👋 Open to new opportunities

Juan D. Hoyos

Control Engineer & Data Scientist

I build forecasting models and predictive controllers for complex, stochastic systems. For 5+ years I have worked on real-world, messy time series, from continuous glucose monitors and smart meters to solar inverters, closing the loop between prediction, optimization and real-time control.

View projects →
5+
Years in R&D and data-driven control
500+
Patients' CGM time series processed
−25%
Monthly bill with demand-forecast recommendations
3
Domains: biomedical, smart grids, power electronics

What I do

From forecast to decision

The common thread in my work: predict a stochastic signal, decide under constraints, and make it run reliably outside a notebook.

📈

Forecasting & uncertainty

Time-series models for glucose, demand and prices, with uncertainty estimates that downstream controllers can use.

🎛️

Model predictive control

MPC and PID+FF controllers under constraints, from glucose regulation to solar inverter DC/AC stages.

🧮

Optimization

LP/QP and mixed-integer formulations for scheduling, plus metaheuristics (GA, PBIL) for model identification.

⚡

Energy systems

Smart-meter analytics, demand forecasting, power electronics and flexible-asset dispatch such as batteries.

🚀

Built to ship

Reproducible Python pipelines, Docker, Git and microservices. I favor the solution that works under real-world noise over the theoretically perfect one.

Portfolio

Selected work

Research, industry and personal projects. Open any card for the approach and my contribution.

🩸
Research

Glucose prediction & control for type 1 diabetes

Machine-learning tools for blood glucose prediction and closed-loop control in people with type 1 diabetes.

MPC System identification MATLAB Python
Closed-loop control from CGM data only
Details
  • Control-oriented glucose-insulin models identified from real patient data, including free-living data without insulin or meal records.
  • Predictors feed model predictive controllers that respect safety constraints on glucose and insulin delivery.
📊
Research

Patient segmentation from 500+ CGM series

Reproducible pipelines to clean continuous glucose monitor data and cluster patients by metabolic variability.

Clustering Pandas scikit-learn R
Better glucose prediction per subgroup
Details
  • Preprocessing and quality control of large, noisy physiological time series for a cohort of more than 500 people with diabetes.
  • Unsupervised clustering in Python to segment patients, improving blood glucose prediction accuracy.
  • Statistical validation in R (ANOVA, PCA, hypothesis testing) to assess the clinical significance of the groups found.
🧬
Research

Metaheuristic optimization for model identification

Population-based and evolutionary algorithms to estimate metabolic parameters and to correct noisy inputs in clinical records.

PBIL Genetic algorithms Identifiability Simulink
Published research
Details
  • Population-Based Incremental Learning to identify blood glucose dynamics models for 75 type 1 diabetes patients.
  • Genetic algorithm to correct noisy carbohydrate-intake records before parameter estimation.
  • Review and analysis of the identifiability of control-oriented glucose-insulin linear models.
🔌
Industry-academia

Smart-meter analytics & demand forecasting

Python and MongoDB architecture to process hundreds of GB of smart-meter time series, with forecasting models behind consumption recommendations.

Demand forecasting Python MongoDB Big time series
↓ up to 25% monthly bill
Details
  • Designed the data architecture for statistical processing of hundreds of GB of smart-meter data from distribution networks (project with Enel).
  • Built predictive models of consumption and a recommender that shifts end-user demand.
  • Co-authored a platform to simulate the impact of distributed generation and electric mobility on distribution networks.
☀️
Industry

MPC for solar inverter DC/AC systems

Design and validation of MPC and PID with feed-forward controllers for power-electronics stages, in MATLAB/Simulink.

MPC PID + FF Power electronics Simulink
Robust across operating conditions
Details
  • Statistical and data-driven validation of control algorithms to check robustness across operating points.
  • Worked with cross-functional teams to troubleshoot data-integration issues; Git-based workflow.
🔋
Personal project

Battery dispatch under price uncertainty

Simulation study that closes the loop: probabilistic price forecasts feed a stochastic MPC that schedules a battery against day-ahead and intraday prices.

Stochastic MPC Quantile forecasting QP / MILP Python
Forecast → optimize → control
Details
  • Quantile forecasts of electricity prices (gradient boosting), evaluated with pinball loss and walk-forward validation.
  • Receding-horizon scheduling with a scenario-based formulation: state-of-charge, power and degradation constraints.
  • Compared against a deterministic MPC and a rule-based strategy to measure the value of modeling uncertainty.

Stack

Skills

Control & optimization

Model predictive control Stochastic MPC PID + feed-forward State-space System identification LP / QP Mixed-integer programming Stochastic programming Genetic algorithms / PBIL

Forecasting & machine learning

Time-series forecasting Probabilistic / quantile forecasts Clustering Gradient boosting PyTorch scikit-learn ANOVA · PCA · hypothesis testing

Energy & power

Demand forecasting Smart-meter data Solar inverters Battery dispatch Electricity markets (day-ahead, intraday, balancing)

Programming & data engineering

Python pandas · NumPy · SciPy R MATLAB / Simulink SQL · NoSQL · MongoDB Docker Git · Bash Microservices Power BI

Contact

Let's talk

Interested in forecasting, control and optimization roles, especially in energy and flexibility.