Wind speed prediction method and system based on time sequence threshold segmentation, and medium

Abstract

The method comprises the following steps: acquiring a weather forecast numerical value, and respectively constructing grayscale images of meteorological characteristic data and target wind speed data; based on the target wind speed data grayscale image, performing secondary threshold segmentation operation to obtain wind speed grayscale images of four numerical intervals; performing feature correlation analysis according to the meteorological feature sequence data and the target wind speed sequence data to obtain influence values of the meteorological features and the target wind speed; carrying out random screening on the meteorological characteristic data through the influence values, and obtaining a mapping relation image of an input image and an output image based on the input data and the wind speed gray level images of the four numerical value intervals; and expanding the mapping relation image into time sequence data, performing target wind speed sequence prediction of four numerical intervals, and performing integrated comparison. According to the method, secondary threshold segmentation and meteorological characteristic random screening are carried out on the image, so that the uncertainty of wind speed prediction is reduced, and the precision of wind speed prediction and the accuracy of numerical weather prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of wind speed prediction, and in particular to a wind speed prediction method, system and medium based on time series threshold segmentation. Background Art

[0002] Wind speed is highly random and intermittent, making its volatility and uncertainty characteristics extremely prominent. The change in wind speed is affected by a variety of complex factors such as topography, atmospheric circulation, and temperature differences. These factors are intertwined and work together to cause strong spatiotemporal dynamic changes in wind speed. When wind speed data is passed as input to the prediction model, this volatility increases the difficulty of the model in processing data.

[0003] Traditional wind speed forecasting methods are usually difficult to accurately capture the inherent laws of wind speed changes. The main reason is that they directly input wind speed data containing uncertain characteristics into the forecasting model for training, but do not take effective methods to reduce the impact of wind speed uncertainty on the forecasting model. For example, when directly using time series forecasting models or machine learning models to learn wind speed data features with uncertainty, although they can grasp the overall trend of wind speed changes to a certain extent, when faced with sudden changes in wind speed and wind speed changes near the maximum and minimum values, these models are difficult to accurately capture the subtle trend of wind speed changes under special circumstances, resulting in reduced accuracy of wind speed forecasting. Low prediction accuracy and reduced model universality are the core problems facing current wind speed forecasting. Low-precision wind speed forecasts have brought many adverse effects on meteorological decision-making and other aspects. For example, in the process of meteorological decision-making, it is difficult for relevant departments to make accurate warnings and effective response measures in a timely manner based on inaccurate forecast results, which in turn has a negative impact on public safety and social and economic activities.

[0004] In order to enhance the accuracy and reliability of wind speed forecasts, it is crucial to construct the temporal regularity characteristics of wind speed data. By deeply exploring the temporal regularity in wind speed data, the prediction model can better handle the changing trend of wind speed, thereby improving the prediction accuracy. However, most current prediction methods have obvious shortcomings in this regard, especially in how to effectively segment uncertain wind speed data to construct temporal regularity characteristics. Further in-depth exploration and innovation are still needed. If this problem can be solved, it will bring a key breakthrough to the development of wind speed prediction technology and help provide a more accurate basis for meteorological decision-making. Summary of the invention

[0005] In view of this, the purpose of the embodiments of the present invention is to provide a wind speed prediction method, system and medium based on time series threshold segmentation, so as to solve the problem that the uncertainty characteristics of wind speed affect the accuracy and reliability of wind speed prediction. By deeply mining the time series laws in wind speed data, the prediction model can better handle the changing trend of wind speed, thereby improving the prediction accuracy.

[0006] On the one hand, an embodiment of the present invention provides a wind speed prediction method based on time series threshold segmentation, comprising:

[0007] Obtain weather forecast values ​​and construct grayscale images of meteorological characteristic data and target wind speed data respectively;

[0008] Based on the target wind speed data grayscale image, a secondary threshold segmentation operation is performed to obtain wind speed grayscale images in four numerical intervals to characterize the regular wind speed sequence in four non-continuous time periods;

[0009] According to the meteorological characteristic sequence data and the target wind speed sequence data, characteristic correlation analysis is performed to obtain the impact value of each meteorological characteristic and the target wind speed;

[0010] Randomly screening the meteorological characteristic data by using the influence value, constructing at least one characteristic matrix, and using the corresponding meteorological characteristic grayscale image as input data for the prediction model;

[0011] Based on the input data and the wind speed grayscale images in the four numerical intervals, a mapping relationship image between the input image and the output image is obtained; after the mapping relationship image is expanded into time series data, target wind speed series in four numerical intervals are predicted and integrated and compared.

[0012] Optionally, the randomly screening the meteorological characteristic data includes:

[0013] Based on the meteorological characteristics corresponding to the impact values, corresponding feature labels are set, and feature labels whose sum of impact values ​​is a target impact value are randomly selected; wherein the target impact value is the sum of the impact values;

[0014] Selecting at least one feature sequence corresponding to a feature label to construct a feature matrix;

[0015] The prediction accuracy based on at least one feature matrix as input is compared, and the meteorological feature grayscale image corresponding to the comparison result with the highest accuracy is used as input.

[0016] Optionally, the step of obtaining the wind speed grayscale images in four value intervals after performing the secondary threshold segmentation operation includes:

[0017] Segmenting the target wind speed data grayscale image once based on the improved OTSU algorithm to obtain a wind speed image threshold of the target wind speed data grayscale image;

[0018] If the first wind speed value is greater than or equal to the wind speed image threshold, the original value is retained, and the wind speed value less than the wind speed image threshold is assigned a value of 0 to obtain a first wind speed image;

[0019] If the second wind speed value is less than the wind speed image threshold, the original value is retained, and the wind speed values ​​greater than or equal to the wind speed image threshold are assigned a value of 0 to obtain a second wind speed image.

[0020] Optionally, the random screening of the meteorological characteristic data further includes:

[0021] The decision tree algorithm is used to train the target wind speed to obtain the wind speed prediction model;

[0022] An interpretative analysis is performed on the wind speed prediction model based on the Shap interpretability model to obtain the contribution of each meteorological feature to the wind speed;

[0023] The contribution of each of the meteorological characteristics to the wind speed is compared and analyzed to obtain a first contribution, and random screening is performed according to the meteorological characteristics corresponding to the first contribution, and the formula is as follows:

[0024]

[0025] In the formula, Represents feature x i The predicted value of model f The contribution degree of f(S∪{x i}) contains feature x i and the model prediction value of the feature subset S, and f(S) is the model prediction value containing only the feature subset S.

[0026] Optionally, the obtaining of weather forecast values ​​and respectively constructing grayscale images of meteorological characteristic data and target wind speed data includes:

[0027] Based on the weather forecast value, the meteorological characteristics are obtained and set to x i , the target wind speed sequence is set to y and normalized by logarithmic transformation, as follows:

[0028]

[0029] Based on the normalized sequence, it is reconstructed into a grayscale image of size b×c.

[0030] Optionally, segmenting the target wind speed data grayscale image once based on the improved OTSU algorithm to obtain a wind speed image threshold of the target wind speed data grayscale image further includes:

[0031] The original threshold segmentation distinguishes the foreground from the background by the maximum inter-class variance, and the formula is as follows:

[0032] g0=w0(u0-u) 2 +w1(u1-u) 2 ;

[0033] Where w0 is the ratio of the number of foreground pixels to the total image, u0 is the average gray value of the foreground image, w1 is the ratio of the number of background pixels to the total image, u1 is the average gray value of the background image;

[0034] The maximum inter-class variance is modified to the maximum inter-class cubic variance, and the improved formula is as follows:

[0035] g1=w0(u0-u) 3 +w1(u1-u) 3 ;

[0036] Where w0 is the ratio of the number of foreground pixels to the total image, u0 is the average gray value of the foreground image, w1 is the ratio of the number of background pixels to the total image, and u1 is the average gray value of the background image.

[0037] Optionally, the first wind speed image and the second wind speed image are respectively subjected to secondary segmentation using an improved OTSU algorithm to obtain wind speed grayscale images in four numerical intervals, including:

[0038] The first wind speed image is subjected to secondary segmentation, that is, the maximum inter-class variance is changed to the maximum inter-class quartic variance, and the improved formula is as follows:

[0039] g2=w 10 (u 10 -u') 4 +w 11 (u 11 -u') 4 ;

[0040] In the formula, w 10 is the ratio of foreground pixels to the total image, u 10 is the average gray value of the foreground image, w 11 is the ratio of background pixels to the total image, u 11 is the average gray value of the background image;

[0041] The second wind speed image is subjected to secondary segmentation, that is, the maximum inter-class variance is changed to the maximum inter-class cubic variance, and the improved formula is as follows:

[0042]

[0043] In the formula, w 20is the ratio of foreground pixels to the total image, u 20 is the average gray value of the foreground image, w 21 is the ratio of background pixels to the total image, u 21 is the average gray value of the background image.

[0044] On the other hand, an embodiment of the present invention provides a wind speed prediction device based on time series threshold segmentation, characterized in that it includes:

[0045] at least one processor;

[0046] at least one memory for storing at least one program;

[0047] When the at least one program is executed by the at least one processor, the at least one processor implements the wind speed prediction method based on time series threshold segmentation as described in any one of the above items.

[0048] On the other hand, an embodiment of the present invention provides a computer-readable storage medium, which stores a program executable by a processor, characterized in that the program executable by the processor is used to execute a wind speed prediction method based on time series threshold segmentation as described in any of the above items when executed by the processor.

[0049] The embodiment of the present invention obtains weather forecast values ​​to construct grayscale images of meteorological feature data and target wind speed data respectively; based on the grayscale image of the target wind speed data, a secondary threshold segmentation operation is performed to obtain wind speed grayscale images of four numerical intervals to characterize the regular wind speed sequence of four non-continuous time periods; feature correlation analysis is performed based on the meteorological feature sequence data and the target wind speed sequence data to obtain the influence value of each meteorological feature and the target wind speed; the meteorological feature data is randomly screened by the influence value to construct at least one feature matrix, and the corresponding meteorological feature grayscale image is used as the input data of the prediction model; a mapping relationship image between the input image and the output image is obtained based on the input data and the wind speed grayscale image of the four numerical intervals; the target wind speed sequence of the four numerical intervals is predicted and integrated comparison is performed after the mapping relationship image is expanded into time series data. The present invention constructs the temporal regularity characteristics of wind speed in each numerical interval through threshold segmentation, converts the uncertainty of wind speed data into regularity, and helps to optimize the training process of the prediction model. At the same time, by integrating the models of each interval, we can fully learn the regular characteristics of wind speed in each period, perform secondary threshold segmentation on the target wind speed image, and randomly screen the meteorological characteristics to reduce the uncertainty of wind speed prediction, improve the accuracy of wind speed prediction and numerical weather forecast, and help to more accurately predict wind speed in the meteorological decision-making process. So as to make corresponding meteorological warnings and effective disaster prevention measures in time, reduce social and economic losses, and improve the safety level of people in coastal and strong convective areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0051] Figure 1 is a flow chart of a wind speed prediction method based on time series threshold segmentation provided by an embodiment of the present invention;

[0052] Figure 2 It is a schematic diagram of the overall process provided by an embodiment of the present invention;

[0053] Figure 3 A schematic diagram of an image composed of an original wind speed sequence provided by an embodiment of the present invention;

[0054] Figure 4 is a schematic diagram of a contribution image provided by an embodiment of the present invention;

[0055] Figure 5is a schematic diagram of a wind speed sequence image consisting of 5840 normalized data provided by an embodiment of the present invention;

[0056] Figure 6 is a schematic diagram of a 73×80 wind speed grayscale image composed of 5840 normalized wind speeds provided in an embodiment of the present invention;

[0057] Figure 7 is a schematic diagram of a low wind speed grayscale image after secondary threshold segmentation provided by an embodiment of the present invention;

[0058] Figure 8 is a schematic diagram of a low wind speed time series provided by an embodiment of the present invention;

[0059] Fig. 9 is a schematic diagram of a medium to low wind speed grayscale image after secondary threshold segmentation provided by an embodiment of the present invention;

[0060] Fig.10 is a schematic diagram of a time series image of medium to low wind speed provided by an embodiment of the present invention;

[0061] Fig.11 is a schematic diagram of a medium to high wind speed grayscale image after secondary threshold segmentation provided by an embodiment of the present invention;

[0062] Fig.12 is a schematic diagram of a time series image of medium to high wind speed provided by an embodiment of the present invention;

[0063] Fig.13 is a schematic diagram of a high wind speed grayscale image after secondary threshold segmentation provided by an embodiment of the present invention;

[0064] Fig.14 is a schematic diagram of a high wind speed time series image provided by an embodiment of the present invention;

[0065] Fig.15 It is a schematic diagram of a test set comparison of the wind speed sequence predicted by the prediction method of the present invention, the original wind speed sequence, and the wind speed sequence predicted by the LSTM method;

[0066] Fig.16 It is a comparison bar chart of error indicators between the prediction method of the present invention and the LSTM method;

[0067] Fig.17 is a structural block diagram of a system provided by an embodiment of the present invention;

[0068] Fig.18 It is a structural block diagram of a device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application 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 application and are not used to limit the present application.

[0070] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0071] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0072] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0073] The block diagrams shown in the accompanying drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or in one or more hardware charging modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0074] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0075] See also Figures 1 to 16 As shown, an embodiment of the present invention provides a wind speed prediction method based on time series threshold segmentation, including:

[0076] S1, obtain weather forecast values ​​and construct grayscale images of meteorological characteristic data and target wind speed data respectively;

[0077] Exemplarily, the meteorological characteristic data include temperature, humidity, air pressure, etc. The target wind speed data include time, average wind speed, maximum wind speed, wind direction, temperature, air pressure, humidity, etc.

[0078] Specifically, the step of obtaining weather forecast values ​​and constructing grayscale images of meteorological characteristic data and target wind speed data respectively includes:

[0079] Based on the weather forecast value, the meteorological characteristics are obtained and set to x i , the target wind speed sequence is set to y and normalized by logarithmic transformation, as follows:

[0080]

[0081] Based on the normalized sequence, it is reconstructed into a grayscale image of size b×c.

[0082] For example, constructing corresponding features based on existing meteorological data and converting meteorological feature data and target wind speed data into grayscale images can more accurately characterize the characteristics of uncertainty and volatility.

[0083] S2, based on the target wind speed data grayscale image, performing a secondary threshold segmentation operation to obtain wind speed grayscale images in four numerical intervals to characterize the regular wind speed sequence in four non-continuous time periods;

[0084] Specifically, the process of obtaining the wind speed grayscale images in four value intervals after performing the secondary threshold segmentation operation includes:

[0085] Segmenting the target wind speed data grayscale image once based on the improved OTSU algorithm to obtain a wind speed image threshold of the target wind speed data grayscale image;

[0086] If the first wind speed value is greater than or equal to the wind speed image threshold, the original value is retained, and the wind speed value less than the wind speed image threshold is assigned a value of 0 to obtain a first wind speed image;

[0087] If the second wind speed value is less than the wind speed image threshold, the original value is retained, and the wind speed values ​​greater than or equal to the wind speed image threshold are assigned a value of 0 to obtain a second wind speed image.

[0088] Exemplarily, the first wind speed image is a high wind speed image divided once, and the second wind speed image is a low wind speed image divided once. First, the OTSU algorithm is used to perform a segmentation to obtain the first wind speed image and the second wind speed image. Then, the OTSU method is used to perform improved secondary segmentation on the first wind speed image and the second wind speed image to obtain grayscale images of the four wind speed intervals. Finally, the grayscale images of the four wind speed intervals are expanded in time series to obtain regular wind speed sequences in the four non-continuous time periods.

[0089] Specifically, segmenting the target wind speed data grayscale image once based on the improved OTSU algorithm to obtain the wind speed image threshold of the target wind speed data grayscale image also includes:

[0090] The original threshold segmentation distinguishes the foreground from the background by the maximum inter-class variance, and the formula is as follows:

[0091] g0=w0(u0-u) 2 +w1(u1-u) 2 ;

[0092] Where w0 is the ratio of the number of foreground pixels to the total image, u0 is the average gray value of the foreground image, w1 is the ratio of the number of background pixels to the total image, u1 is the average gray value of the background image;

[0093] The maximum inter-class variance is modified to the maximum inter-class cubic variance, and the improved formula is as follows:

[0094] g1=w0(u0-u) 3 +w1(u1-u) 3 ;

[0095] In the formula, w0 is the ratio of the number of foreground pixels to the total image, u0 is the average gray value of the foreground image, w1 is the ratio of the number of background pixels to the total image, and u1 is the average gray value of the background image; the segmentation effect is made more uniform by increasing the difference between the foreground and the background.

[0096] Since the grayscale difference between the first wind speed image and the second wind speed image after one segmentation is small, the segmentation method is improved as follows:

[0097] Specifically, the first wind speed image and the second wind speed image are respectively subjected to secondary segmentation using the improved OTSU algorithm to obtain wind speed grayscale images in four numerical intervals, including:

[0098] The first wind speed image is subjected to secondary segmentation, that is, the maximum inter-class variance is changed to the maximum inter-class quartic variance, and the improved formula is as follows:

[0099] g2=w 10 (u 10 -u') 4 +w 11 (u 11 -u') 4 ;

[0100] In the formula, w 10 is the ratio of foreground pixels to the total image, u 10 is the average gray value of the foreground image, w 11 is the ratio of background pixels to the total image, u 11is the average gray value of the background image;

[0101] The second wind speed image is subjected to secondary segmentation, that is, the maximum inter-class variance is changed to the maximum inter-class cubic variance, and the improved formula is as follows:

[0102]

[0103] In the formula, w 20 is the ratio of foreground pixels to the total image, u 20 is the average gray value of the foreground image, w 21 is the ratio of background pixels to the total image, u 21 is the average gray value of the background image; by increasing the difference between the foreground and background, the segmentation effect is made more uniform.

[0104] Exemplarily, the first wind speed image is finally segmented into a third wind speed image and a fourth wind speed image; the second wind speed image is segmented into a fifth wind speed image and a sixth wind speed image; that is, after the second segmentation, there are four wind speed image intervals.

[0105] It can be understood that the third wind speed image is a high wind speed image after secondary segmentation, the fourth wind speed image is a medium to high wind speed image after secondary segmentation, the fifth wind speed image is a medium to low wind speed image after secondary segmentation, and the sixth wind speed image is a low wind speed image after secondary segmentation.

[0106] S3, performing feature correlation analysis based on the meteorological feature sequence data and the target wind speed sequence data to obtain the impact value of each meteorological feature and the target wind speed;

[0107] For example, the filtered feature image and the segmented target wind speed image are used as input and output respectively, and the relationship between the input and output is extracted using CGAN (conditional generative adversarial network). Then, it is expanded into sequence data and predicted using BiGRU (bidirectional gated recurrent unit), and the wind speed training models of four intervals with different feature groups as input are obtained.

[0108] Exemplarily, an input feature score image is obtained, defined as c, noise z is randomly generated, and the feature score image c and the noise z are input into the generator G together to obtain an image x=G(c, z);

[0109] Obtain a wind speed interval image, defined as d, and input the wind speed interval image d, the generated image x, and its corresponding feature score image c into the discriminator D. The discriminator D makes a judgment based on the input and obtains the probability that the input image is a real image;

[0110] The generator generates an output wind speed image based on the input feature score image. The discriminator obtains the discrimination probability through learning and distinguishes the real output image from the image generated by the generator. Then the loss function is calculated based on the discriminator output and the real label (real image is 1, fake image is 0). The generator goal is to minimize the discrimination probability, while the discriminator goal is to maximize the discrimination probability. The target loss function is calculated as follows:

[0111]

[0112] The trained generator is used as a feature extractor to extract the features of the hidden layer, and the features are expanded in time series. The expanded sequence is used as the input feature and output sequence of the BiGRU network.

[0113] S4, randomly screening the meteorological characteristic data according to the influence value, constructing at least one characteristic matrix, and using the corresponding meteorological characteristic grayscale image as input data for a prediction model;

[0114] Specifically, the random screening of the meteorological characteristic data includes:

[0115] Based on the meteorological characteristics corresponding to the impact values, corresponding feature labels are set, and feature labels whose sum of impact values ​​is a target impact value are randomly selected; wherein the target impact value is the sum of the impact values;

[0116] Selecting at least one feature sequence corresponding to a feature label to construct a feature matrix;

[0117] The prediction accuracy based on at least one feature matrix as input is compared, and the meteorological feature grayscale image corresponding to the comparison result with the highest accuracy is used as input.

[0118] Exemplarily, the meteorological characteristic grayscale image corresponding to the optimal characteristic matrix is ​​selected as input according to the comparison result.

[0119] Specifically, the random screening of the meteorological characteristic data also includes:

[0120] The decision tree algorithm is used to train the target wind speed to obtain the wind speed prediction model;

[0121] An interpretative analysis is performed on the wind speed prediction model based on the Shap interpretability model to obtain the contribution of each meteorological feature to the wind speed;

[0122] The contribution of each of the meteorological characteristics to the wind speed is compared and analyzed to obtain a first contribution, and random screening is performed according to the meteorological characteristics corresponding to the first contribution.

[0123] For example, given n meteorological characteristics x i are x1, x2, …, xn , the target wind speed is y, and the value predicted by the decision tree model is Let each feature x i The contribution of According to the Shap model, the decision tree model process formula is as follows:

[0124]

[0125] In the formula, Represents feature x i The predicted value of model f The contribution degree of f(S∪{x i}) contains feature x i and the model prediction value of the feature subset S, and f(S) is the model prediction value containing only the feature subset S.

[0126] For example, taking the European Centre for Medium-Range Weather Forecasts (ECMWF) as an example, meteorological feature data and wind speed data are obtained from ECMWF. Meteorological features may include: temperature, humidity, air pressure, wind direction, etc., and wind speed data is the target variable. Clean and preprocess the data: that is, handle missing values, standardize or normalize (if the decision tree model does not require standardization, it can be skipped). Use meteorological feature data as input and wind speed data as the target variable to train a decision tree regression model, namely, DecisionTree Regressor, to fit the relationship between features and targets. Use SHAP (SHapley AdditiveexPlanations) to interpret the decision tree model, that is, calculate the contribution of each meteorological feature to the target wind speed forecast and visualize the feature contribution. According to the feature contribution ranking, identify the meteorological features with the first contribution (i.e., high contribution) (such as Importance>0.1), and further optimize the feature set through random screening for subsequent modeling.

[0127] By randomly screening meteorological characteristics, the negative impact of meteorological characteristic input on the wind speed prediction model can be reduced.

[0128] S5, obtaining a mapping relationship image between the input image and the output image based on the input data and the wind speed grayscale images in the four numerical intervals; and performing target wind speed series prediction in four numerical intervals and performing integrated comparison after expanding the mapping relationship image into time series data.

[0129] Based on the training models of wind speed in the four intervals with different feature groups as input, the errors of each feature group in a single interval are compared, and the features corresponding to the prediction model with the best prediction effect are selected as input, and the optimal prediction models of the four intervals are selected respectively. Then the bagging ensemble learning model is used to integrate the wind speed prediction models of the four intervals, and the prediction results are denormalized to predict the final wind speed prediction value. The integration process is as follows:

[0130] For the optimal training model of wind speed in four intervals, input its training data and test data into four training models respectively to obtain training prediction sequences and test prediction sequences. Sum the four training and test prediction sequences, and use the summed training prediction data as input; sum the original training data and original test data of wind speed in four intervals, and use the summed original training data as output to train the Bagging model. Use the trained model to predict the summed test prediction sequence to obtain integrated test prediction data, and the trained Bagging model obtained is the final prediction model;

[0131] The present invention firstly utilizes grayscale image construction technology to intuitively display the intrinsic characteristics of meteorological characteristics and wind speed data. Then, through in-depth analysis of wind speed uncertainty, the improved secondary threshold segmentation is adopted to divide the wind speed grayscale image into four numerical intervals. In this way, the regular characteristics of wind speed in four non-continuous time periods are characterized, and the wind speed images of the four intervals are used as target prediction images. Subsequently, the deep mapping relationship between the input image and the target image is accurately extracted through the conditional generative adversarial network (CGAN), and then the mapping image is expanded into time series data, and the bidirectional gated recurrent unit (BiGRU) is used for efficient training, so as to obtain the wind speed prediction model for each numerical interval. Finally, through Bagging ensemble learning, the models of each interval are integrated to construct the final wind speed prediction model. Through the time series threshold secondary segmentation method, the accuracy of wind speed prediction is improved, and the accuracy of numerical weather forecast is enhanced.

[0132] This application constructs wind speed grayscale images of different degrees of refinement as input when predicting short-term and medium- and long-term wind speeds. Since the secondary threshold segmentation can transform uncertainty features into regular features, and CGAN (conditional generative adversarial network) can process images with a large number of samples, BiGRU (bidirectional gated recurrent unit) can learn long-term series data more fully, so it can effectively predict both short-term wind speed series and medium- and long-term wind speed series. There will be no problems of underfitting or large deviations in late data prediction due to the large sample size and long time span of medium- and long-term wind speed data.

[0133] Among them, this method uses four performance indicators, namely, mean absolute error (MAE): mean absolute percentage error (MAPE), root mean square error (RMSE), and mean square error (MSE); the formula for evaluating the mean absolute error (MAE) of the prediction model in the test sample is as follows:

[0134]

[0135] The formula for evaluating the mean absolute percentage error (MAPE) of the prediction model in the test sample is as follows:

[0136]

[0137] The formula for evaluating the root mean square error (RMSE) of the prediction model in the test sample is as follows:

[0138]

[0139] The mean square error (MSE) formula for evaluating the prediction model in the test sample is as follows:

[0140]

[0141] Where n represents the number of samples, y i represents the true value, represents the predicted value, Represents the mean of the true values.

[0142] MAE, MAPE, RMSE and MSE respectively represent the prediction error of the prediction model, that is, the gap between the predicted value and the actual value. The smaller the value, the smaller the prediction error of the model and the higher the prediction accuracy of the model.

[0143] See also Fig.17 The embodiment of the present invention provides a wind speed prediction device based on time series threshold segmentation, characterized in that the device comprises:

[0144] The first module is used to obtain weather forecast values ​​and construct grayscale images of meteorological characteristic data and target wind speed data respectively;

[0145] The second module is used to perform a secondary threshold segmentation operation based on the target wind speed data grayscale image to obtain wind speed grayscale images in four numerical intervals to characterize the regular wind speed sequence in four non-continuous time periods;

[0146] The third module is used to perform feature correlation analysis based on the meteorological feature sequence data and the target wind speed sequence data to obtain the impact value of each meteorological feature and the target wind speed;

[0147] The fourth module is used to randomly screen the meteorological characteristic data according to the influence value, construct at least one characteristic matrix, and use the corresponding meteorological characteristic grayscale image as the prediction model input data;

[0148] The fifth module is used to obtain a mapping relationship image between the input image and the output image based on the input data and the wind speed grayscale images in the four numerical intervals; after expanding the mapping relationship image into time series data, the target wind speed series in the four numerical intervals is predicted and integratedly compared.

[0149] It can be seen that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0150] See also Fig.18 The embodiment of the present invention provides a nonlinear signal time-frequency spectrum generation system based on LSTM, characterized in that it includes:

[0151] at least one processor;

[0152] at least one memory for storing at least one program;

[0153] When the at least one program is executed by the at least one processor, the at least one processor implements the wind speed prediction method based on time series threshold segmentation as described in any one of the above items.

[0154] It can be seen that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0155] In addition, the embodiment of the present application also discloses a computer program product or a computer program, and the computer program product or the computer program is stored in a computer-readable storage medium. The processor of the computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device performs the above method. Similarly, the contents in the above method embodiment are all applicable to the storage medium embodiment, and the functions specifically implemented by the storage medium embodiment are the same as those in the above method embodiment, and the beneficial effects achieved are also the same as those achieved by the above method embodiment.

[0156] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the charging modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0157] It will be appreciated by those skilled in the art that all or some of the methods disclosed above, the system can be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, magnetic tape, disk storage or other magnetic storage device, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0158] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0159] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A wind speed prediction method based on time series threshold segmentation, characterized in that: include: Obtain weather forecast values ​​and construct grayscale images of meteorological characteristic data and target wind speed data respectively; Based on the target wind speed data grayscale image, a secondary threshold segmentation operation is performed to obtain wind speed grayscale images in four numerical intervals to characterize the regular wind speed sequence in four non-continuous time periods; According to the meteorological characteristic sequence data and the target wind speed sequence data, characteristic correlation analysis is performed to obtain the impact value of each meteorological characteristic and the target wind speed; Randomly screening the meteorological characteristic data by using the influence value, constructing at least one characteristic matrix, and using the corresponding meteorological characteristic grayscale image as input data for the prediction model; A mapping relationship image between an input image and an output image is obtained based on the input data and the wind speed grayscale images in the four numerical intervals; after the mapping relationship image is expanded into time series data, target wind speed series in four numerical intervals are predicted and integratedly compared.

2. The method according to claim 1, characterized in that: The random screening of the meteorological characteristic data comprises: Based on the meteorological characteristics corresponding to the impact values, corresponding feature labels are set, and feature labels whose sum of impact values ​​is a target impact value are randomly selected; wherein the target impact value is the sum of the impact values; Selecting at least one feature sequence corresponding to a feature label to construct a feature matrix; The prediction accuracy based on at least one feature matrix as input is compared, and the meteorological feature grayscale image corresponding to the comparison result with the highest accuracy is used as input.

3. The method according to claim 1, characterized in that The process of obtaining the wind speed grayscale images in four value intervals after performing the secondary threshold segmentation operation comprises: Segmenting the target wind speed data grayscale image once based on the improved OTSU algorithm to obtain a wind speed image threshold of the target wind speed data grayscale image; If the first wind speed value is greater than or equal to the wind speed image threshold, the original value is retained, and the wind speed value less than the wind speed image threshold is assigned a value of 0 to obtain a first wind speed image; If the second wind speed value is less than the wind speed image threshold, the original value is retained, and the wind speed values ​​greater than or equal to the wind speed image threshold are assigned a value of 0 to obtain a second wind speed image.

4. The method according to claim 1, characterized in that: The random screening of the meteorological characteristic data also includes: The decision tree algorithm is used to train the target wind speed to obtain the wind speed prediction model; An interpretative analysis is performed on the wind speed prediction model based on the Shap interpretability model to obtain the contribution of each meteorological feature to the wind speed; The contribution of each of the meteorological characteristics to the wind speed is compared and analyzed to obtain a first contribution, and random screening is performed according to the meteorological characteristics corresponding to the first contribution, and the formula is as follows: In the formula, Represents feature x i The predicted value of model f The contribution degree of f(S∪{x i }) contains feature x i and the model prediction value of the feature subset S, and f(S) is the model prediction value containing only the feature subset S.

5. The method according to claim 1, characterized in that: The obtaining of weather forecast values ​​and constructing grayscale images of meteorological characteristic data and target wind speed data respectively include: Based on the weather forecast value, the meteorological characteristics are obtained and set as xi, the target wind speed sequence is set as y, and normalized by logarithmic transformation, as shown in the following formula: Based on the normalized sequence, it is reconstructed into a grayscale image of size b×c.

6. The method according to claim 3, characterized in that The target wind speed data grayscale image is segmented once based on the improved OTSU algorithm to obtain the wind speed image threshold of the target wind speed data grayscale image, and further includes: The original threshold segmentation distinguishes the foreground from the background by the maximum inter-class variance, and the formula is as follows: g 0=w0(u0-u)2+w1(u1-u)2; Where w0 is the ratio of the number of foreground pixels to the total image, u0 is the average gray value of the foreground image, w1 is the ratio of the number of background pixels to the total image, u1 is the average gray value of the background image; The maximum inter-class variance is modified to the maximum inter-class cubic variance, and the improved formula is as follows: g 1=w0(u0-u)3+w1(u1-u)3; Where w0 is the ratio of the number of foreground pixels to the total image, u0 is the average gray value of the foreground image, w1 is the ratio of the number of background pixels to the total image, and u1 is the average gray value of the background image.

7. The method according to claim 6, characterized in that The first wind speed image and the second wind speed image are respectively subjected to secondary segmentation using the improved OTSU algorithm to obtain wind speed grayscale images in four numerical intervals, including: The first wind speed image is subjected to secondary segmentation, that is, the maximum inter-class variance is changed to the maximum inter-class quartic variance, and the improved formula is as follows: g 2=w10(u10-u')4+w11(u11-u')4; In the formula, w 10 is the ratio of foreground pixels to the total image, u 10 is the average gray value of the foreground image, w 11 is the ratio of background pixels to the total image, u 11 is the average gray value of the background image; The second wind speed image is subjected to secondary segmentation, that is, the maximum inter-class variance is changed to the maximum inter-class cubic variance, and the improved formula is as follows: In the formula, w 20 is the ratio of foreground pixels to the total image, u 20 is the average gray value of the foreground image, w 21 is the ratio of background pixels to the total image, u 21 is the average gray value of the background image.

8. A wind speed prediction system based on time series threshold segmentation, characterized in that: The system comprises: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the wind speed prediction method based on time series threshold segmentation as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to execute the wind speed prediction method based on time series threshold segmentation as described in any one of claims 1 to 7 when executed by the processor.

Images