Skip to content

VPP Lean RL English Site

Safe RL + formal verification

Train. Filter. Verify.

This site is for engineers and researchers who want more than a static README: real traces, interactive charts, reproducible examples, and a clear path from Python RL to Lean verification.

4 scenarios
3 RL algorithms
3 shield modes
v3 trace schema

Where To Start

New to the Project

Use the shortest path from zero to a verified rollout.

Open the English Quick Start

Want to See Real Behavior

Open the runtime demo and inspect SoC, prices, rewards, actions, and intervention reasons.

Open the Demo Page

Need Concrete Recipes

Use runnable command sequences for baseline, projection, Lean table, solar export, and stress test runs.

Open Experiment Playbooks

Architecture Overview

flowchart LR
    A[Python Agent<br/>Q-learning / SARSA / Double Q] --> B[Safety Shield<br/>none / project / table_project]
    B --> C[VPP Environment<br/>battery + solar + load + grid]
    C --> D[Trace Export<br/>proposedAction / appliedAction / reasons]
    D --> E[Lean Verifier]
    F[Lean Rule Table] --> B

Example 1

Peak Shaving With No Shield

Use this as the raw baseline to measure how much value the policy extracts before safety intervention.

Run the baseline recipe

Example 2

Projection Shield Comparison

Compare proposed versus applied actions and inspect where online projection changes the rollout.

Inspect the interactive comparison

Example 3

Solar Export Workflow

Switch to a midday-surplus profile and observe how charging and export behavior changes.

See the solar export recipe

Example 4

Stress Test Constraints

Run tighter limits and use the trace schema pages to inspect battery and grid margins.

Open trace schema reference

Suggested Path By Goal

  1. Learn the system: Quick Start -> Demo
  2. Extend the system: User Manual -> Command Reference
  3. Design experiments: Playbooks -> Trace Schema

Live Entrypoints

  • Root language portal: https://tigerneil.github.io/vpp/
  • English site: https://tigerneil.github.io/vpp/en/
  • Chinese site: https://tigerneil.github.io/vpp/zh/