Xinyang Sun
I'm |
About
I'm a passionate and self-motivated PhD Candidate in Engineering at King’s College London. I dedicate to unravelling the complexities of the ever-evolving world through the lens of interdisciplinary research. My current research project delves into a dynamic intersection, where telecommunications, machine learning, and sustainability converge. As a PhD Candidate, I thrive on the challenges presented by diverse fields and find immense joy in bridging the gaps between them.
I received my BSc degree in Telecommunications Engineering with Management from Queen Mary University of London and MSc degree in Telecommunications and Internet Technology from King’s College London. My academic journey has equipped me with a versatile skill set, including programming and data analysis. As a believer in the potential of interdisciplinary collaboration, I'm excited to contribute to projects that push the boundaries of traditional academic fields.
Language
English
Bilingual Proficiency
Chinese
Bilingual Proficiency
Spanish
Limited Working Proficiency
Resume
Education
Doctor of Philosophy - PhD, Engineering
Feb 2024 - Present
King's College London, England, United Kingdom
Area of Research: Digital product passport and DPP systems hold great promise to enable digital transition of manufacturing sectors towards circular economy. My project aims to develop a novel modelling approach underpinned by Machine Learning algorithms, optimisation theory to enable DPP technologies and simulate and optimise material and energy flows, support supply chain decision-making on waste recovery maximisation.
Master of Science, Telecommunications and Internet Technology
Sep 2022 - Sep 2023
King's College London, England, United Kingdom
Grade: Pass with Distinction
Key Modules: Network Theory (90), Topics on Data and
Signal Analysis (88), Fundamentals of Digital Signal Processing (83),
Telecommunications Networks I (82), Telecommunications Networks II (79), Digital
Communications (76).
Bachelor of Science, Telecommunications Engineering with Management
Sep 2018 - Jun 2022
Queen Mary University of London, England, United Kingdom
Grade: First Class Honours
Key Modules: Engineering Mathematics 2 (90), Internet
Protocols (79), Image and Video Processing (79), Engineering Environment (Telecom)
(78), Communication Skills (76), Product Development and Marketing (76).
Experience
Data Scientist
Jan 2025 - Present
Carlsberg Group, England, United Kingdom
Developed and deployed a machine learning-based demand forecasting model for a wide
range of SKUs using algorithms such as Prophet, LSTM and XGBoost, achieving over 90%
prediction
accuracy.
Designed and implemented an AWS SageMaker pipeline integrated with S3, enabling
seamless
model training, deployment, and scaling across GB and Ireland markets.
Secured additional funding by demonstrating the success and business impact of the
demand forecasting project, showcasing its potential to drive operational and
financial
improvements.
Data Scientist
Jul 2024 - Jan 2025
Britvic Soft Drinks, England, United Kingdom
Developed and delivered a demand forecasting model for diverse SKUs using advanced
machine learning algorithms, including LSTM and XGBoost, achieving over 80%
prediction
accuracy.
Designed a user-friendly web interface to provide real-time insights and simplify
interaction with forecasting models, empowering stakeholders to make data-driven
decisions effectively.
Built digital twins for the manufacturing lines in Rugby factories, enabling
predictive
maintenance and reducing downtime by identifying potential issues before they
occurred.
Skill
Machine Learning
Data Modelling
Mathematical Optimisation
Artificial Intelligence
Cloud Computing
Digital Sustainability
Research
Digital Product Passport for Manufacturing and Supply Chain
This project develops a Digital Product Passport (DPP) system to advance circular economy objectives by enabling full supply chain traceability, resource optimisation, and environmental impact reduction. The DPP serves as a dynamic data framework that captures and communicates product information across lifecycle stages. The system incorporates edge computing to enable decentralised, real-time data processing at the source of information generation, supporting responsive and energy-efficient decision-making. This architecture enhances data integrity, reduces transmission delays, and minimises cloud dependency. The system is designed to be scalable and adaptable across industries, demonstrating practical applications through targeted case studies. These studies assess the system’s impact on transparency, accountability, and sustainability by tracking environmental metrics such as carbon footprint and material reuse rates. The outcomes highlight the DPP’s potential to improve both economic and environmental performance, empowering stakeholders with actionable insights for responsible production and consumption.
A KAN-based Interpretable Framework for Process-Informed Prediction of Global Warming Potential
This project developed an integrative model for predicting Global Warming Potential by combining molecular and process-level data. Traditional GWP models focus mainly on molecular structure, but this approach incorporated molecular descriptors alongside process metadata to enhance accuracy. A deep neural network achieved an R² of 86% using Mordred descriptors and process details, marking a 25% improvement over the previous benchmark. Explainable AI techniques revealed that embeddings of process titles significantly influenced predictions. To improve interpretability, a Kolmogorov-Arnold Network generated a symbolic expression for GWP, offering a transparent alternative to black-box models. Error analysis indicated strong reliability in dense data regions, with higher uncertainty for extreme GWP values. By integrating structural and contextual features, this model advances GWP prediction accuracy and interpretability, providing a practical tool for sustainable chemical process design. Future developments aim to expand this framework to additional environmental impact metrics.
Temporal Dynamics of Microbial Communities in Anaerobic Digestion: Influence of Temperature and Feedstock Composition on Reactor Performance and Stability
This project applied detailed chemical and biological fingerprinting to enhance anaerobic digestion (AD) of carbon-rich fermentation wastewater. Using mycoprotein-derived effluent, eighteen reactors were operated under varied conditions to examine how specific wastewater compounds and operational settings shape microbial communities and reactor performance. Chemical analysis tracked sugars, sugar alcohols, and volatile fatty acids (VFAs) throughout the process, while microbial sequencing identified temperature- and configuration-dependent biodiversity shifts. Mesophilic conditions supported more connected and diverse microbiomes. Machine learning models predicted performance with high accuracy using operational data and microbial profiles, identifying Oscillibacter and Clostridium species as key biogas contributors. These findings highlight the value of integrating molecular-level insights with AI tools to optimise AD processes, offering pathways toward more efficient, sustainable waste-to-energy solutions.
Life Cycle Optimisation Tool Development for Process Systems and Supply Chain Design
This project developed a life cycle optimisation tool to support environmentally and economically balanced decision-making in supply chain management. By combining life cycle assessment with advanced optimisation methods, the tool quantifies environmental impacts—such as greenhouse gas emissions—while identifying strategies to maximise overall system profitability. The optimisation framework addresses a constrained multi-objective problem using two distinct approaches. The first applies a weighted sum method, converting environmental costs into economic terms by assigning a price to emissions. The second uses NSGA-II, a genetic algorithm, to generate a Pareto front that presents a spectrum of optimal trade-offs between environmental impact and profit. A case study focused on the UK electricity generation sector validated the framework, incorporating constraints such as national electricity demand and emissions targets across the supply chain. The results showed that the tool effectively supports strategic planning for net-zero transitions without sacrificing financial performance.
Swarm Intelligence for Firefighting Drones
This project applied swarm intelligence to optimise drone deployment in firefighting scenarios. Inspired by the decentralised coordination seen in social insects like ants and bees, the approach enabled autonomous drones to adjust their movement based on the speed and location of neighbouring drones within a defined range. This prevented redundant coverage and collisions, allowing drones to collectively cover a wider area more efficiently. Each drone dynamically calculated proximity and adjusted velocity accordingly, supporting decentralised behaviour without central control. Python was used to simulate fire scenarios, and Matplotlib generated real-time visualisations of drone trajectories, speeds, and cumulative coverage. Compared to randomised movement, the swarm-based strategy significantly improved area coverage within the same timeframe. The results demonstrated how swarm intelligence can enhance autonomous coordination and contribute to more effective firefighting strategies in complex, real-world environments. The framework also lays the groundwork for scaling to larger drone fleets and integrating real-time sensor data for adaptive response in evolving fire conditions.
Publication
- All
- Journal Articles
- Conference Papers
Coding
Contact
Birthday
11th July 2000
Location
London, England, United Kingdom
Phone
+44 7864 584558
Hometown
Beijing, China