A model of a power transformer describes a transfer behavior for input data into output data as a function of model parameters. In successive time windows, measurement dataset is received from first and second sensors of the power transformer. The model parameters are optimized by executing, for each of the time windows, the following group of steps a) to c), repeatedly: a) determining the output of the model using the input data defined by first sensor data, for the first execution of predefined parameters are used; b) determining a target value, which includes at least one squared error, weighted by a first weighting factor, the squared error is between the second sensor data from the measurement dataset assigned to the given time window and a previously determined output data, and c) determining optimization parameters as new model parameters based on the target value.