A thunderstorm time window capture method based on fuzzy C-means clustering

Abstract

The present invention discloses a method for capturing a thunderstorm time window based on fuzzy C-means clustering. The method first obtains time-series atmospheric electric field related attribute information and constructs a data set that characterizes the weather. Secondly, the data set is clustered using a fuzzy C-means algorithm to obtain a fuzzy set, and then the optimal granularity is determined to obtain the optimal information granule. Finally, the optimal information granule is cut to capture the time start and end points of the granule elements in the information granule representing the thunderstorm dominant obtained by cutting, and the start and end points correspond to the thunderstorm time window. The present invention fully considers the fuzzy characteristics of the atmospheric electric field signal, ensures the effectiveness of the information granules that reflect the weather conditions, improves the reliability of thunderstorm monitoring, and promotes accurate thunderstorm monitoring and tracking.

Description

Technical Field

[0001] The invention belongs to the field of lightning detection, and in particular relates to a thunderstorm time window capturing method based on fuzzy C-means clustering. Background Art

[0002] There is always an atmospheric electric field in the atmosphere. The three-dimensional atmospheric electric field measured by the three-dimensional atmospheric electric field instrument contains two horizontal electric fields Ex and Ey and one vertical electric field Ez. The value of the atmospheric electric field on a sunny day is small, especially the electric field in the horizontal direction can be ignored. However, the value and direction of the atmospheric electric field during a thunderstorm have obvious irregular changes. The strong correlation between the atmospheric electric field and weather changes, especially thunderstorms, has prompted a series of thunderstorm monitoring studies based on the atmospheric electric field, which has important practical significance for lightning protection and disaster reduction under the background of global extreme climate.

[0003] Atmospheric electric field signals carry multiple attributes such as time, statistics (mean, standard deviation, threshold), and rate of change. These attributes and related methods such as polarity reversal, fast-changing jitter, and threshold warning have prompted the study of atmospheric electric fields in the time domain. However, different attributes have different characterization capabilities for weather, which greatly affects the accurate assessment of the occurrence and end time of thunderstorms. In addition, there are many methods for dividing the time of atmospheric electric field signals, and the conventional method is equal-time division. However, this division is difficult to meet the needs of accurate thunderstorm monitoring. For example, a thunderstorm occurs at 8:13 in the morning. If the interval of equal-time division is 10 minutes, then the ideal situation is to monitor thunderstorm activity at 8:10, but not at 8:13. If this process is regarded as an information granule, the main reason for the inability to monitor is that the granularity of the information granule does not reach the minute level. In short, the static thunderstorm development time assessment brought about by fixed division is difficult to meet the accuracy requirements of thunderstorm monitoring and tracking. Summary of the invention

[0004] The purpose of the present invention is to address the shortcomings of the existing technology, explore the fuzzy weather attributes carried by the atmospheric electric field, accurately estimate the time window for thunderstorm occurrence, propose a thunderstorm time window capture method based on fuzzy C-means clustering, and apply it to lightning detection to further improve the lightning detection capability.

[0005] The technical solution of the present invention is to provide a thunderstorm time window capture method based on fuzzy C-means clustering, comprising the following steps:

[0006] Step 1: Obtain the time-series atmospheric electric field related attribute information and construct a data set representing the weather.

[0007] The time-series atmospheric electric field related attribute information specifically includes: the time-series atmospheric electric field signal and its first-order deviation; the mean, standard deviation, and threshold of the time-series atmospheric electric field signal; and time attributes.

[0008] Step 2: Apply the fuzzy C-means clustering algorithm to the data set to obtain a fuzzy set, and then determine the optimal granularity to obtain the optimal information granule.

[0009] Step 3: Use the optimal information particle The cutting algorithm captures the time start and end points of the particle elements in the information particles representing the thunderstorm dominance obtained by cutting, and the start and end points correspond to the thunderstorm time window.

[0010] Beneficial effects of the present invention:

[0011] 1. The method of the present invention utilizes fuzzy characteristics to construct information granules to reflect weather conditions, and fully considers the fuzzy characteristics of atmospheric electric field signals.

[0012] 2. The present invention constructs an optimal data set based on atmospheric electric field signals, ensures the validity of information particles reflecting weather conditions, and improves the reliability of thunderstorm monitoring.

[0013] 3. The thunderstorm time window dynamic capture method adopted by the present invention has obvious advantages over the thunderstorm activity assessment after isochronous atmospheric electric field signal division, and promotes accurate thunderstorm monitoring and tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the thunderstorm time window capturing method based on fuzzy C-means clustering of the present invention;

[0015] Figure 2 is the Ez data in four events;

[0016] Figure 3 Statistical property results for visualizing numerical differences;

[0017] Figure 4(a) shows the distribution of Ez and its first-order deviation on April 17;

[0018] Figure 4(b) shows the distribution of Ez and its first-order deviation on April 18;

[0019] Figure 4(c) shows the distribution of Ez and its first-order deviation on April 4;

[0020] Figure 4(d) shows the distribution of Ez and its first-order deviation on October 9;

[0021] Figure 5 It is the Ez data of two thunderstorm events;

[0022] Figure 6(a) shows information granules of different granularities under event 1;

[0023] Figure 6(b) shows information granules of different granularities under event 2. DETAILED DESCRIPTION

[0024] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0025] Embodiment 1:

[0026] The present invention provides a thunderstorm time window capture method based on fuzzy C-means clustering, the specific process is as follows: Figure 1 As shown, the following steps are included:

[0027] (1) Obtain the relevant attribute information of the time series atmospheric electric field and construct a data set representing the weather. Verify the differences in the attributes of the atmospheric electric field signal that can reflect different weather conditions, construct the attributes into a data set, and input it into the fuzzy C-means clustering algorithm, which is as follows:

[0028] The mean, standard deviation and threshold of the atmospheric electric field signal are often selected as statistical attributes for thunderstorm detection. At the same time, the time series signal and its first-order deviation can intuitively reflect its changes, which is also true for the atmospheric electric field signal. Before considering the data set consisting of the five attributes of mean, standard deviation, threshold, time series electric field signal and its first-order deviation plus a time attribute as the input of fuzzy C-means clustering, the attribute differences under different weather conditions were tested. This difference leads to differences in the temporal membership and information granularity after clustering. The atmospheric electric field events of two sunny days (18:30 to 20:30 on April 17 and April 18) and two thunderstorms (18:30 to 20:30 on April 4 and October 9) measured at the Nanjing University of Information Science and Technology Station (NUIST Station) in 2018 were selected (see Figure 2 ) to visualize the numerical differences of these attributes. Figure 3 Numerical results are presented at 30-min intervals (time windows numbered from 1 to 4). Note that the atmospheric electric field consists of horizontal electric field components Ex, Ey and a vertical electric field component Ez.

[0029] Figure 2 The change of Ez on a sunny day is relatively stable with less fluctuations, while the Ez driven by a thunderstorm changes dramatically, with the maximum amplitude exceeding 15kV / m. Figure 3 The attribute value of sunny weather in is much smaller than that of thunderstorm weather, showing a significant difference. Taking the threshold as an example, the maximum threshold of sunny weather is only 2.652 kV / m, while the minimum threshold of thunderstorm is as high as 6.434 kV / m. Further, suppose the time series atmospheric electric field signal is , and its first-order difference time series is ,So:

[0030] (1)

[0031] Where n is the length of the time series atmospheric electric field signal, and .Will and Doing the Cartesian product forms a feature space ( , ). Figure 4(a) to Figure 4(b) The distribution of the time series Ez and its first-order deviation in this space under each event is presented.

[0032] Figure 4(a) and 4(b) The electric field value (horizontal axis) and the first-order difference (vertical axis) in the equation are in the same small order of magnitude. Figure 4(c) and 4(d) The range of the electric field value and the first-order difference in Figure 4(a) and 4(b) The obvious magnitude difference in different weather conditions supports the feasibility of using these two attributes in dataset construction. Figures 2 to 3 The analysis determined that the data set consisting of the atmospheric electric field signal itself, the first-order deviation, mean, standard deviation, threshold and time of the atmospheric electric field signal was used as the input set of fuzzy C-means clustering to obtain fuzzy sets, temporal membership and information granules. Each attribute was normalized to [0,1] mapping to ensure that the numerical magnitudes of different attributes were the same.

[0033] ② Construct information particles that can reflect weather conditions and determine the optimal granularity of information particles, which are as follows:

[0034] Weather is fuzzy. In other words, whether it is sunny or thunderstormy is fuzzy. In fact, thunderstorm means that thunderstorm dominates the weather, and non-thunderstorm is less dominant than thunderstorm. Similarly, non-thunderstorm weather means non-thunderstorm dominates the weather, and thunderstorm is less dominant than non-thunderstorm. The weather attributes (the dominance of thunderstorms and non-thunderstorms) carried by the atmospheric electric field signal are also fuzzy (i.e., membership). The fuzzy C-means clustering algorithm has the ability to convert atmospheric electric field attributes into fuzzy sets and use multiple memberships to characterize multiple dominant weather information. The number of categories, maximum number of iterations, fuzziness coefficient, and threshold in fuzzy C-means clustering are set to 2, 100, 2, and 10, respectively. .

[0035] Assume that the domain X = {non-thunderstorm dominated, thunderstorm dominated}, fuzzy set A t represents the weather conditions in time period t. Among them, non-thunderstorm dominance µ1 represents the degree of dominance of non-thunderstorm weather in this time period (membership degree, called A t (1) = MD1), thunderstorm dominant µ2 is similar to it, called A t(2) = MD2. Different from conventional membership, the present invention inputs the data set of time series electric field signals into fuzzy C-means clustering to obtain the time series membership group {MD1, MD2} in the context of all-weather thunderstorm monitoring. The fuzzy set composed of MD1 and MD2 constitutes an information particle used to describe the weather at time t.

[0036] The granularity of information is crucial to the description of the thunderstorm monitoring process and related problems, as well as the solutions to the problems. An optimal granularity helps to focus on the details of thunderstorm-related problems at the most appropriate level. The atmospheric electric field data measured in different regions have different characteristics. The optimal granularity needs to be determined based on the atmospheric electric field measured in a specific region. Two typical events of thunderstorms from scratch with known thunderstorm occurrence time information recorded at the NUIST station are used to test the time deviation at different granularities. Figure 5 As shown in the figure, the first event (event 1) occurred from 12:30 to 14:30 on June 1, 2018, with a thunderstorm occurring around 13:59 (corresponding to the 5340th sample point) (marked by the dashed line on the right). The second event (event 2) occurred from 10:30 to 12:30 on August 17, 2018, with a thunderstorm occurring around 11:03 (corresponding to the 1980th sample point) (marked by the dashed line on the left).

[0037] The granularity is set to 1min, 3min, 5min, 10min, and 30min, and the information granules shown in Figure 6(a) and Figure 6(b) are obtained. The dotted line in Figure 6(a) corresponds to a membership value of 0.5. The time points of the thunderstorm occurrence of event 1 corresponding to the granularity of 1min, 3min, 5min, 10min, and 30min are respectively located near the 89th minute (5340 / 60), the 30th minute (5340 / 180), the 18th minute (5340 / 300), the 9th minute (5340 / 600), and the 3rd minute (5340 / 1800). Similarly, in Figure 6(b), the time points of event 2 are respectively located near the 33rd minute, the 11th minute, the 7th minute, the 4th minute, and the 2nd minute.

[0038] Table 1 Deviation rate between measured time points and standard time points

[0039]

[0040] Tests on labeled samples show that when the fuzziness factor is equal to 2, a membership value of 0.5 (also applicable to ) better matches the transition from non-thunderstorm to thunderstorm-dominated. The transition from the absence of thunderstorms corresponds to the transition of the temporal membership from no more than 0.5 to greater than 0.5. The window number corresponding to the first occurrence of this transition in Figures 6(a) and 6(b) is taken as the time point of the thunderstorm occurrence (the membership point needs to be above the dotted line). As shown in Table 1, the measured time points of the particle elements With standard time point Deviation rate , determine the optimal granularity and the minimum deviation rate Corresponding to the optimal information granularity, deviation rate The calculation formula is as follows:

[0041] (2)

[0042] In Table 1, for event 1, the lowest DR of 11.11% was achieved at both 5-min and 10-min granularities. For event 2, the lowest DR of 0% was achieved at both 10-min and 30-min granularities. Although the 30-min granularity achieved the most competitive DR in event 2, it achieved the maximum DR of 33.33% in event 1. The main reason for the extreme deviation and fluctuation is that the 30-min granularity is too large, considering that a thunderstorm usually lasts about 2 hours. Larger granularity makes it difficult to accurately track the ever-changing weather. In addition, smaller granularity will be disturbed by local values ​​caused by excessive refinement. The resulting mixed unreliable information and objects lead to an unstable DR. In summary, the 10-min granularity is determined to be the optimal granularity.

[0043] ③ Propose a The thunderstorm time window capture method is as follows:

[0044] As an important defuzzification tool in the fuzzy rule system, it is used to capture and lock the time start and end points (i.e., time window) of the information particles Aw that represent the dominant particles of thunderstorms. The value is 0.5 The tangent (dashed line corresponding to the value of 0.5) represents the thunderstorm-dominated particle elements located above the tangent.

[0045] Usually, the starting and ending points of each particle element area corresponding to time t are the possible thunderstorm windows captured. However, the large fluctuation of atmospheric electric field signals driven by thunderstorms is very likely to cause a jump in the temporal membership. Based on the principle of maximizing the time window, Table 2 lists the main weather conditions and the corresponding information particle legends and capture results for reference.

[0046] Table 2 Capture results with information granule legend

[0047]

[0048] NT-T, T-NT, and NT-T-NT in Table 2 represent the transition from non-thunderstorm to thunderstorm, from thunderstorm to non-thunderstorm, from non-thunderstorm to thunderstorm and then to non-thunderstorm. PT and PNT represent pure thunderstorm dominance (the particle elements of the thunderstorm dominance part are all above the tangent) and pure non-thunderstorm dominance (the particle elements of the non-thunderstorm dominance part are all below the tangent). TCNT and NTCT represent thunderstorm dominance including non-thunderstorm and non-thunderstorm dominance including thunderstorm.

Claims

1. A thunderstorm time window capture method based on fuzzy C-means clustering, characterized in that: The following steps are involved: Step 1: Obtain the time-series atmospheric electric field related attribute information and construct a data set representing the weather; The time series atmospheric electric field related attribute information specifically includes: time series atmospheric electric field signal and its first-order deviation; mean, standard deviation, threshold of time series atmospheric electric field signal; time attribute; Step 2: Use the fuzzy C-means clustering algorithm to cluster the data set to obtain a fuzzy set, and then determine the optimal granularity to obtain the optimal information granule; The specific process of obtaining the fuzzy set by using the fuzzy C-means clustering algorithm is as follows: Let the domain X = {non-thunderstorm-dominated, thunderstorm-dominated}, the fuzzy set A t represents the weather conditions in time period t; among them, non-thunderstorm dominance represents the degree of dominance of non-thunderstorm weather in this time period, that is, the degree of membership, called A t (1) = MD1; thunderstorm dominant, called A t (2) = MD2; the data set of the time series electric field signal is input into the fuzzy C-means clustering to obtain the time series membership group {MD1, MD2}. The fuzzy set composed of MD1 and MD2 constitutes an information particle, describing the weather at time t; The specific operation of determining the optimal granularity to obtain the optimal information granule is: determining the optimal granularity through the deviation rate DR between the actual measurement time point MTP of the granule element and the standard time point STP, and the minimum deviation rate DR corresponds to the optimal information granularity; The calculation formula of the deviation rate DR is as follows: Step 3: Use the α-cut algorithm for the optimal information granule to capture the time start and end points of the granule elements in the information granule representing the thunderstorm dominant obtained by cutting, and the start and end points correspond to the thunderstorm time window.

Images