.png)
Map Interface
The entry view connects field location, weather, nectar sources and system status in one map-based workspace.
Website System / Predictive Support
Bee Cue
A Predictive Support System for Beekeeping under Phenological Mismatch
A web-based system that translates weather, flowering and bee activity data into readable decision support for beekeeping practice.
.png)
Bee Cue is a graduation design project that translates ecological prediction into a readable, interactive and working web system for beekeeping practice.
Project Type
Graduation Design / Interactive Website / Data Visualization System
My Contribution
I led the end-to-end design and development of Bee Cue’s digital system, from system architecture and prediction logic to interaction design, data visualization and front-end implementation.
Team Collaboration
Collaborators contributed the ecological corridor design and hive sensing component.
Duration
2025 - 2026
Deliverables
Working Website / Data Dashboard / Prediction Logic / Project Film / Research Report
Tools & Methods
Python, HTML, CSS, JavaScript, Weather API, Sensor Data, Rule-based Models, Machine Learning Correction
Context
China Academy of Art, Art & Technology Graduation Project
Project Boundary
Bee Cue supports judgment by making ecological change readable, traceable and comparable while leaving final decisions to beekeepers.
Final Product
Working Web System
The final outcome is a functional web system connecting a site-based map, ecological indexes, bee activity data, scenario comparison and AI-assisted follow-up.
.png)
.png)
Environmental Layer
This layer presents nectar plants, flowering periods and environmental conditions, showing how resource availability changes across time and space.
.png)
Behavior Layer
Bee activity is translated into heatmaps, sensor records and activity curves, making behavioral changes visible and comparable.
.png)
Strategy Page
Strategy cards organize possible responses based on the current ecological state, allowing users to compare references before taking action.
.png)
Strategy Detail Page
Each strategy expands into detailed guidance, including conditions, operation steps, risk notes and an entry to AI-assisted conversation.
Product Walkthrough
Working Prototype Demo
A complete walkthrough of the live flow—from environmental layers and bee activity to mismatch risk and scenario-based references.
Design Challenge
Ecological timing is becoming harder to read.
Climate change destabilizes the timing between flowering and bee activity, making nectar windows and mismatch risk harder to judge through experience alone.
Temperature, rainfall and extreme weather are making ecological rhythms less stable.
Earlier, delayed or shortened blooms make optimal nectar-flow periods harder to anticipate.
Weather, flowering and colony condition must be interpreted together in the field.
Scattered sources are difficult to translate into timely and practical judgment.
Research
Grounded in ecological and field context
Literature, field observation and data mapping established the ecological and operational context for the system.
01 Literature Review
The research began with phenological mismatch, pollination, climate change and agricultural decision support systems.
02 Field Context
Field observation focused on the beekeeping site, nectar plant distribution, seasonal changes and local environmental conditions.
03 Data Sources
The system was built around three main types of information: weather data, flowering data and bee activity data.
Design Decisions
Make ecological change visible, comparable and traceable.
Three principles guided the system: reveal ecological change, explain prediction clearly and preserve the user’s final judgment.
System Logic
From Data to Decision Support
Bee Cue combines transparent rule-based ecological models with a machine learning correction layer. A daily automated pipeline synchronizes new weather and hive observations, recalculates ecological indexes, retrains the correction model and exports updated results to the website.
Machine Learning Boundary
Machine learning is used as a correction layer, not as a black-box replacement or automatic decision-making engine.
System Evolution
Two system-level changes shaped the final product.
Single Suggestion→Scenario Comparison + Detailed Guidance + AI Follow-up
The system moved away from one fixed answer and now lets users compare possibilities while retaining practical judgment.
Rule-based Prediction→Rule-based Models + ML-assisted Correction
Transparent ecological rules remain the foundation, while observed hive activity is used to recalibrate prediction deviation.
Working Web System
A connected map, dashboard, strategy and AI experience.
Automated Data Pipeline
Daily synchronization, index recalculation, model retraining and web export.
Historical Correction
706 samples · MAE 0.1085 → 0.0875 after ML-assisted correction.
Observed bee activity is compared with rule-based prediction and ML-adjusted output to evaluate how the correction layer changes prediction deviation.
706Historical samples0.1085Rule MAE0.0875ML-adjusted MAE
Preliminary historical validation; further cross-season and cross-site testing is required.Field Evaluation
Testing with beekeeping practitioners
Members of a beekeeping cooperative completed a task-based walkthrough of a pre-final desktop prototype, covering the full website flow.
Use Familiar Language
Finding Technical strategy explanations were difficult to follow.
Design response Added role context, beekeeping vocabulary and clearer AI response constraints.
Reduce Map Overload
Finding Displaying all plant information at once made the map difficult to read.
Design response Prioritized current nectar-source plants and revealed other plant information through interaction.
Make Guidance Easier to Scan
Finding Users tended to skip long blocks of operational instructions.
Design response Divided guidance into shorter steps and paired it with line-drawing illustrations.
Project Film
A complete narrative supplement
A short film introducing the ecological background, system logic, interface design and application scenario of Bee Cue.
Reflection
This project strengthened my ability to connect ecological research, data structures, prediction logic, interaction design and front-end implementation within one working system.
Future development will focus on longer-term field data, cross-season and cross-site validation, improved mobile use and stronger feedback loops with practitioners.
