LG M50 Single Pulse Parameter Identification¶

This example presents an experimental parameter identification method for a two-RC circuit model. The data for this notebook is located within the same directory and was obtained from [1].

Setting up the Environment¶

Before we begin, we need to ensure that we have all the necessary tools. We will install PyBOP for the fitting and Pandas for the data ingestion and manipulation:

Importing Libraries¶

With the environment set up, we can now import PyBOP alongside other libraries we will need:

Importing Data¶

The data is imported as a dictionary with the following key level:

• ["LGM50_5Ah_Pulse"]
  • Temperature-> ["T0"]
    • State of Charge ->["SoC3"]
      • Cell number -> ["Cell1"]
        • Corresponding data -> ["data"]

Using SciPy's loadmat, we import the data from the MAT containers:

Convert to Dataframes¶

Next, we construct a dataframe from a selected pulse. In this case, we select the data for zero degrees (T0) with a state-of-charge of 90% (SoC9) and the 19th cell (Cell19). This can be extended to multiple pulses, or a different selection.

Additionally, we apply two filters to the dataframe to ensure the data contains only monotonically increasing time samples without duplicates.

A plot of time vs voltage confirms the data looks correct for fitting. In this situation, we would prefer to have additional samples from the relaxation, but as we will show below, PyBOP is still able to identify parameter values that fit this system.

Next, we construct the OCV function from the imported OCV data. This is completed with a wrapper method on the pybamm.Interpolant function,

We can construct the two RC parameter set with initial values as listed. Note, the initial SOC is shifted slightly to better match the zero degree data.

Identifying the Parameters¶

Now that the initial parameter set is defined, we can start the PyBOP fitting process. First, we construct the model class with two RC elements and the corresponding parameter set.

In this example, we are going to try to fit all five parameters at once. To do this, we define a pybop.parameter for each fitting parameter as,

We can now form the pybop.Dataset from the experimental data. In this situation, the applied current convention is different between the experimental data and PyBOP's convention. This is solved by negating the vector on construction.

The FittingProblem class provides us with a class to hold all of the objects we need to evaluate our selected SumSquaredError cost function.

The cost function can be interrogated manually via the cost([params]) API. In this example, that would look like the following,

Next, we construct the optimisation class with our algorithm of choice and run it. In this case, we select the PSO method as it provides global optimisation capability. After optimisation, we can compare the initial parameters to the identified parameters.

Plotting and Visualisation¶

PyBOP provides various plotting utilities to visualise the results of the optimisation. The quick_plot method provides a fast scatter of two variables. For this example, we plot the identified parameter set to the target dataset. The default plot is voltage vs time as shown below.

Convergence and Parameter Trajectories¶

To assess the optimisation process, we can plot the convergence of the cost function and the trajectories of the parameters:

Validating the Fit¶

We've identified parameters from a single discharge pulse. Now, let's ask ourselves how well these parameters extrapolate to other operational conditions. To accomplish this, let's load a new pulse dataset at 80% SOC and compare it to our identified forward model.

Next, we construct a new pybop.Dataset from the second pulse data,

Now that we have a new dataset, we update the target within the problem class as well as the Initial SoC value. Once that has been completed, we rebuild the model.

Let's plot the parameterised forward model against the new pulse data:

As expected, when identifying parameters from a single pulse, extrapolation to different operating conditions is challenging. To solve this issue, parameter identification with various pulse datasets is recommended.

Conclusion¶

This notebook illustrates how to perform circuit model parameter identification using PSO in PyBOP, providing insights into the optimisation process through various visualisations.