Subspace Identification with Multiple Data Records: unlocking the archive
For control engineers and data analysts, this method enables system identification from fragmented operational data archives, solving a practical bottleneck in real-world applications.
The paper develops a subspace system identification method that uses multiple non-contiguous data records, enabling identification from archives where individual records are insufficient. It provides a rank-based test to determine if the combined data uniquely identifies the system model.
We develop an approach to subspace system identification using multiple data records and present a simple rank-based test for the adequacy of these data for fitting the unique linear, noise-free, dynamic model of prescribed state-vector, input-vector and output-vector dimensions. The approach is motivated by the prospect of sorting through archives of operational data and extracting a sequence of not-necessarily-contiguous data records individually insufficient for providing identifiability but collectively making this possible. The test of identifiability then becomes the sorting criterion for accepting or rejecting new data records. En passant, the familiar Hankel structure of the data matrices of subspace system identification is reinterpreted and revised.