A storm cell tracking and forecasting method and system based on machine learning

Abstract

The invention discloses a method and system for tracking and forecasting a storm cell based on machine learning, and relates to the field of machine learning development. The method comprises the following specific steps: S1 is used to sample data samples of the same weather; S2 is used to obtain the data of the previous and next storm cells by using a volume scanning interval algorithm to determine whether they are continuous; S3 is used to determine whether they are in the same weather process by the characteristics and distance of the storm cells; S4 is used to track and match the storm cells in a time series to fit the motion trajectory of the time series; S5 is used to fit a straight line using a weighted least square method to forecast the position of the storm cell at the next moment; the method of the invention can forecast more cells, and the average forecast error of the obtained cell position is smaller, so that the tracking and forecasting of the storm cells can be better realized.

Description

Technical Field

[0001] The invention discloses a storm cell tracking and forecasting method and system based on machine learning, and relates to the technical field of machine learning development. Background Art

[0002] Machine learning is based on an algorithm that learns from data without relying on rule-based programming. Machine learning emerged as a scientific discipline in the late 1990s as advances in digitization and cheaper computing power enabled data scientists to stop building models and instead train computers to do the job. The huge and unmanageable volume and complexity of big data, which is attracting worldwide attention, has increased the potential for machine learning—and the need for it. The core idea of ​​weighted least squares is weighting. Generally speaking, recent data has a greater impact on the future than distant data. Therefore, a more reasonable approach is to use a weighted approach, giving greater weights to recent data and smaller weights to distant data. China has a complex and diverse climate, and severe convective weather such as hail and heavy rain often occurs, which has a great impact on regional economic development and people's production and life. In order to reduce the impact of severe convective weather, manual forecasting has received more and more attention. In order to reduce the forecast error and improve the forecast accuracy, we need to better use machine learning algorithms to track and forecast storm cells.

[0003] Among the existing storm cell tracking and forecasting algorithms, including the TITAN algorithm and the SCIT algorithm, some effects have been achieved in actual business. However, since the overall structure and local information of the storm cell are not well utilized, the forecast accuracy is not high enough and the forecast timeliness is unstable, so it has certain limitations.

[0004] Therefore, a storm cell tracking and forecasting method and system based on machine learning are invented to solve the above problems. Summary of the invention

[0005] The present invention aims at solving the problems of the prior art and provides a method and system for tracking and forecasting a storm cell based on machine learning. The technical solution adopted is: a method for tracking and forecasting a storm cell based on machine learning, wherein the specific steps of the method include:

[0006] S1 is sampled through data samples of the same weather;

[0007] S2 uses the volume scanning interval algorithm to obtain the data of the previous and next storm cells to determine whether they are continuous;

[0008] S3 determines whether they are in the same weather process based on the characteristics and distance of the storm cells;

[0009] S4 uses the tracking and matching of storm cells within a time series to fit the motion trajectory of the time series;

[0010] S5 uses the weighted least squares method to fit the straight line to predict the location of the storm cell at the next moment.

[0011] The specific steps of S2 using the volume scanning interval algorithm to obtain the front and rear storm monomer data and determine whether they are continuous include:

[0012] S201 checks the time interval between the body scans of the two preceding and following storm cells;

[0013] S202 compares the time threshold and the time interval;

[0014] S203: if the time threshold is less than the time interval, the data is judged to be discontinuous;

[0015] S213 If the time threshold is greater than the time interval, it is determined that the data is continuous.

[0016] The specific steps of S3 judging whether the storm cells are in the same weather process by the characteristics and distance of the storm cells include:

[0017] S301 calculates the mean value data of the single scan intensity by using a feature similarity algorithm;

[0018] S302 calculates the monomer distance using a distance priority algorithm;

[0019] S303 determines whether the calculated storm cells belong to the same weather process.

[0020] The S301 uses a feature similarity algorithm to calculate the single-cell scan intensity mean data using the difference in the storm single-cell intensity mean and the ratio of the area.

[0021] A storm monomer tracking and forecasting system based on machine learning, the system specifically includes a data sampling module, a continuous judgment module, a weather judgment module, a trajectory fitting module and a position prediction module:

[0022] Data sampling module: sampling through data samples of the same weather;

[0023] Continuity judgment module: Use the volume scanning interval algorithm to obtain the previous and next storm monomer data to determine whether it is continuous;

[0024] Weather judgment module: judge whether they are in the same weather process based on the characteristics and distance of the storm cell;

[0025] Trajectory fitting module: uses the storm monomer tracking and matching within a time series to fit the motion trajectory of the time series;

[0026] Position forecast module: Use the weighted least squares method to fit the straight line to predict the position of the storm cell at the next moment.

[0027] The continuity judgment module specifically includes an interval checking module, an interval comparing module, an interval judgment module A and an interval judgment module B:

[0028] Interval check module: Check the time interval between the two storm cells;

[0029] Interval comparison module: compares time threshold and time interval;

[0030] Interval judgment module A: if the time threshold is less than the time interval, the data is judged to be discontinuous;

[0031] Interval judgment module B: If the time threshold is greater than the time interval, the data is judged to be continuous.

[0032] The weather judgment module specifically includes an intensity calculation module, a distance calculation module and a weather judgment module:

[0033] Intensity calculation module: calculates the mean intensity data of the single scan using the feature similarity algorithm;

[0034] Spacing calculation module: calculates the monomer spacing using the distance priority algorithm;

[0035] Weather determination module: determines whether the calculated storm cells belong to the same weather process.

[0036] The intensity calculation module uses a feature similarity algorithm to calculate the storm cell intensity mean data using the difference in storm cell intensity mean values ​​and the ratio of the area.

[0037] The beneficial effects of the present invention are as follows: the method of the present invention utilizes a machine learning algorithm to address the problems of large forecast errors and a small number of predictable cells. First, representative data is obtained from a large amount of data. The embodiment of the method of the present invention can select the most representative hail and rainstorms in 75 weather processes in a certain city from 2013 to 2018 as research objects; then, the tracking and matching of cells in a time series are achieved according to the three algorithms of "volume scanning interval, feature similarity, and close distance priority", and finally the position of the storm cell at the next moment is predicted using the weighted least squares method; from the evaluation results, the method can predict more cells, and the average forecast error of the obtained cell position is smaller, which can better realize the tracking and forecasting of storm cells; the method of the present invention improves the accuracy of storm cell forecast, enhances the timeliness of storm cell forecast, increases the diversity of storm cell forecast, and enriches the applicability of machine learning algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are 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.

[0039] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is a structural schematic diagram of the system of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0041] Embodiment 1:

[0042] A storm cell tracking and forecasting method based on machine learning, wherein the method comprises the following specific steps:

[0043] S1 is sampled through data samples of the same weather;

[0044] S2 uses the volume scanning interval algorithm to obtain the data of the previous and next storm cells to determine whether they are continuous;

[0045] S3 determines whether they are in the same weather process based on the characteristics and distance of the storm cells;

[0046] S4 uses the tracking and matching of storm cells within a time series to fit the motion trajectory of the time series;

[0047] S5 uses the weighted least squares method to fit the straight line to predict the location of the storm cell at the next moment;

[0048] Furthermore, the specific steps of obtaining the front and rear storm monomer data by using the volume scanning interval algorithm and determining whether the data are continuous include:

[0049] S201 checks the time interval between the body scans of the two preceding and following storm cells;

[0050] S202 compares the time threshold and the time interval;

[0051] S203: if the time threshold is less than the time interval, the data is judged to be discontinuous;

[0052] S213 If the time threshold is greater than the time interval, the data is judged to be continuous;

[0053] Furthermore, the specific steps of determining whether the storm cells are in the same weather process based on their characteristics and distances in S3 include:

[0054] S301 calculates the mean value data of the single scan intensity by using a feature similarity algorithm;

[0055] S302 calculates the monomer distance using a distance priority algorithm;

[0056] S303 determines whether the calculated storm cells belong to the same weather process;

[0057] Further, the S301 uses a feature similarity algorithm to calculate the single-cell scan intensity mean data using the difference between the storm single-cell intensity means and the ratio of the area;

[0058] The method of the present invention takes the data samples of big data as the main research object, and uses the machine learning algorithm to process the data samples to realize the effectiveness tracking and forecasting of storm cells; firstly, the most representative hail and rainstorms in 75 weather processes are selected as the research objects from a large number of data samples. Generally speaking, for the forecast of storm cells in the weather process, the tracking of cells in the mid- and early-stage body scans is the basis of fitting and extrapolation forecasting, so we must first track the cells of the same weather process;

[0059] Then, according to the volume scan interval algorithm, since the time interval for acquiring monomer data is 6 minutes, the volume scan time interval of the two monomers before and after is checked, and the time threshold is used to compare the size. If the threshold is less than the time interval, the data is discontinuous this time, and no time correlation analysis is performed on it. On the contrary, the two monomers are considered to be continuous. Then, using the feature similarity algorithm, the difference in the mean intensity of the monomers and the ratio of the areas are used. Through statistical analysis of a large number of monomers, it is found that in the same weather process, the size change of the mean intensity of the volume scans of the front and rear monomers is less than 5dBZ, and the size of the area ratio does not exceed 40%. This shows that the front and rear monomers meet the feature similarity algorithm and can be considered to be in the same weather process.

[0060] Finally, through the short-distance priority algorithm, the distance that any object moves within a certain period of time is limited, and the same is true for monomers. In the same weather process, when the time is 6 minutes, the movement distance is less than 20 kilometers, it can be determined that the two monomers are in the same weather process. In this way, the tracking and matching of monomers in a time series is completed through the above three algorithms; when this time series meets the number of 5 monomers, we can fit the movement trajectory of this time series. Statistical analysis of multiple time series samples found that in the same weather process, within a certain period of time, the monomer moves in a straight line;

[0061] Therefore, the weighted least squares method is used to fit the centroids of the monomers in the same time series to obtain a fitted straight line. After obtaining the fitted straight line, the straight line is used for extrapolation forecasting. We use the positions of 5 monomers in a time series to obtain the speed and direction of the monomer movement; according to the speed multiplied by time, the position of the subsequent monomer can be predicted, thus realizing the tracking and forecasting of storm monomers;

[0062] The method of the present invention uses a machine learning algorithm to address the problems of large forecast errors and a small number of predictable monomers. First, representative data is obtained from a large amount of data. The embodiment of the method of the present invention can select the most representative hail and rainstorms in 75 weather processes in a city from 2013 to 2018 as research objects; then, the tracking and matching of monomers in a time series are achieved according to the three algorithms of "volume scanning interval, feature similarity, and close distance priority", and finally the weighted least squares method is used to predict the position of the storm monomer at the next moment; from the evaluation results, the method can predict more monomers, and the average forecast error of the obtained monomer position is smaller, which can better realize the tracking and forecasting of storm monomers.

[0063] Embodiment 2:

[0064] A storm monomer tracking and forecasting system based on machine learning, the system specifically includes a data sampling module, a continuous judgment module, a weather judgment module, a trajectory fitting module and a position prediction module:

[0065] Data sampling module: sampling through data samples of the same weather;

[0066] Continuity judgment module: Use the volume scanning interval algorithm to obtain the previous and next storm monomer data to determine whether it is continuous;

[0067] Weather judgment module: judge whether they are in the same weather process based on the characteristics and distance of the storm cell;

[0068] Trajectory fitting module: uses the storm monomer tracking and matching within a time series to fit the motion trajectory of the time series;

[0069] Position forecast module: Use the weighted least squares method to fit the straight line to predict the position of the storm cell at the next moment;

[0070] Furthermore, the continuity judgment module specifically includes an interval checking module, an interval comparing module, an interval judgment module A and an interval judgment module B:

[0071] Interval check module: Check the time interval between the two storm cells;

[0072] Interval comparison module: compares time threshold and time interval;

[0073] Interval judgment module A: if the time threshold is less than the time interval, the data is judged to be discontinuous;

[0074] Interval judgment module B: if the time threshold is greater than the time interval, the data is judged to be continuous;

[0075] Furthermore, the weather determination module specifically includes an intensity calculation module, a distance calculation module and a weather determination module:

[0076] Intensity calculation module: calculates the mean intensity data of the single scan using the feature similarity algorithm;

[0077] Spacing calculation module: calculates the monomer spacing using the distance priority algorithm;

[0078] Weather determination module: determines whether the calculated storm cells belong to the same weather process;

[0079] Furthermore, the intensity calculation module uses a feature similarity algorithm to calculate the single-cell scan intensity mean data using the difference between the storm single-cell intensity means and the ratio of the area.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A storm cell tracking and forecasting method based on machine learning, characterized by The specific steps of the method include: S1 is sampled through data samples of the same weather; S2 uses the volume scanning interval algorithm to obtain the data of the previous and next storm cells to determine whether they are continuous; S3 determines whether they are in the same weather process based on the characteristics and distance of the storm cells; S4 uses the tracking and matching of storm cells within a time series to fit the motion trajectory of the time series; S5 uses the weighted least squares method to fit the straight line to predict the location of the storm cell at the next moment; The specific steps of S2 using the volume scanning interval algorithm to obtain the front and rear storm monomer data and determine whether they are continuous include: S201 checks the time interval between the body scans of the two preceding and following storm cells; S202 compares the time threshold and the time interval; S203: if the time threshold is less than the time interval, the data is judged to be discontinuous; S213 If the time threshold is greater than the time interval, the data is judged to be continuous; The specific steps of S3 judging whether the storm cells are in the same weather process by the characteristics and distance of the storm cells include: S301 calculates the mean value data of the single scan intensity by using a feature similarity algorithm; S302 calculates the monomer distance using a distance priority algorithm; S303 determines whether the calculated storm cells belong to the same weather process; According to the volume scan interval algorithm, since the time interval for acquiring monomer data is 6 minutes, the volume scan time interval of the two monomers before and after is checked, and the time threshold is used to compare the size. When the threshold is less than the time interval, the data is discontinuous this time, and no time correlation analysis is performed on it. On the contrary, the two monomers are considered to be continuous. Using the feature similarity algorithm, the difference in the mean intensity of the monomers and the ratio of the areas are used. Through statistical analysis of a large number of monomers, it is found that in the same weather process, the size change of the mean intensity of the volume scans of the front and rear monomers is less than 5dBZ, and the size of the area ratio does not exceed 40%. The front and rear monomers meet the feature similarity algorithm and are considered to be in the same weather process. Through the short-distance priority algorithm, in the same weather process, when the time is 6 minutes, the movement distance is less than 20 kilometers. The above three algorithms can complete the tracking and matching of the monomer in a time series; when this time series meets the number of 5 monomers, the movement trajectory of this time series can be fitted; statistical analysis of multiple time series samples found that in the same weather process, within a certain period of time, the monomer moves in a straight line.

2. The method according to claim 1, characterized in that The S301 uses a feature similarity algorithm to calculate the single-cell scan intensity mean data using the difference in the storm single-cell intensity mean and the ratio of the area.

3. A system using the storm monomer tracking and forecasting method based on machine learning according to claim 1, characterized in that The system specifically includes a data sampling module, a continuous judgment module, a weather judgment module, a trajectory fitting module and a location forecast module: Data sampling module: sampling through data samples of the same weather; Continuity judgment module: Use the volume scanning interval algorithm to obtain the previous and next storm monomer data to determine whether it is continuous; Weather judgment module: judge whether they are in the same weather process based on the characteristics and distance of the storm cell; Trajectory fitting module: uses the storm monomer tracking and matching within a time series to fit the motion trajectory of the time series; Position forecast module: Use the weighted least squares method to fit the straight line to predict the position of the storm cell at the next moment; The continuous judgment module specifically includes an interval checking module, an interval comparison module, an interval judgment module A and an interval judgment module B: Interval check module: Check the time interval between the two storm cells; Interval comparison module: compares time threshold and time interval; Interval judgment module A: if the time threshold is less than the time interval, the data is judged to be discontinuous; Interval judgment module B: if the time threshold is greater than the time interval, the data is judged to be continuous; The weather judgment module specifically includes an intensity calculation module, a distance calculation module and a weather judgment module: Intensity calculation module: calculates the mean intensity data of the single scan using the feature similarity algorithm; Spacing calculation module: calculates the monomer spacing using the distance priority algorithm; Weather determination module: determines whether the calculated storm cells belong to the same weather process.

4. The system according to claim 3, characterized in that The intensity calculation module uses a feature similarity algorithm to calculate the storm cell intensity mean data using the difference in storm cell intensity mean values ​​and the ratio of the area.

Images