Numeric Simulator for state space models

It simulates ordinary differential equations with a state space representation and external input. scipy.integrate.solve_ivp integrates continuous systems. Discrete systems are iterated with a fixed step size.

State space description, given an initial condition $x(0)$ and a fixed time horizon $T$, for discrete system with a step size $\eta$

• for linear continuous models ($t = [0, T]$): math \begin{pmatrix} \dot{x}(t) \\ y(t) \end{pmatrix} = \begin{pmatrix} A & B \\ C & D \end{pmatrix} \begin{pmatrix} x(t) \\ u(t) \end{pmatrix}
• for nonlinear continuous models ($t = [0, T]$): $$\dot{x}(t) = f(x(t), u(t)),\qquad y(t) = g(x(t), u(t))$$
• for linear discrete models ($k = 0, \ldots, (T/\eta)-1$): math \begin{pmatrix} x(k+1) \\ y(k) \end{pmatrix} = \begin{pmatrix} A & B \\ C & D \end{pmatrix} \begin{pmatrix} x(k) \\ u(k) \end{pmatrix}

Simulation

The continuous simulator will evaluate the function depending on the method chosen. The input sequence is always discrete with a constant distance of step_size. The output sequence has the same step size as the input sequence. Therefore the result of the continuous integrator is resampled (or evaluated) to the step_size.

Systems

The systems that can be simulated with statesim are described by nonlinear differential equations. Each system has a nonlinear symbolic expression that can be used for simulation and is considered the ground truth data. From the nonlinear symbolic expressions, linearizations can be derived and evaluated at an equilibrium point with SymPy.

Currently, the following systems are implemented - CartPole: Zero input, the initial state is around the upright position of the pole (Barto AG, Sutton RS, Anderson CW. Neuronlike adaptive elements that can solve difficult learning control problems. IEEE Transactions on systems, man, and cybernetics. 1983 Sep(5):834-46.)

• Coupled mass spring damper system: states of 4 coupled masses

• Inverted pendulum with torque input: Zero input, the initial state is around the upright position of the pole

Input sequences

Random input sequences can be generated to excite the system.

Currently, the following types of input generation are implemented: - Random Static: Static inputs that jump to another static value after a random time from a given interval.

Generate data for a continuous linear system

To generate a dataset for a continuous linear system, you can use the script run_datagen_linear.py. This will use the matrices defined in config/linear.json. The following output is generated for a double integrator with Gaussian measurement noise. The input sequence is a random static sequence in the range $u \in [-1, 1]$

Example

In scripts/notebooks, some examples of how to use statesim are shown in jupyter notebooks. To generate .csv files from the simulations, the script scripts/run_cartpole_datagen.py offers this for the cart pole example. The configuration can also be external as a .json file with the fields: json { "result_directory": "str, Directory where the .csv files will be stored, must exist", "base_name": "str, Folder name within the result directory, will be created", "seed": "int, a random seed that will be set to numpy", "K": "int, Number of samples", "T": "float, Simulation end time start is always 0, e.g. [0, T]", "step_size": "float, step between two measurements", "input": "Optional: InputGeneratorConfig, configuration for generating random input sequences, if not defined there will be no input", "system": "SystemConfig, specific configuration for the system to be simulated", "simulator": "SimulatorConfig, configuration for the simulator, requires an initial state" "measurement_noise": "Optional: NoiseConfig, configuration for measurement noise, if not defined there will be no measurement noise" }