after denormalizing the decision results. The process of building the local model includes: The data X of site i will be predicted i Normalize to between 0 and 1 and divide the training set X i-train and the test set X i-test ; When selecting the deep neural network model, the time window length is set to a when predicting site i, and the training set input data and actual data are divided; The process of building the global model includes: Analyze the lead / delay relationship between sites and keep the data X of the predicted site i i The rest of the site data are moved within ±T steps, and the Pearson correlation coefficient between each site and site i after the movement is calculated. The site data ahead / behind when the correlation coefficient is the largest are selected as input features.

2. The wind power prediction method based on time-space fusion according to claim 1, characterized in that: The predetermined area is within 80-120km of the prediction site, and the historical wind power data of n adjacent sites in the same time period are: X=[X1,X2,X3,...,X n ]; Where X is the set of historical wind power data of all sites, and the data length of each site is p; X1 represents the historical wind power data of site 1 within a given time period; X2 represents the historical wind power data of site 2 within a given time period; and so on, X n Represents the historical wind power data of site n within a given time period.

3. The wind power prediction method based on time-space fusion according to claim 1, characterized in that: The preprocessing includes data cleaning, and the data cleaning includes filling missing values, deleting duplicate data or abnormal data processing; The missing value filling includes mean filling, median filling, mode filling, interpolation and filling based on regression model; the abnormal data processing includes deletion, mean replacement, median replacement, mode replacement, binning processing and identification and replacement based on deep learning model.

4. The wind power prediction method based on time-space fusion according to claim 1, characterized in that: Input local_X is: ; Among them, the time window shape is (a, 1, p / 2), a represents the time window size, 1 is the number of input features, p / 2 is the input data length, and the actual value y is: ; Among them, x i_1 Indicates the first data of site i, x i_2 Indicates the second data of site i, x i_a represents the ath data of site i; and so on, x i_ j represents the j-th data of site i, x i_ j+1 represents the j+1th data of site i, x i_ j+(a-1) Indicates the j+(a-1)th data of site i; represents the first data, represents the first data, represents the first individual data; x i_1 ~x i_a Construct the first time window, for the a+1th data x of site i i_(a+1) Make a prediction, x i_2 ~x i_(a+1) Construct the second time window, for the a+2th data x of site i i_(a+2) Make a prediction; and so on, x i_ j ~x i_ j+(a-1) Construct the jth time window, for the j+ath data x of site i i_( j+a) Make predictions; ~ Composition time window, for site i Data Make predictions.

5. The wind power prediction method based on time-space fusion according to claim 4, characterized in that: Use data from site i from 1 to a to predict the (a+1)th data, data from 2 to (a+1) to predict the (a+2)th data, data from j to (j+(a-1)) to predict the (j+a)th data, and so on. Set the training batch and training cycle to train the local model. Use the trained model to train the test set data X i-test Make a prediction and get the prediction result y local .

6. The wind power prediction method based on time-space fusion according to claim 5, characterized in that: Normalize all site data to between 0 and 1 and divide the training set X train and the test set X test ; When using the deep neural network model to predict site i, the time window length is set to b, dividing the training set input data and actual data; Among them, the input global_X is: ; Among them, the time window shape is (b, n, p / 2), b represents the time window size, n is the number of input features, p / 2 is the input data length, and the actual value y is: ; Among them, x1 contains the first data of all sites, x2 contains the second data of all sites, and x b Contains the bth data of all sites; and so on, x j Contains the jth data of all sites, x j+1 Contains the j+1th data of all sites, x j+(b-1) Contains the j+(b-1)th data of all sites; Contains all sites data, Contains all sites data, Contains all sites individual data; x1~x b Construct the first time window, for the b+1th data x of site i i_(b+1) Make predictions, x2~x b+1 Construct the second time window, for the b+2th data x of site i i_(b+2) Make a prediction; and so on, x j ~x j+(b-1) Construct the jth time window, for the j+bth data x of site i i_(j+b) Make predictions; ~ Composition time window, for site i Data Make predictions.

7. The wind power prediction method based on time-space fusion according to claim 6, characterized in that: Use 1 to b data of all sites to predict the (b+1)th data of site i, 2 to (b+1)th data to predict the (b+2)th data, j to (j+(b-1))th data to predict the (j+b)th data, and so on. Set the training batch and training cycle to train the global model. Use the trained model to train the test set data X test Make a prediction and get the prediction result y global .

8. A wind power prediction system based on time-space fusion, used to implement the wind power prediction method based on time-space fusion as claimed in any one of claims 1 to 7, characterized in that: include: Data processing module, model building module and final result prediction module; The data processing module is used to perform preprocessing based on historical wind power data within a predetermined area; The model building module is used to build a local model and obtain a local prediction result and to build a global model and obtain a global prediction result; The final result prediction module is used to build a decision model to perform spatiotemporal fusion of the temporal correlation of the prediction site itself and the spatiotemporal correlation between adjacent sites to obtain the final prediction result.