In the era of ever-increasing social and technological interconnectedness, complex socio-technical systems (STS) engineering and design have emerged as a pivotal discipline. These systems, characterized by their intricate interdependencies and dynamic interactions, permeate various facets of modern life, ranging from smart power grid and electric vehicle (EV) charging systems to vehicle (EV & non-EV) market systems and beyond. The challenges inherent in the engineering and design of such systems necessitate a paradigm shift in methodology and tools. To this end, my long-term research goal is to establish the theoretical foundation for STS engineering and design so as to tackle the rapidly increasing complexity. In my further research, I aim to build the foundation by synergistically integrating knowledge from the following disciplines: network science, machine learning, optimization, and statistics.

Figure 1: Research Overview

Research Experience and Contributions

As shown in Figure 1, the central hypothesis of my dissertation research is that the statistically significant local connections of individual entities, embedding collective behaviors of entities, are crucial function units of STSs, influencing global-level system performance. A paradigmatic example is the shared mobility network, such as bike-sharing systems. Compared to individual stations that only record users’ rental and return behaviors, local service systems formed by several stations (e.g., a triangle of three stations) capture users’ travel behaviors within the local network (e.g., a triangle transit). A better understanding of such a subsystem can help better probe into users’ mobility patterns and local resource distribution, which are critical for the optimal decision-making of system engineers and operators. Despite the significance of local network structures and behaviors, existing work has primarily focused on investigating individual entities and the impact of their behaviors on system-level performance. Therefore, a fundamental knowledge gap exists in understanding and utilizing meaningful local network structures for STS engineering and design. My dissertation has contributed to the current literature by creating a novel local-level network-based framework for STS engineering and design, combining three advanced network theories and models, including network motif theory, exponential random graph model (ERGM), and graph neural network (GNN) model. In particular, this framework contains two steps. In Step 1, I developed an approach to identifying statistically significant local network motifs (i.e., recurring network patterns) and analyzing their impacts on system structure and performance. In Step 2, I developed a new STS engineering and design approach based on the local information obtained from Step 1.

1. Step One: Significant Local Network Motifs Identification and Impact Analysis

One of my dissertation research outcomes is an analytical framework to address the challenge of extracting local network motifs embedded within the global network. The first step involves an appropriate abstraction of STS structure as well as intricate interactions using network representations. The second step entails the application of network motif theory to identify statistically significant subsystems. In step three, I developed a systematic approach to evaluate the impact of those identified subsystems based on ERGM and GNN. In a case study on market systems with a particular example of household vacuum cleaners, I first defined a co-consideration network representation where nodes represent product models and links denote the co-consideration relations between two products (i.e., two products are co-considered at least once). This network model has been proven to be a powerful representation of product competition, from which essential competition patterns were successfully identified. For example, it was discovered that the pattern, where products from three brands form a closed triangle competition, positively influences participating products’ competitiveness. This enables enterprises as market players to measure their competitive standing by quantifying the frequencies of their products engaged in this pattern. In another case study on bike-sharing systems, I developed a GNN-based model to predict whether a pair of stations would have sufficient travel demand. This model was found to have a ~10% improvement over traditional neural network models because of the incorporation of local network structures in machine learning.

2. Step Two: Using The Identified Local-Level System Information to Guide STS Engineering and Design

In this step, I developed a new approach to incorporating the significant local-level system information obtained from Step 1 into the STS design process. In the bike-sharing case study, I discovered that the imbalanced number of rental and return bikes within a local service system (e.g., a local system consists of three stations) is sensitive to seasonal effects, e.g., serious imbalance issues occur more often in summertime, leading to user dissatisfaction. I further discovered that this sensitivity positively correlates to the differences in docking capacities between stations within those identified local systems. Based on these discoveries, I formulated an optimization problem to improve dock planning (i.e., how many docks shall be designed in a station) and seasonal robustness. The developed approach has been successfully demonstrated in Chicago’s Divvy Bike. Moreover, in my ongoing research, I developed a network-based methodology to integrate local-level system information into the optimal design of individual entities’ attributes and behaviors. Taking the household vacuum cleaner market system as an example, I first defined the competitiveness vector of a vacuum cleaner model based on its involvement in significant competition patterns identified in Step 1. This vector is a function of the model’s configurations, such as suction power and weight. Then, the relationship between the competitiveness vector and the model’s market share was estimated. Finally, I conducted an optimal design of the vacuum cleaner model’s configuration with the objective of maximizing its market share. The preliminary findings demonstrate the applicability of the proposed method in enhancing the market share of a targeted vacuum cleaner model, indicating the potential for real-world application to guide product design such as electrical vehicles(EVs).

Future Research

my future research objective is to extend my dissertation work by delving into socio-technical systems (STS) of greater complexity because of temporal effects and the embrace of artificial intelligence (AI) technologies. This involves advancing our understanding of AI-integrated STSs (e.g., smart power grids) and the temporal dynamics of STSs to inform effective engineering and design practices. In particular, I propose two research projects outlined below.

1. An Innovative Framework For AI-Integrated STS Engineering and Design

As AI increasingly permeates our daily lives, its interactions with both humans and machines are reshaping the socio-technical landscape. For example, the integration of autonomous vehicles in transportation systems allows for real-time traffic analysis and prediction, enabling efficient route optimization, reduced congestion, and improved overall traffic management. However, the active role of AI also raises concerns about data security and protecting systems from potential threats and breaches. In this project, my research objective will be to develop an advanced framework guiding the engineering design of machine and AI entities for an STS with the aim of serving humans better. In the context of the smart power grid, an illustrative instance of machine design involves end-users determining the type and quantity of solar panels for home installation. In terms of AI design, wherein AI represents a combination of machine learning algorithms assisting end-users in energy management decisions, the design considerations encompass the level of AI involvement in each decision-making process. For instance, the designed AI can autonomously decide to deactivate predefined non-essential electronics during energy shortages, but it does not assist end-users in financial decisions such as selling stored energy. To achieve the research goal, I propose three detailed research tasks: (1) Develop a general mathematical representation, considering machine and AI design parameters as independent variables, to model diverse human-AI-machine interactions within a minimal functioning unit of STS. For example, taking a single smart home as a representative unit in the smart power grid, these interactions may include instances where humans grant AI partial authority to control energy output, and output terminals provide feedback to both AI and humans. This modeling approach will be grounded in probability, control, and utility theories, etc. (2) Use network theory to model the interactions among different functioning units and assess the robustness of the STS at a global level, such as evaluating the reliability of smart power grids at the community level when facing cybersecurity threats. (3) Apply optimization theory to optimize machine and AI design to maximize human satisfaction and global-level system reliability. This study contributes to understanding the relationships between humans, AI, and machines, emphasizing the role of AI in an STS. The potential applications are AI-integrated future mobility networks (e.g., autonomous vehicles) and smart power grids.

2. STS Engineering and Design Considering Temporal Dynamics

There is a consensus that STSs continuously evolve due to ever-changing human behaviors. Design strategies would not be effective without thoroughly understanding such dynamic changes. For example, in the customer-product market system, customer preferences could be different from time to time, necessitating design strategies that accommodate this evolution for future market success. However, temporal changes from system structure (e.g., the construction of new EV charging stations at a new city center) to individual attributes (e.g., an increase in battery capacity for a new version of an EV model) pose one of the most challenging problems in STS. To tackle this challenge, my research objective is to develop a network-based analysis framework to uncover temporal patterns (e.g., weekly or monthly traffic peaks in transportation systems) of STSs to guide systems design decisions. Detailed research tasks include: (1) Apply a complex network to simulate STS evolution (preservation, evolution, formulation, dissolution as illustrated in Figure 1). For example, in an EV charging system, define stations as nodes and power transmissions between them as links. Both nodes and links exhibit temporal dynamics, such as station expansion and time-specific power transmission. (2) Use network motif theory and Separable Temporal ERGM to identify statistically significant evolution patterns (e.g., power transferring between three charging stations only in period t) and key factors driving system evolution (e.g., population increase motivating charging station expansion). (3) Develop a GNN model for STS evolution prediction considering identified patterns and factors. (4) Optimize STS configurations for resilience amid evolving complexities, e.g., optimizing charging capacity and power transmission to meet dynamic demand. The expected outcome includes a network-based visualization tool illustrating STS evolution and an STS design framework with a verification function based on the proposed predictive model. The potential applications span from EV charging infrastructure to design for market systems.