dTwin 1.1.9

Python bindings for ModelSolver library

dTwin

dTwin is a Python interface to a high‑performance C++ nonlinear model solver capable of solving:

• Static problems

  • Nonlinear algebraic equations (NLE)
  • Weighted Least Squares (WLS)

• Dynamic problems

  • Ordinary Differential Equations (ODE)
  • Nonlinear Differential‑Algebraic Equations (DAE)

It is built using the C++ modelSolver library (part of the dTwin framework).

This README explains how to install, understand, and use dTwin step by step, with real working examples.

---

1. Installation

Install from TestPyPI / PyPI

pip install --index-url https://test.pypi.org/simple/ dTwin

or

pip install dTwin --index-url https://test.pypi.org/simple/

(or from PyPI once released)

pip install dTwin

Verify installation

import dTwin
from dTwin import modelSolver
print(dTwin.__doc__)

---

2. What does dTwin.modelSolver solve?

Problem type	Description
Static (NLE)	Systems of nonlinear algebraic equations
Static (WLS)	Weighted least squares estimation
Dynamic (ODE)	Time‑domain simulation of ODE systems
Dynamic (DAE)	Time‑domain simulation of nonlinear DAEs

Models are described using .dmodl files, a domain‑specific language for defining variables, parameters, equations, limits, controllers, and submodels.

---

3. Core Python API overview

Enumerations

dTwin.StaticProblem.NLE
dTwin.StaticProblem.WLS

dTwin.DynamicProblem.ODE
dTwin.DynamicProblem.DAE

dTwin.Solution.OK

Vector types (C++ backed)

These are zero‑copy wrappers around C++ memory:

dTwin.DoubleVector
dTwin.StringVector
dTwin.UintVector

They behave like Python vectors:

v = dTwin.DoubleVector(3)
v[0] = 1.0
print(len(v), v)

---

##4. Examples

There are several examples in the models subfolder.

4.1 Static model example (Power‑flow NLE)

Model file: PF_PV_03.dmodl

This model solves a power‑flow problem with:

• PV and PQ buses
• Generator voltage regulation (PV node with limits)
• Reactive power limits

(Details: see associated arXiv paper.)

Python usage

import dTwin
from pathlib import Path

p_log = dTwin.getConsoleLogger()

# Create real‑valued static model (NLE)
p_model = dTwin.createRealStaticModel(
    dTwin.StaticProblem.NLE,
    p_log
)

# Load model
p_model.initFromFile("PF_PV_03.dmodl")

# Obtain solver
p_solver = p_model.getSolverInterface()

# Solve
status = p_solver.solve()
assert status == dTwin.Solution.OK

# Extract outputs
out_indices = p_model.getOutputSymbolIndices()
out_names   = p_model.getOutputSymbolNames(out_indices)
out_values  = p_model.getOutputSymbolValues(out_indices)

for i in range(len(out_names)):
    print(out_names[i], out_values[i])

p_model.release()

Parameter manipulation

idx = p_model.getParameterIndex("P3_inj")
vals = p_model.getParameterValues(dTwin.UintVector([idx]))
vals[0] -= 0.1
p_model.setParameterValues(dTwin.UintVector([idx]), vals)

p_solver.solve()

---

4.2 Dynamic DAE example (Generator frequency regulation)

Model file: FreqReg_01.dmodl

This model simulates:

• Synchronous generator swing equation
• PI frequency controller
• Nonlinear network algebraic equations
• Initial power‑flow solved via a SubModel
• Fully nonlinear time‑domain simulation

Python usage

import dTwin

p_log = dTwin.getConsoleLogger()

p_model = dTwin.createRealDynamicModel(
    dTwin.DynamicProblem.DAE,
    p_log
)

p_model.initFromFile("FreqReg_01.dmodl")

p_solver = p_model.getSolverInterface()

# Time step (optional)
dt = p_solver.getStepSize()
if dt <= 0:
    p_solver.setStepSize(0.001)    

# Initialize
p_solver.reset(0.0)

out_idx   = p_model.getOutputSymbolIndices()
out_names = p_model.getOutputSymbolNames(out_idx)

# Time loop
t = 0.0
T_END = 20.0

while t <= T_END:
    #change model parameter to simulate transients
    sol = p_solver.step()
    assert sol == dTwin.Solution.OK

    values = p_model.getOutputSymbolValues(out_idx)
    print(t, list(values))
    t += dt

p_model.release()

---

4.3 Dynamic model with transfer functions (Dorf's example)

Model file: TF_Dorf_E760_PD_Disk_RK4.dmodl.

Features:

• Laplace transfer functions (TFs)
• PID (PD) controller
• Implicit RK4/RK6 integration

No Python code changes are required — simply load the model and simulate as shown above.

---

5. Memory management (important)

Models are C++ objects.

• They are automatically released when garbage‑collected
• You may explicitly free memory (optional):

p_model.release()

This is recommended for long simulations or batch runs.

---

6. Design notes (advanced users)

• All vectors are backed by natID's fast cnt::SafeFullVector<T>
• No implicit memory copies between Python and C++
• td::String (natID) is transparently convertible to/from Python str
• Dynamic solvers support implicit DAE solving
• SubModels allow hierarchical initialization and reuse

---

7. Requirements

• Python 3.13+
• Windows: visual C++ runtimes
• NumPy (optional, for post‑processing)
• OS‑specific native libraries bundled in internally

---

8. License & citation

If you use dTwin in academic work, please cite the accompanying paper (see arXiv link).

---

9. Summary

dTwin enables research‑grade nonlinear modeling directly from Python while retaining ultra-fast C++ performance.

It is suitable for:

• Digital twinning
• Simulation
• Dynamic simulation
• Power systems
• Control systems
• Nonlinear estimation

Happy modeling 🚀