Bonus Days 13 & 14: Making Model 4 Production-Safe — Guardrails, EMA, and End-to-End Logic скачать в хорошем качестве

Bonus Days 13 & 14: Making Model 4 Production-Safe — Guardrails, EMA, and End-to-End Logic

In today’s video, we finish the bonus Model 4 series by turning theory into something you could actually trust in a real building. These two lessons were recorded back-to-back because they belong together. First, we focus on guardrails and stability. Even a great model can do dumb things if the data is noisy, the building behaves strangely, or it’s a classic “Monday morning” scenario. Model 4 is no exception. Without guardrails, small estimation errors can explode into massive start times. We talk through the real failure modes and how to prevent them with simple, practical rules. Then we put everything together in a full end-to-end Model 4 loop. This is the point where all the previous bonus lessons finally click: first-order intuition, exponentials, logs, deadband, estimating c, and smoothing parameters over time. Nothing fancy. Nothing hidden. Just clean logic you can reason about. In this combined lesson we cover: • Why c drifting toward 1 breaks predictions • Why day-to-day noise causes twitchy start times • How simple clamping prevents runaway behavior • Why max runtime limits are essential • How EMA smoothing stabilizes learned parameters • Why a stable model beats a perfect model • A full mini Model 4 workflow: – build error signals – estimate c from data – compute t_opt with logs – apply guardrails – smooth parameters with EMA All examples are shown in simple Python with no NumPy, reinforcing that this logic is realistic for BAS platforms, Niagara Program Objects, and edge devices. By the end of this video, Model 4 is no longer “advanced math.” It’s a practical, adaptive, physics-inspired optimal start algorithm that can actually survive real buildings and real data. Lesson references https://github.com/bbartling/hvac-opt... Vibe Coding (Niagara Program Objects) https://github.com/bbartling/niagara4... HVAC Optimal Start Math Playground https://github.com/bbartling/hvac-opt... #hvac #optimalstart #buildingautomation #niagaraframework #smartbuildings #hvacmath #firstordersystems #controlsystems #engineeringeducation #vibecoding #hvacoptimization #hvaccontrols