A method, system, device and medium for predicting ultra-short-term wind power generation

Abstract

The present invention discloses a method, system, device and medium for predicting ultra-short-term wind power generation, which relates to the field of wind power technology. The method includes: obtaining meteorological data corresponding to ultra-short-term wind power generation and the original time series M of power generation; selecting the power at the current time point and the time point before the first preset time period of the current time point, and calculating the difference to obtain the time window; reorganizing the original time series M to obtain a new time series N, sampling once every second preset time period, integrating adjacent sampling points into sampling data, and drawing a wind power generation power diagram through the sampling data; calculating financial factors KDJ and RSI, and using the LightGBM algorithm to build a prediction model; inputting the data group consisting of meteorological data, new time series N, financial factors KDJ and RSI into the prediction model to obtain the prediction result of wind power generation. The present invention reduces the dependence of wind power prediction on high-precision meteorological data, and can quickly and accurately predict wind power generation.

Description

Technical Field

[0001] The present invention relates to the field of wind power technology, and in particular to a method, system, equipment and medium for predicting ultra-short-term wind power generation. Background Art

[0002] Wind power generation has become the fastest growing renewable energy source in recent years. In the future, wind power will account for more than 12% of the world's electricity supply. As the proportion of wind power in the power system continues to increase, the volatility and intermittency of wind power have an increasingly significant impact on the power grid. Accurately predicting the power of wind turbines can help dispatching agencies better manage and dispatch wind farms, reduce the impact of wind power on the power grid, and improve the stability of the power grid. By predicting the power of wind turbines, it can help wind power companies better participate in market competition and improve the economic benefits and market competitiveness of wind power.

[0003] In the prior art, the power of wind turbines is usually predicted using meteorological data and geographic data as inputs to the prediction model. Specifically, the method includes: applying the theory of atmospheric boundary layer dynamics and boundary layer meteorology, refining the numerical weather prediction (NWP) data into the wind speed and wind direction at the hub height of the wind turbine under the actual terrain and topographic conditions of the wind farm, and then applying the predicted wind speed to the power curve of the wind turbine after considering the influence of the wake, thereby obtaining the predicted power of the wind turbine. Finally, the predicted power of all wind turbines is summed to obtain the predicted power of the entire wind farm.

[0004] The defects of the above-mentioned prior art are: due to the instability of wind, even changes in wind speed and wind direction in a short period of time may cause large fluctuations in wind power, making the data required for wind power prediction more accurate and difficult to indicate the time series data characteristics in the prediction of wind turbine power generation strategy. Summary of the invention

[0005] Based on this, it is necessary to provide a method, system, equipment and medium for predicting ultra-short-term wind power generation in response to the above technical problems.

[0006] An embodiment of the present invention provides a method for predicting ultra-short-term wind power generation power, comprising:

[0007] Obtain the original time series M of meteorological data and power generation corresponding to ultra-short-term wind power generation;

[0008] Select the power of the current time point and the power of the time point before the current time point by the first preset time period in the original time series M, calculate the difference between the two powers to obtain the time window; reorganize the original time series M through the time window to obtain the new time series N;

[0009] Sampling is performed once every second preset time period in the new time series N, adjacent sampling points are integrated into sampling data, and a wind power generation diagram is drawn through the sampling data;

[0010] The financial factors KDJ and RSI are calculated based on the wind power diagram, and the LightGBM algorithm is used to build a forecasting model for extended financial factors;

[0011] The data set consisting of meteorological data, new time series N, financial factors KDJ and RSI is input into the prediction model to obtain the prediction result of wind power generation.

[0012] In addition, the reorganization of the original time series M to obtain the new time series N specifically includes:

[0013] For time series data M, let the data value at time m be t m+1 is the data value at the end of sampling, and the data value at time u+1 is t u+1 is the data value to be predicted, and the time series data value t at time u is taken u , time series data value at time u-1 t u-1 , the data difference between time u and time u-1▽(t u-1 -t u ) constitutes a time window;

[0014] Combine the time window with the remaining covariates q at time u u 、w u 、e u Transform to form new time series data N. The transformation algorithm is:

[0015]

[0016] Among them, t u is the time series data value at time u, t m+1 is the data value at time m+1, t u+1 is the predicted data value.

[0017] In addition, the drawing of the wind power generation diagram by sampling data specifically includes:

[0018] The maximum value in the sampled data is regarded as the highest price in the financial market, and the minimum value is regarded as the lowest price in the financial market; the power output value of the first sampling point is regarded as the opening price in the financial market, and the power output value of the last sampling point is regarded as the closing price in the financial market;

[0019] A plane rectangular coordinate system is established with time as the horizontal coordinate and wind power generation power as the vertical coordinate;

[0020] In the plane rectangular coordinate system, if the closing price is higher than the opening price, it means that the opening price is below and the closing price is above, and a black rectangular column is drawn; if the closing price is lower than the opening price, it means that the opening price is above and the closing price is below, and a gray rectangular column is drawn to obtain the wind power generation power diagram.

[0021] In addition, the formula for calculating the financial factor KDJ of wind power generation is:

[0022]

[0023] J u =2K u -D u

[0024] Among them, RSV is the immature random value, G 3 The power at the end of 3 hours, E 3 is the maximum power within 3 hours, F 3 is the minimum power within 3 hours, K u is the K value at the uth hour, D u is the D value at the uth hour, J u is the J value at hour u, RSV u is the RSV value at hour u, K u-1 is the K value of the first three hours, D u-1 It is the D value of the first three hours.

[0025] In addition, the financial factor RSI for calculating wind power generation power specifically includes:

[0026] Let the magnitude of the upward power fluctuation within 6 hours be A, let the magnitude of the downward power fluctuation within 6 hours be B, and the power value at the end of x hours is C x ; If C x -C x-1 > 0, then A+=C x -C x-1 ; If C x -C x-1 <0, then B+=C x-1 -C x ; Continuously calculate the values ​​of 6 sampling points, the calculation formula is as follows:

[0027]

[0028] Among them, RS is relative strength, A is the amplitude of upward power fluctuation within 6 hours, B is the amplitude of downward power fluctuation within 6 hours, and RSI1X is the value of 6 consecutive sampling points.

[0029] In addition, a system for predicting ultra-short-term wind power generation comprises:

[0030] A data acquisition module is used to obtain the meteorological data corresponding to ultra-short-term wind power generation and the original time series M of the generated power;

[0031] A time window module is used to select the power of the current time point and the power of the time point before the current time point by a preset time period in the original time series M, calculate the difference between the two powers to obtain the time window; and reorganize the original time series M through the time window to obtain a new time series N;

[0032] A drawing module, used for sampling once every second preset time period in the new time series N, integrating adjacent sampling points into sampling data, and drawing a wind power generation diagram through the sampling data;

[0033] Model building module, used to calculate the financial factors KDJ and RSI based on the wind power diagram, and use the LightGBM algorithm to build a forecasting model for extended financial factors;

[0034] The prediction module is used to input the data group consisting of meteorological data, new time series N, financial factors KDJ and RSI into the prediction model to obtain the prediction result of wind power generation.

[0035] In addition, a computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned ultra-short-term wind power generation prediction method when executing the computer program.

[0036] In addition, a storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the above-mentioned ultra-short-term wind power generation prediction method are implemented.

[0037] Compared with the prior art, the above-mentioned ultra-short-term wind power prediction method, system, device and medium provided by the embodiments of the present invention have the following beneficial effects:

[0038] The present invention obtains the meteorological data and the original time series M of the power generation corresponding to the ultra-short-term wind power generation; selects the power at the current time point and the time point before the first preset time period of the current time point in the original time series M, and calculates the difference between the two powers to obtain the time window; reorganizes the original time series M through the time window to obtain a new time series N; samples are taken once every second preset time period in the new time series N, and adjacent sampling points are integrated into sampling data, and a wind power generation power diagram is drawn through the sampling data; the financial factors KDJ and RSI are calculated according to the wind power generation power diagram, and a prediction model for extending the financial factors is constructed using the LightGBM algorithm; the data group consisting of the meteorological data, the new time series N, the financial factors KDJ and RSI is input into the prediction model to obtain the prediction result of the wind power generation power.

[0039] The above technical solution uses a data set consisting of meteorological data, a new time series N, financial factors KDJ and RSI to replace high-precision meteorological data and geographic data as the input of the prediction model, which solves the problem that existing variables are difficult to indicate the characteristics of time series data, reduces the dependence of wind power prediction on high-precision meteorological data, and can quickly and accurately predict wind power generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic flow chart of a method for predicting ultra-short-term wind power generation provided in an embodiment;

[0041] Figure 2 A wind power generation power diagram of Example 1 of a method for predicting ultra-short-term wind power generation power provided in an embodiment;

[0042] Figure 3 A comparison diagram of actual values ​​and predicted values ​​of Example 1 of a method for predicting ultra-short-term wind power generation provided in an embodiment;

[0043] Figure 4 A comparison diagram of actual values ​​and predicted values ​​of Example 2 of a method for predicting ultra-short-term wind power generation provided in an embodiment;

[0044] Figure 5 A comparison diagram of actual values ​​and predicted values ​​of Example 3 of a method for predicting ultra-short-term wind power generation provided in an embodiment;

[0045] Figure 6 This is a diagram showing the statistical results of the error range of Example 3 of a method for predicting ultra-short-term wind power generation provided in an embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] In one embodiment, a method for predicting ultra-short-term wind power generation is provided, such as Figure 1 As shown, the method includes:

[0048] Step 1: Obtain meteorological data and original time series M from the wind farm or data platform.

[0049] Step 2: In the original time series M, select the power at the current time point and the power 15 minutes before the current time point, and then calculate the difference between the two powers. This difference is the required time window. The method of constructing the time window is as follows:

[0050] For the original time series M, let the data value at time m be t m+1 is the data value at the end of sampling, and the data value at time u+1 is t u+1 is the data value to be predicted, and the time series data value t at time u is taken u , time series data value at time u-1 t u-1 , the data difference between time u and time u-1 Construct a time window. Combine the time window with the remaining covariates q at time u u 、w u 、e u To form a new time series data N. The conversion algorithm is as follows:

[0051]

[0052] Among them, t u is the time series data value at time u, t m+1 is the data value at time m+1, t u+1 is the predicted data value.

[0053] Step 3: Integrate the five adjacent sampling points into sampling data, where the maximum value is regarded as the highest price in the financial market and the minimum value is regarded as the lowest price in the financial market. The power output value of the first sampling point is regarded as the opening price in the financial market, and the power output value of the last sampling point is regarded as the closing price in the financial market. Draw a wind power diagram based on these data, as shown below:

[0054] The first step is to integrate five adjacent sampling points into one piece of data at a frequency of sampling every fifteen minutes. A plane rectangular coordinate system is established with time as the horizontal axis and wind power as the vertical axis.

[0055] In the second step, the maximum value of the power output values ​​of the five sampling points is regarded as the highest price in the financial market, and the minimum value is regarded as the lowest price. The power output value of the first sampling point is used as the opening price of the financial market, and the power output value of the last sampling point represents the closing price. In the plane rectangular coordinate system, if the closing price is higher than the opening price, it means that the opening price is below and the closing price is above. At this time, we use black rectangles for drawing. On the contrary, if the closing price is lower than the opening price, it means that the opening price is above and the closing price is below. At this time, we use gray rectangles for drawing. The drawing result is as follows: Figure 2 shown.

[0056] Step 4, based on the drawn wind power diagram, calculate the financial factors KDJ and RSI of wind power expansion.

[0057] A scheme called KDJ method is used to calculate wind power. This method combines the concept of moving average with the concept of momentum and strength index to form a more accurate basis for power increase and decrease. The design focus of KDJFP indicator is to study the relationship between the highest value, lowest value and sampling end value of power output, and integrates some advantages of moving average. Therefore, this method can quickly and intuitively judge the trend of wind power generation.

[0058] KDJ wind power generation is based on ultra-short-term forecasts, with a forecast period of 3 hours. Assume that the power at the end of 3 hours is G 3 ; The maximum power within 3 hours is E 3 ; The minimum power within 3 hours is F 3 ; The immature random value is RSV; The K value at the uth hour is K u ; The D value at the uth hour is D u ; The J value at the uth hour is J u ; The RSV value at hour u is RSV u ; Ku-1 is the K value of the previous three hours. If it does not exist, use 40 instead; D u-1 The D value of the previous three hours. If it does not exist, use 40 instead. At the same time, the value range of K and D is 0 to 100. The calculation method is as follows:

[0059]

[0060] J u =2K u -D u

[0061] Among them, RSV is the immature random value, G 3 The power at the end of 3 hours, E 3 is the maximum power within 3 hours, F 3 is the minimum power within 3 hours, K u is the K value at the uth hour, D u is the D value at the uth hour, J u is the J value at hour u, RSV u is the RSV value at hour u, K u-1 is the K value of the first three hours, D u-1 It is the D value of the first three hours.

[0062] The RSI method is proposed for wind power forecasting, which further extracts more valuable data and enriches the wind power time series data set, providing a large number of covariates that can replace meteorological data. The addition of RSIFWP greatly improves the forecast accuracy.

[0063] RSI wind power generation is based on ultra-short-term forecasts and is calculated as follows:

[0064] Let the magnitude of the upward power fluctuation within 6 hours be A, let the magnitude of the downward power fluctuation within 6 hours be B, and the power value at the end of x hours is C x If C x -C x-1 > 0, then A+=C x -C x-1 If C x -C x-1 <0, then B+=C x-1 -C x . Calculate the values ​​of 6 sampling points continuously.

[0065]

[0066] Among them, RS is relative strength, A is the amplitude of upward power fluctuation within 6 hours, B is the amplitude of downward power fluctuation within 6 hours, and RSI1X is the value of 6 consecutive sampling points.

[0067] Normally, you need to calculate the values ​​of RSI1X, RSI2X, and RSI3X. RSI1X represents the value of 6 consecutive sampling points, RSI2X represents the value of 12 consecutive sampling points, and RSI3X represents the value of 24 consecutive sampling points.

[0068] Step 5, use the LightGBM algorithm to build a prediction model for extending financial factors, input the data group consisting of meteorological data, new time series N, financial factors KDJ and RSI into the prediction model, and obtain the prediction results of wind power generation.

[0069] In one embodiment, a system for predicting ultra-short-term wind power generation is provided, the system comprising:

[0070] A data acquisition module is used to obtain the meteorological data corresponding to ultra-short-term wind power generation and the original time series M of the generated power;

[0071] A time window module is used to select the power of the current time point and the power of the time point before the current time point by a preset time period in the original time series M, calculate the difference between the two powers to obtain the time window; and reorganize the original time series M through the time window to obtain a new time series N;

[0072] A drawing module, used for sampling once every second preset time period in the new time series N, integrating adjacent sampling points into sampling data, and drawing a wind power generation diagram through the sampling data;

[0073] Model building module, used to calculate the financial factors KDJ and RSI based on the wind power diagram, and use the LightGBM algorithm to build a forecasting model for extended financial factors;

[0074] The prediction module is used to input the data group consisting of meteorological data, new time series N, financial factors KDJ and RSI into the prediction model to obtain the prediction result of wind power generation.

[0075] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned ultra-short-term wind power generation prediction method when executing the computer program.

[0076] A storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned ultra-short-term wind power generation prediction method.

[0077] In a specific embodiment of the present invention, the hardware is a computer, and the configuration includes the hardware environment: processor: Intel Core i7-8750H@2.20GHz six-core; motherboard: HP 84DB; memory: 16GB (Samsung DDR42666MHz); main hard disk: Samsung MZVLW128HEGR-000H1 (128GB / solid state drive); main graphics card: Nvidia GeForce GTX 1060 (6GB / HP); software environment: operating system: Ubantu 16.04; machine learning framework: LightGBM; language and development environment: Python 3.6, Anaconda3.

[0078] In the invention, the model first combines a wind power data set in a certain place in China with a wind power data set of Amprion, a German power company, to confirm the positive impact of financial factors on the prediction effect; secondly, the prediction effect of the model is demonstrated using a wind power data set for the entire Netherlands. Therefore, in order to confirm the positive impact of financial factors on prediction accuracy, and that the LightGBM algorithm has a high prediction accuracy and the fastest training speed, it is necessary to compare it with a model based on the XGBoost algorithm, a model based on the decision tree algorithm, and a model based on the random forest algorithm after the experiment. The basic information of the data set is shown in Table 1:

[0079] Table 1 Basic information of the dataset

[0080]

[0081] Embodiment 1

[0082] This study used a wind power dataset from a wind farm in a certain place in China for ablation experiments. The dataset consists of two parts: NWP data and wind farm power data.

[0083] There are 13493 entries in the NWP data, each of which includes 9 fields: date and time occupy 2 fields, and meteorological characteristics (such as wind speed, wind direction, air density and atmospheric pressure) occupy the other 7 fields. When processing empty data and abnormal data, the interpolation method of adding the power value of the first 15 minutes and the power value of the last 15 minutes is adopted, and the sampling records with zero data caused by wind turbine maintenance are deleted. Among the features that affect the output power of wind power generation, wind direction and wind speed directly affect the output power. Wind is caused by uneven solar radiation on the earth's surface. Changes in temperature will affect changes in air density, thereby affecting the airflow through the wind turbine blades, resulting in changes in the output power of wind power generation equipment. Therefore, temperature, wind direction and wind speed are selected as input features in meteorological data. At the same time, the four financial factors derived from wind power time series data proposed in this study are all used as input features, namely K, D, J in KDJFWP and RSI in RSIFWP. Through the above analysis, 8 features (wind speed, wind direction, temperature, K, D, J, RSI and current wind power output value) were left in 11990 valid and continuous entries, and they were converted into the initial data set, and the prediction target was the wind power 15 minutes after the current time. The experiment was divided into four groups, and the specific details are shown in Table 2. The model prediction evaluation indicators are shown in Table 3 (the bold items in the table are the best results, and the underlined items are the second best results), and the feature importance scores are shown in Table 4. The machine learning method used in this study on this data set is LightGBM, and the comparison between the true value and the predicted value is shown in the figure below. Figure 3 shown.

[0084] Table 2 Experimental groups and details in a wind power generation dataset in a certain place in China

[0085]

[0086] Table 3 Experimental prediction results of each group in a certain wind power data set

[0087]

[0088] Table 4 Importance score of each group of experimental features in a wind power generation dataset in a certain place in China

[0089]

[0090]

[0091] As shown in Table 3, the experimental group with only financial factors achieved the best prediction results. The mixed experimental group with financial factors and meteorological data achieved the second best prediction results, and the experimental group with only meteorological data had poor prediction results. Figure 3 (a) in the figure is a comparison chart of the true value and the predicted value. Different model input features have different degrees of improvement on the prediction accuracy of the model. Among them, the improvement effect of Experiment III is the most obvious, with MSE reduced by about 45.1%; RMSE reduced by about 25.9%; MAE reduced by about 57.4%; R2score increased by about 5%; AR increased by about 10.5%; NB / T increased by about 1.4%; and prediction time saved by about 70.7%. By comparing Experiment II and Experiment III, it can be found that the addition of financial factors reduces MSE by about 29.6%; RMSE by about 16.1%; MAE by about 29.6%; R2score by about 2.5%; AR by about 3.1%; NB / T by about 1.4%. At the same time, benefiting from the logical design of LightGBM feature selection, the increase in input features reduces the calculation time of the model and saves about 54.8% of the prediction time. By analyzing the prediction results of Experiment II and Experiment III, it is confirmed that the addition of financial factors can improve the accuracy of the prediction model. Comparing Experiment III with Experiment IV, the mixed input of financial factors and meteorological data failed to improve the model's prediction accuracy, but instead slightly reduced it. Figure 3 (b) Figure 3 (c) and Figure 3 In the comparison of the true value and predicted value in (d), the predicted image of Experiment III has the best fit with the standard image. In Table 4, the feature importance score refers to the number of times the feature is used as a split node in all gradient boosting regression trees. The table shows the importance scores of the first three features in each group of experiments. From the feature importance scores of Experiments III and IV, it can be seen that the financial factor is used as an important basis for the LightGBM algorithm to split the node. Therefore, the positive role of the financial factor in model prediction is once again confirmed.

[0092] Embodiment 2

[0093] The present invention obtains the wind power data set of Amprion, a German power company, through ENTSO-E. The data in this data set are wind power data from January 1, 2015 to October 31, 2018, and the collection frequency is recorded every 15 minutes. 96 data points can be obtained every day, totaling 200,998 data. Because the data comes from the company's entire wind power generation system, the data integrity is extremely high. Similarly, because the data covers the company's entire wind power generation system, it is not possible to use meteorological data as input features during prediction. Therefore, in the model proposed in this study, this data set can only use the financial factors derived from the wind power time series data as input features. In summary, the financial factors derived from the wind power time series data are spliced ​​with the power data. After splicing, the time and date fields are converted into indexes and are not used for machine learning. Five features (K, D, J, RSI, current wind power output value) are left in 200,998 valid and continuous entries, which are converted into the initial data set, and the prediction target is the wind power generation 15 minutes after the current time. This study conducted multiple experiments on this dataset using different prediction models, proving that adding financial factors to different prediction models can improve the prediction accuracy of the models to varying degrees. The training set accounts for 90%, the test set accounts for 10%, the experimental grouping and details are shown in Table 5, the model prediction evaluation indicators are shown in Table 6, and the comparison between the true value and the predicted value is shown in Table 6. Figure 4 shown.

[0094] Table 5 Experimental grouping and details in the Amprion wind power dataset of the German power company

[0095]

[0096]

[0097] Table 6 Experimental prediction results for each group in the German power company Amprion wind power data set

[0098]

[0099] As shown in Table 6, benefiting from the addition of financial factors, the prediction accuracy of the prediction models composed of different algorithms has been improved to varying degrees. Taking the MSE indicator as an example, in the decision tree prediction model (experiments I and II), the MSE is reduced by about 51.8%. In the random forest prediction model (experiments III and IV), the MSE is reduced by about 24.4%. In the XGBoost prediction model (experiments V and VI), the MSE is reduced by about 39.1%. In the LightGBM prediction model (experiments VII and VIII), the MSE is reduced by about 67.8%. Comparing Experiment VII with Experiment VIII, the prediction model proposed in this study is affected by financial factors. The RMSE is reduced by about 43.6%; MAE is reduced by about 61%; R2score is increased by about 0.3%; AR is increased by about 2.4%; NB / T is increased by about 0.5%.

[0100] From the data in Table 6, we can see that the prediction model based on the XGBoost algorithm in Experiment VI achieved the best prediction effect, but this study did not use it as a prediction algorithm. Comparing Experiment VI with Experiment VIII, it can be seen that the MSE, RMSE, and MAE of the LightGBM prediction model are slightly lower than those of the XGBoost prediction model; R2score, AR, and NB / T are on par with the XGBoost prediction model. However, the LightGBM prediction model saves about 86% of the computing time, so this study uses the LightGBM algorithm as the prediction algorithm of the model.

[0101] like Figure 4 As shown in the figure, Experiment II, Experiment IV, Experiment VI, and Experiment VIII are prediction models containing financial factors, and their fitting effects are significantly better than the ordinary prediction models without financial factors in Experiment I, Experiment III, Experiment V, and Experiment VII. The experiment shows that the positive impact of financial factors is not only reflected in the LightGBM algorithm, but also has universality and can improve the prediction accuracy of various prediction algorithms. In summary, the financial factors derived from wind power time series data have a good effect on the prediction model.

[0102] Embodiment 3

[0103] Windmills are considered a symbol of Dutch culture, and by 2050 all energy used in the Netherlands must come from renewable energy. During this period, wind power is key to the Netherlands' transition to a zero-carbon energy supply, so predicting its future wind power generation has certain practical significance. This study obtained the dataset through ENTSO-E. The data collection time started on January 1, 2015 and ended on September 14, 2018, with a total of 199,998 data, of which the training set accounted for 90% and the test set accounted for 10%. Because the data comes from the company's entire wind power system, the data integrity is extremely high and there are no missing values. Similarly, because the data covers the company's entire wind power system, it is not possible to use meteorological data as input features during prediction. Therefore, in the model proposed in this study, this dataset can only use the financial factors derived from wind power time series data as input features. In summary, the financial factors derived from wind power time series data are spliced ​​with power data. After splicing, the time and date fields are converted into indexes and are not used for machine learning. Five features (K, D, J, RSI, current wind power output) were retained from 200,998 valid and continuous entries, and the prediction target was the wind power output 15 minutes after the current time. The comparison models were XGBoost model, random forest model, and decision tree model. The prediction evaluation indicators of each model are shown in Table 7; the comparison between the predicted value and the true value is shown in the figure below. Figure 5 As shown in the figure, the prediction error statistics are as follows Figure 6 As shown (DT in the table and figure represents decision tree model, RF represents random forest model).

[0104] Table 7 Prediction and evaluation indicators of each model in the Netherlands wind power dataset

[0105]

[0106] As shown in Table 7, the prediction effect of the model proposed in this study is significantly better than that of the random forest model and the decision tree model. The prediction effect of the model based on the XGBoost algorithm on this data set is close to that of the model proposed in this study. Comparing the model proposed in this study with the XGBoost model, the RMSE difference is nearly 4.5%; MAE, MSE, R2score, AR, and NB / T are basically the same, but the prediction model proposed in this study saves about 88.51% of the calculation time.

[0107] like Figure 5 As shown in Figure 2, the model proposed in this study has a significantly better fitting effect than the random forest model and the decision tree model. Figure 6 As shown in the figure, it is a graph of error range statistics. Compared with the model based on XGBoost algorithm, the error range statistics of the model proposed in this study are basically consistent with the fitting effect. Compared with the random forest model and decision tree model, the error range within 5% has been greatly improved.

[0108] In summary, in terms of prediction effect, the prediction model based on the LightGBM algorithm proposed in this study has achieved good performance on the German 50Hertz Power Company wind power dataset and the Netherlands global wind power dataset; in terms of model training speed, the prediction model based on the LightGBM algorithm proposed in this study has achieved excellent performance in the above datasets, which is significantly better than the comparison models.

[0109] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for predicting ultra-short-term wind power generation, characterized in that: include: Obtain the original time series M of meteorological data and power generation corresponding to ultra-short-term wind power generation; Select the power of the current time point and the power of the time point before the current time point by the first preset time period in the original time series M, calculate the difference between the two powers to obtain the time window; reorganize the original time series M through the time window to obtain the new time series N; The process of reorganizing the original time series M to obtain the new time series N specifically includes: For time series data M, let the data value at time m+1 be t m+1 is the data value at the end of sampling, and the data value at time u+1 is t u+1 To predict the data value, take the time series data value t at time u u 、Time series data value at time u-1 t u-1 , the data difference between time u and time u-1 Constitute a time window; Combine the time window with the remaining covariates q at time u u 、w u 、e u Transform to form new time series data N. The transformation algorithm is: Among them, t u is the time series data value at time u, t m+1 is the data value at time m+1, t u+1 is the predicted data value; Sampling is performed once every second preset time period in the new time series N, adjacent sampling points are integrated into sampling data, and a wind power generation diagram is drawn through the sampling data; The financial factors KDJ and RSI are calculated based on the wind power diagram, and the LightGBM algorithm is used to build a forecasting model for extended financial factors; Among them, the formula for calculating the financial factor KDJ of wind power generation is: <h2 style=";text-align:left;direction:ltr">J<h2 style=";text-align:left;direction:ltr"> u <h2 style=";text-align:left;direction:ltr"> <2K<h2 style=";text-align:left;direction:ltr"> u <h2 style=";text-align:left;direction:ltr"> -D<h2 style=";text-align:left;direction:ltr"> u Among them, RSV is the immature random value, G3 is the power at the end of 3 hours, E3 is the maximum power within 3 hours, F3 is the minimum power within 3 hours, K u is the K value at the uth hour, D u is the D value at the uth hour, J u is the J value at hour u, RSV u is the RSV value at hour u, K u-1 is the K value of the first three hours, D u-1 is the D value of the first three hours; Among them, the financial factor RSI for calculating wind power generation specifically includes: Let the magnitude of the upward power fluctuation within 6 hours be A, let the magnitude of the downward power fluctuation within 6 hours be B, and the power value at the end of x hours is C x ; If C x -C x-1 > 0, then A+=C x -C x-1 ; If C x -C x-1 <0, then B+=C x-1 -C x ; Continuously calculate the values ​​of 6 sampling points, the calculation formula is as follows: Among them, RS is relative strength, A is the amplitude of the power upward fluctuation within 6 hours, B is the amplitude of the power downward fluctuation within 6 hours, and RSI1X is the value of 6 consecutive sampling points; The data set consisting of meteorological data, new time series N, financial factors KDJ and RSI is input into the prediction model to obtain the prediction result of wind power generation.

2. The method for predicting ultra-short-term wind power generation according to claim 1, characterized in that: Drawing a wind power generation diagram by sampling data specifically includes: The maximum value in the sampled data is regarded as the highest price in the financial market, and the minimum value is regarded as the lowest price in the financial market; the power output value of the first sampling point is regarded as the opening price in the financial market, and the power output value of the last sampling point is regarded as the closing price in the financial market; A plane rectangular coordinate system is established with time as the horizontal coordinate and wind power generation power as the vertical coordinate; In the plane rectangular coordinate system, if the closing price is higher than the opening price, it means that the opening price is below and the closing price is above, and a black rectangular column is drawn; if the closing price is lower than the opening price, it means that the opening price is above and the closing price is below, and a gray rectangular column is drawn to obtain the wind power generation power diagram.

3. A system for predicting ultra-short-term wind power generation based on the method for predicting ultra-short-term wind power generation according to claim 1, characterized in that: include: A data acquisition module is used to obtain the meteorological data corresponding to ultra-short-term wind power generation and the original time series M of the generated power; A time window module is used to select the power of the current time point and the power of the time point before the current time point by a preset time period in the original time series M, calculate the difference between the two powers to obtain the time window; and reorganize the original time series M through the time window to obtain a new time series N; A drawing module, used for sampling once every second preset time period in the new time series N, integrating adjacent sampling points into sampling data, and drawing a wind power generation diagram through the sampling data; Model building module, used to calculate the financial factors KDJ and RSI based on the wind power diagram, and use the LightGBM algorithm to build a forecasting model for extended financial factors; The prediction module is used to input the data group consisting of meteorological data, new time series N, financial factors KDJ and RSI into the prediction model to obtain the prediction result of wind power generation.

4. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the ultra-short-term wind power generation prediction method described in any one of claims 1-2 are implemented.

5. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the ultra-short-term wind power generation prediction method according to any one of claims 1 to 2 are implemented.

Images