The pervasive application of Internet of Things (IoT) has been seen in many aspects in human daily life and industrial production. The concept of IoT originates from traditional machine-to-machine (M2M) communications which aimed at solving domain-specific and applicationspecific problems. Today, the rapid progress of communication technologies, the maturation of Internet infrastructures, the continuously reduced cost of sensors, and emergence of more open standards, have witnessed the approaching of the expected IoT era, which envisions full connectivity between the physical world and the digital world via the Internet protocol. The popularity of cloud computing technology has enhanced this IoT transform, benefiting from the superior computing capability and flexible data storage, let alone the security, reliability and scalability advantages.

However, there are still a series of obstacles confronted by the industry in deployment of IoT services. First, due to the heterogeneity of hardware devices and application scenarios, the interoperability and compatibility between link-layer protocols, sub-systems and back-end services are significantly challenging. Second, the device management requires a uniform scheme to implement the commissioning, communication, authorization and identity management to guarantee security. Last, the heterogeneity of data format, speed and storage mechanism for different services pose a challenge to further data mining.

This thesis aims to solve these aforementioned challenges by proposing a data-centric IoT framework based on public cloud platforms. It targets at providing a universal architecture to facilitate the deployment of IoT services in massive IoT and broadband IoT categories. The framework involves three representative communication protocols, namely WiFi, Thread and Lo-RaWAN, to enable support for local, personal, and wide area networks. A security assessment taxonomy for wireless communications in building automation networks is proposed as a tool to evaluate the security performance of adopted protocols, so as to mitigate potential network flaws and guarantee the security. Azure cloud platform is adopted in the framework to provide device management, data processing and storage, visualization, and intelligent services, thanks to the mature cloud infrastructure and the uniform device model and data model. We also exhibit the value of the study by applying the framework into the digitalization procedure of the green plant wall industry. Based on the framework, a remote monitoring and management system for green plant wall is developed as a showcase to validate the feasibility. Furthermore, three specialized visualization methods are proposed and a neuron network-based anomaly detection method is deployed in the project, showing the potential of the framework in terms of data analytics and intelligence.

The primary goal for collecting and analyzing temporal data differs between individuals and their domain of expertise e.g., forecasting might be the goal in meteorology, anomaly detection might be the goal in finance. While the goal differs, one common denominator is the need for exploratory analysis of the temporal data, as this can aid the search for useful information. However, as temporal data can be challenging to understand and visualize, selecting appropriate visual representations for the domain and data at hand becomes a challenge. Moreover, many visual representations can show a single variable that changes over time, displaying multiple variables in a clear and easily accessible way is much harder, and inference-making and pattern recognition often require visualization of multiple variables. Additionally, as visualization aims to gain insight, it becomes crucial to investigate whether the representations used help users gain this insight. Furthermore, to create effective and efficient visual analysis tools, it is vital to understand the structure of the data, how this data can be represented, and have a clear understanding of the user needs. Developing useful visual representations can be challenging, but through close collaboration and involvement of end-users in the entire process, useful results can be accomplished. 

This thesis aims to investigate the usability of different visual representations for different types of multivariate temporal data, users, and tasks. Five user studies have been conducted to investigate different representation spaces, layouts, and interaction methods for investigating representations’ ability to facilitate users when analyzing and exploring such temporal datasets. The first study investigated and evaluated the experience of different radial design ideas for finding and comparison tasks when presenting hourly data based on an analog clock metaphor. The second study investigated 2D and 3D parallel coordinates for pattern finding. In the third study, the usability of three linear visual representations for presenting indoor climate data was investigated with domain experts. The fourth study continued on the third study and developed and evaluated a visual analytics tool with different visual representations and interaction techniques with domain experts. Finally, in the fifth study, another visual analytics tool presenting visual representations of temporal data was developed and evaluated with domain experts working and conducting experiments in Antarctica. 

The research conducted within the scope of this thesis concludes that it is vital to understand the characteristics of the temporal data and user needs for selecting the optimal representations. Without this knowledge, it becomes much harder to choose visual representations to help users gain insight from the data. It is also crucial to evaluate the perception and usability of the chosen visual representations.