In this study, an IoT-based indoor environment monitoring system is designed. The system consists of a monitoring sensor module that collects real-time indoor environment information, a calculation and data processing module that analyzes information and transmits control signals, and a Wi-Fi-based communication module. Each sensor collects information about the indoor environment, integrates the collected information within the sensor module, and delivers it to the indoor environment integrated management module server through a wireless network.

Figure 1 shows the system architecture showing the overall operational flow of the system to be developed and the configuration of each module.

Figure 1: 
System architecture

The measurement equipment for monitoring the indoor environment of a ship and maintaining and improving the indoor environment at an appropriate level must acquire data on indoor environment information (temperature, humidity, and fine dust) and indoor activity information (vibration and noise). The information collected from each sensor module is presented in Table 2.

The actual indoor environment changes continuously at a slow rate. However, in the case of an indoor environment monitoring system in a ship, it is an environment in which the outliers are prone to occur because of challenges such as communication errors due to closed spatial characteristics.

Figure 2 shows the actual collected PM 2.5 (particulate matter less than 2.5) data. In this figure, the x-axis is time in minutes, and the y-axis is the concentration of PM 2.5. The overall signal is stable; however, a signal with an exceptionally high transient value can be observed, as represented by the red box. The majority of these are communication errors that occur in the shielding environment of ships, which makes it difficult to monitor and manage the indoor environment through this system.

Figure 2: 
Collected signals and noises

To address this problem, an outlier removal algorithm is proposed, as shown in Figure 3. The mean and standard deviation were calculated by statistically analyzing the data of the previous 24 h based on the measurement time, and the calculated statistical values were converted into a Z-score. Based on this result, it is determined whether the data at the current time point is normal or abnormal, and if it is abnormal, the outlier is replaced by a moving average filter.

Figure 3: 
Proposed algorithm

Using Equation (1), the Z-score value Z_t can be derived.

where, X_t is the air condition at time t, μ_(24h) and σ_(24h) are the mean and standard deviation, respectively. Z_t becomes a criterion for filtering, as shown in Equation (2).

The estimation value of Equation (2) is determined through the threshold, and X_t is the moving average.

To verify the validity of the proposed outlier removal algorithm, it was compared and analyzed with widely used IQR and DBSCAN [8, 9]. The experiment was based on the data collected for a single day by installing the actual equipment on the ship. As shown in Figure 4, the Z-score applied in this study displayed the best accuracy.

Figure 4: 
Comparison of accuracy by method

To evaluate the actual performance of the proposed indoor environment monitoring system, it was installed in an actual vessel environment, and experiments were conducted.

Figure 5 and 6 show the raw measured data of PM 2.5 and the same filtered result, respectively. It can be seen that the outliers generated in Figure 5 are all removed as a result of the filter in Figure 6.

Figure 5: 
The measurement data of PM 2.5 (raw data)

Figure 6: 
The measurement data of PM 2.5 (filtered data)