Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

A Low Complexity Algorithm for Earthquake

Detection System
Meng-Yun Hsu* and Shiann-Tsong Sheu+
Department of Communication Engineering, National Central University
*
mandyhsu1995@gmail.com, +stsheu@ce.ncu.edu.tw

Abstract ─ Predicting occurrence of earthquakes is very determine the motion type. Thus, in order to shorten decision
difficult today. Instead of, the contemporary Earthquake Early latency, this paper will propose a low complexity algorithm
Warning System (EEWS) only relies on the earthquake detection
technology. Deploying more stations to sense seismic wave could meanwhile keeping the low probability of issuing false alarms.
obtain higher accuracy of earthquake detection but requiring III. LOW COMPLEXITY EARTHQUAKE DETECTION
longer detection latency. This paper proposes a low complexity ALGORITHM
algorithm for fast detecting earthquake events from a continuously
series of sensed acceleration information. Simulation results show Generally, the wobble appearing in collected z-axis
that about 90% and 93.5% real earthquake samples could be acceleration information (denoted as data) may be incurred
detected within 1s and 2s respectively
from device itself, human user, or earthquake. Observations
I. INTRODUCTION revealed that the differences among three motion types are
two-fold: 1) the increase of magnitude and 2) long-duration
The movement of earth plates often causes the seismic movement. It implies that the standard deviation would be a
activities, which may also incur the loss of lives and properties. good tool to observe the dispersion degree among data. For
To minimize damages, the Earthquake Early Warning System example, a real earthquake example is shown in Fig. 1 and the
(EEWS) has been deployed to alert people by means of distribution of observed data during the first 5 seconds in this
detecting earthquake events from raw information collected example is shown in Fig. 2.
from acceleration sensors. The seismic wave mainly falls into For eliminating the effects from ripple and short-term
two categories. One travels along the surface of the earth is movement, the series data is transformed as a new series data’
called surface wave, the other travels through the interior of by utilizing exponentially weighted moving average (EWMA)
the earth is called body wave. The body wave includes scheme, where weighting coefficient is set as 1/8. We have
primary wave (P-wave, which is longitudinal) and secondary
data'(t) = 7/8×data'(t-1) + 1/8×data(t).
wave (S-wave, which is transverse). Usually, the P-wave
travels faster than the S-wave, while the S-wave is more It is worthwhile noting that the shape of data’ is smoother
destructive than the P-wave. Thus, the main idea of the EEWS than the original data and the division operation could be
is that if we can detect the P-wave in advance, the alert easily accomplished by bit shifting operation. For earthquake
message, which is broadcast in nearly light speed on wired or decision, we use sliding window W to find out the extreme
wireless media, should reach users before S-wave. However, values (i.e., both peak and valley points) [4] from series data’.
the issue would be the latency of determining the occurrence These extreme values are important because they represent the
of P-wave. A longer decision delay will result in less chance occurrences of significant changing on input samples, as
to alert people in advance. Thus, this paper aims to propose an shown in Fig. 3. To do statistics, we need another queue-based
algorithm to fast and correctly find out the earthquake events sliding window, denoted as E, to record these detected
by means of observing the vertical (i.e., z-axis) acceleration extreme values. From our observations, the window sizes of W
information collected from devices with G-sensors. and E are set as N and M samples respectively. Whenever
detecting a new extreme value, say x, from window W, it is
II. RELATED WORKS inserted into window E (i.e., advancing window E by one
There are many smart technologies [1][2], such as Ratio of sample) and the mean μ, standard deviation σ and variance (σ2)
Vertical power to Total power (RV2T), Short Term Averaging of these extreme values in window E are calculated.
over Long Term Averaging (STA/LTA), Support vector Recall a lower standard deviation means that most of the
machine (SVM) and so on, have been proposed for values are very close to the average and the standard
determining the earthquake while avoiding false alarms. deviation is the square root of the variance. In order to
Intuitively, a longer observation window should obtain a determine whether the window E contains significant
higher accuracy of earthquake detection. For example, in order vibration or not, the variance of window E is used as the
to let smartphones participate EEWS, authors in [3] have weight. If the average weight of the last M windows is beyond
developed an artificial neural network approach to identify the a specified threshold, denoted as TH, we recognize there is an
different characteristics of earthquake and human motions. earthquake event. The proposed algorithm is listed as
The proposed algorithm assesses 2s windows of data to following:
Algorithm ACKNOWLEDGEMENT
1: Let set X={x1, x2,…, xN} denote the last N input z-axis
This research was supported by the National Science
acceleration samples.
Council of Taiwan under grant 104-2221-E-008-042 for
2: Let set W={w1, w2,…, wN} denote the last N values which
funding this work.
have been smoothed by EWMA scheme.
3: Let set E={e1, e2,…, eM} denote the last M extreme values. REFERENCES
4: Let set V={v1, v2,…, vM} denote the last M variances in set E. [1] Y. Sherki, N. Gaikwad, and J. Chandle, A. Kulkarni, ”Design of Real
5: loop Time Sensor System for Detection and Processing of Seismic Waves for
6: get new sample x of z-axis acceleration from G-sensor; Earthquake Early Warning System”, Proc. of IEEE ICPACE, pp. 285 -
289, Bangalore, Aug. 2015.
7: insert x into set X and remove the oldest one from set X; [2] A. Zollo, O. Amoroso, M. Lancieri, Y.-M. Wu and H. Kanamori, ”A
8: calculate the moving average ω = 7/8×wN + 1/8×x; threshold-based earthquake early warning using dense accelerometer
9: insert ω into set W and remove the oldest one from set W; networks”, Geophysical Journal International, Vol. 183, No. 2, pp.963-
974, 2010.
10: if (ω is an extreme point in set W) then [3] Q. Kong, Richard M. Allen, Louis Schreier, Y.-W. Kwon, ”MyShake: A
11: insert ω into set E and remove the oldest one from set E; smartphone seismic network for earthquake early warning and beyond”,
12: calculate the variance ν of set E; Science Advances, Vol. 2, No.2, 2016.
[4] T. Z. Redhwan, M. Chowdhury and H. a. Rahman, ”A Neyman-Pearson
13: insert ν into set V; Approach to the Development of Low Cost Earthquake Detection and
14: calculate the mean value k of set V; Damage Mitigation System using Sensor Fusion”, Procs. Of IEEE
15: if (k > threshold) then International conference on ICECS, pp.191-194, Marseille, Dec. 2014.
16: issue alarm;
17: end if
21: end if
22: forever

IV. SIMULATIONS Fig. 1. Z-axis acceleration values in an earthquake sample.

For evaluating the feasibility of proposed algorithm, we test


293 earthquake cases recorded in Taiwan Central Weather
Bureau. The common sampling rate of these cases is 200Hz
and the gal (cm/s2) is used as unit. From our carefully
observations, we set N=50, M=25 and TH=0.003 gal2. Fig. 2. The distribution of z-axis acceleration values in the first 5 seconds.
For each case, we extract the z-axis acceleration
information as input data and transform it as data’ for
detecting earthquake. Taking Fig. 1 for example again, Fig. 4
shows how the original series of data is transformed into new
series of smoothed data’ during the period from 15s to 16.4s.
Furthermore, Fig. 5 portrays the time instance (at 16.7s) that
the proposed algorithm detects the earthquake event.
Table 1 summaries the overall simulation results of
proposed algorithm. We can easily find that almost 90% and Fig. 3. Illustration of retrieving extreme values from a series of input data.
93.5% real earthquake cases could be detected by proposed
low complexity algorithm within 1s and 2s respectively

V. DISCUSSION
This paper proposed a low complexity algorithm for Fig. 4. The original data (blue line) and smoothed data’(green line) during the
detecting the earthquake event from a series of collected z- period from 15s to 16.4s in the example shown in Fig. 1.
axis acceleration information. The exponentially weighted
moving average (EWMA) scheme is first used to filter short-
term movements and then all extreme points are extracted for
doing statistics in order to measure the degree of variances.
Simulation results reveal that as the average variance is Fig. 5. The red line indicates the time instance of detecting earthquake event.
beyond a certain threshold, it is highly possible that there is an Table 1. Performance of proposed algorithm.
earthquake event. detection within 1s detection within 2s
Correctness
90.1% (264/293) 93.5% (274/293)
Probability

You might also like