Entry Name: PKU-Jiang-MC2

VAST Challenge 2017
Mini-Challenge 2

Team Members:

Ruike Jiang, Peking University, jiangrk.pku@gmail.com PRIMARY

Wei Huang, QIHOO 360, huangweigrace03@gmail.com

Nan Ma, Peking University, 616012777@qq.com

Hong Fan, Peking University, fan.hong@pku.edu.cn

Ying Zhao, Central South University, zhaoying@csu.edu.cn

Xiaoru Yuan, Peking University, xiaoru.yuan@pku.edu.cn

Student Team: No

Tools Used:

D3

Visual analytic system developed by our team.

Approximately how many hours were spent working on this submission in total?

200 hours

May we post your submission in the Visual Analytics Benchmark Repository after VAST Challenge 2017 is complete? YES

Video

http://vis.pku.edu.cn/vast2017-mc2.wmv

Questions

MC2.1 – Characterize the sensors’ performance and operation. Are they all working properly at all times? Can you detect any unexpected behaviors of the sensors through analyzing the readings they capture? Limit your response to no more than 9 images and 1000 words.

The sensors are not always working properly. We found that there exist data missing, data redundancy and data exception in the readings that sensors captured, listed as follows:

1. Data Missing

The data missing phenomenon is very interesting, we summed up three modes: a). periodicity, b). globality for all sensors and all chemicals, and c). locality for partial sensors and partial chemicals.

a. Periodicity, the data missing phenomenon periodicity occurred at the same point of per month. For example, at 2/4 00:00, 2/8 00:00, and 2/12 00:00, all chemical readings captured by all sensors are missing, as shown in Figure 1.

Fig. 1 Periodicity data missing, at 2/4 00:00, 2/8 00:00, and 2/12 00:00, all chemical readings captured by all sensors are missing.

b. Globality, the readings of all chemicals captured by all sensors are missing at some time, such as 00:00 on April 6, 00:00 on August 4, 00:00 on August 7 and 00:00 on December 7 shown in Fig.2.

Fig. 2 Globality data missing. All chemicals captured by all sensors are missing at some time, such as 6/4/2016 00:00.

c. Locality, at some point, the readings of a certain chemical captured by all sensors are missing.  

Fig. 3 Locality data missing. At some time the readings of some chemical captured by some sensors are missing.

2. Data redundancy

There also exist data redundancy phenomenon in the sensor readings, that is, at a certain moment, there appeared multiple data records in the readings of a same chemical captured by one sensor. We further found that all data redundant are appeared when sensors capture chemical AGOC-3A, and the repetition number is twice. The statistical data of redundant that each sensor captures chemical AGOC-3A are shown as Fig 4.

Fig 4. Data redundancy statistics, all data redundant are appeared when sensors capture chemical AGOC-3A, and the repetition number is twice. Sensor 1, 2, 7, 8 have much less number of redundant readings than other 4 sensors.

3. Abnormal Patterns of Readings

a. Sensor 4: its readings show the staircase shape in April, August, and December.

Original data is on the left; y-axis’ range is related to the sub-graph's maximum value. Reading distribution view is on the right. To illustrate the result, we select a threshold, values larger than threshold will turn red in this view. In August, 55% readings larger than the threshold. In April, each reading smaller than the threshold. In December, each reading larger than the threshold. As we can see, from April to August, the baseline of readings increases apparently.

Fig. 5

b. Maximum values:

April: most maximums values of all 4 chemicals presented on Sensor 6

August:

Appluimonia: Sensor 6

AGOC-3A: Sensor 9

Methylosmolene: Sensor 3

Chlorodinine: Sensor 2

December:

Appluimonia: Sensor 9

Others: Sensor 6

c. Large readings on certain sensors

Readings on Sensor 3 are mostly large, and the maximum and minimum values on it are very close. Same behavior exists on Sensor 7. Other sensors usually have significantly differences between peak values and non-peak values.

For example, selecting the same threshold to compare readings of Appluimonia in April, Sensor 3 and Sensor 6. Since Sensor 6 is located between 4 factories and the maximum value appears on it, but it non-peak is smaller than the Sensor 3, so we treat it as anomaly. The reason of the abnormal maybe the broken of Sensor 3, geographical environment there or other factors.

Fig. 6

MC2.2 – Now turn your attention to the chemicals themselves. Which chemicals are being detected by the sensor group? What patterns of chemical releases do you see, as being reported in the data? Limit your response to no more than 6 images and 500 words.

All four chemicals are being detected.

We observed the following patterns:

1. Temporal distribution in each day

Peaks of Methylosmolene usually exist at night (22:00 - 5:00). (Not considering data missing)

Fig. 7

For all chemicals, there is no correlation observed between peaks and weekdays/weekends.

Fig. 8

2. Influences of wind speed and directions

From the readings, we hypothesize that the chemicals reside in the air for a long time, and advect with the wind. It can be observed that multiples sensors show peaks values at different time, which has very close relations with the wind direction.

For Appluimonia, we observe that when the wind speed is low and stable, it diffuse with the wind direction to wide area of regions:

2016/4/16, 9:00-12:00, peaks are found on Sensor 1, 6, 7, and 8, with current wind direction is north-east 0.2m/s-0.9m/s.

In the following picture, red lines mean this reading is higher than the threshold user chooses, while the blue line indicates time 9:00. April 16, 2016.

Fig. 9

2016/4/29, 4:00-9:00, peaks are found on Sensor 7 when the wind direction changes from north-east to north. Then wind direction changes back to north, and sensor 6 shows peaks. Wind speed is always low.

2016/4/17 3:00-5:00, peaks are found on Sensor 5,6, and 7, and at 9:00 on Sensor 9. The wind direction is south-east from 3:00-6:00, and changes to south at 6:00-9:00.

The facts above support the release pattern of Appluimonia we recognize.

In the following figure, the blue line indicates time 0:00. April 17, 2016.

Fig. 10

Similar pattern exists for AGOC-3A: when wind speed is low, it will diffuse with the wind direction with a certain angle; while when the wind speed is high, it will be blown away.

2016/4/6 6:00-7:00, peaks are shown on Sensor 6, then on Sensor 5 at 13:00, 14:00 on Sensor 9, with wind direction changing from west to south-west. When the wind speed increases, no peaks detected by sensors in that day.

In the following figure, the blue line indicates time 5:00, April 6, 2016.

Fig. 11

3. Influences of distances to the factories

Usually, sensors closer to the factories have larger readings, as well as larger peaks, while distant sensors have smaller readings.

MC2.3 – Which factories are responsible for which chemical releases? Carefully describe how you determined this using all the data you have available. For the factories you identified, describe any observed patterns of operation revealed in the data. Limit your response to no more than 8 images and 1000 words.

Through analyzing, the relationship between factories and chemical releases are listed as follows.

• Radiance: [ AGOC-3A ]
• Roadrunner: [ Chlorodinine ]
• Indigo: [ Appluimonia ]
• Kasios: [ Chlorodinine, Methylosmolene ]

Our visual analysis system is shown in the Fig. 12, including a). Calendar View, b). ISOMap View, c). Projection View and d). Sensor Read, e). Wind Direction Difference View, and f). Wind View, each view and interactions between views are keys to solve the problem.

Fig. 12 System Overview. a). Calendar View, b). ISOMap View, c). Projection View and d). Sensor Read, e). Wind Direction Difference View，and f). Wind View

Our analyses are listed as follows:

1. Firstly, we obtain peaks of azimuth-wind direction difference view, then we can find factories and chemicals whose sensor readings are consistent with the peaks. Considered that wind direction has a great impact on the diffusion of chemicals, we designed an azimuth (factory- sensor) -wind direction difference view, shown as fig. , in which horizontal axis represents time, vertical axis represents the difference value between wind direction and the azimuth angle of factory and sensor, the formulas for calculating difference is given by equation (1): ydiff  = 1 / (|a(factory, sensor) - wind| + b)

Where a(factory, sensor) denotes the azimuth angle between factory and sensor, b is a constant. The larger the difference value is, the more consistent the current wind direction with the azimuth angle between factory and sensor becomes. If the peaks in azimuth (factory F and sensor M) - wind direction difference view is consistent with the peaks of chemical A’s readings captured by sensor M, then we think factory F should be responsible for the release of chemical A.

According to wind direction-wind speed view and calendar view we select some time, and be confirmed by the ISO Line and multiples in ISOMap View. The ISOMap view contains ISOLine Layer and Multiples views. We first obtain readings of one chemical captured by 9 sensors, then apply Kriging interpolation algorithm to estimate this chemical’s readings that around sensors and factories for the ISOLine drawing. As the current wind may affect sensors’ readings at the next moment, we utilize multiple views to display the trend of each sensor’s readings. The centre point of trend view represents the current time, the recording numbers before and after the current time point can be configured through the configuration panel.

Detailed analysis is as follows：
1. Radiance releases AGOC-3A

It is found that peaks in the wind direction-azimuth difference view of Radiance and S5 corresponds to peaks in S5-AG readings view (Sensor5 records the AGOC-3A chemicals), as shown in Fig.13-a. Similarly, peaks in Radiance and S9 view and peaks in S9-AG view also appear consistent, as shown in Fig.13-b. In this case, we assume that Radiance plant release AGOC-3A chemicals.

We first selected S5 and AGOC-3A chemical in calendar view, as shown in Fig.13-c. It can be found that several peaks appeared at April 11. We then clicked on the peak time 6:00 to observe the ISOMap view, from which we can see that the wind direction is northwest, and the focus of peak contour is located at the northwest direction of the Radiance factory. All above observations have preliminarily validated our hypothesis.

As shown in Fig.13-x, we can see that 1) the wind direction is 345.5  and wind speed is 3 at 3 pm. on April 24, and 2) the trends of S6, S7, and S8 located in the southeast direction are on the rise in the ISOMap's multiple trend view.

Fig. 13 Radiance release AGOC-3A a). Peaks in Radiance-Sensor5 direction difference view and Sensor5 for AGOC－3A are consistent; b). Peaks in Radiance-Sensor9 direction difference view and Sensor9 AGOC－3A are consistent; c). ISOMapView at 6:00, April 11, 2016; d). ISOMap View at 15:00, April 24, 2016.

2. Roadrunner releases Chlorodinine

We continue the analysis with similar methods, it is found that peaks in the wind direction- azimuth difference view of Roadrunner and S5 and peaks in S5-CH readings view are basically consistent, which can be verified by the wind direction-azimuth difference view of S1, S7, and S6.

As shown in Fig.14-e, the wind direction is 177.9 at 12 pm. on April 29 and S3 located over the Roadrunner is near the peak contour in the ISOMap view, which indicates that Roadrunner plant releases Chlorodinine chemicals.

Since all factories are located below the peak contour, the release of Chlorodinine from any plant will cause the formation of such contour.

It can be found that the wind direction is 211.5 and wind speed is 3.2 at 6 pm. on April 22. From the multiple trend view of the contour, we can see that the trends of S5 and S9 are on the rise. However, the trends of S7 and S8 are declining, whose wind directions are opposite to S5 and S9.

Fig. 14 Roadrunner release Chlorodinine. a) The value of Roadrunner-Sensor1 in direction- azimuth difference view is consistent with the peak value in Sensor1 Chlorodinine; b)The value of Roadrunner-Sensor6 in direction-azimuth difference view is consistent with the peak value in Sensor6 Chlorodinine；c) The value of Roadrunner-Sensor7 in direction-azimuth difference view is consistent with the peak value of Sensor7 Chlorodinine；d) The wind direction in 22/4/2016 18:00 is 211.5. The value of S5 and S9, which is consistent with the wind direction, is increasing, while the value of S7 and S8, which is in the opposite wind direction, are decreasing. e) At 29/4/2016 12:00, the wind direction is 177.9. The peak isovalue is around S3.

3. Indigo releases Appluimonia

From wind direction-azimuth difference view of Indigo-Sensor9 and reading view of APP-Sensor9, wind direction-azimuth difference view of Indigo-Sensor1 and reading view of APP-Sensor1, wind direction-azimuth difference view of Indigo-Sensor7 and reading view of APP-Sensor7, wind direction-azimuth difference view of Indigo-Sensor5 and reading view of APP-Sensor5, we can found that time of peaks in direction-azimuth difference view and reading view appear are almost same.

From the following views, we found that the wind direction is 19.5 on April 15, 18:00 and the wind speed is 2.6. From the ISOMap View, we can find that the APP chemicals from Indigo spread to Sensor 7 through the wind. We can see the value trend in Sensor 7 is increasing.

Fig. 15

Fig. 16 4/15, 18:00 APP chemicals ISO Map

4. Kasios releases Chlorodinine

Compared with the direction-azimuth difference view of Kasios-Sensor4, Kasios-Sensor1, and the value from Sensor4-Chlorodinine、Sensor1-Chlorodinine, we find there peak value is consistent. We can conclude from wind view that on the morning of April 16, wind direction sustains around 60 degrees and readings of Sensor 1 and 6 are increasing which locates in the direction of Kasios. From the ISOMap at 10:00, Apr.16, the peak value is in the following figure.

Fig. 17

Fig. 18

Fig. 19