In the development process of satellite telemetry data analysis software, there
are problems such as significant differences in data protocols among different
types of satellites, difficulty in converting data analysis rules into code, and
the need to reconstruct analysis code when using telemetry analysis software for
different types of satellites. This article proposes a code automatic generation
method that combines reference knowledge alignment and retrieval enhancement,
and constructs a closed-loop process that includes information extraction,
knowledge modeling, code generation, and iterative optimization. Extract packet
structure from the Word/PDF version of the data protocol document, extract field
offsets, data types, units, and formulas from the Excel rule table, and
reconstruct these information into telemetry data knowledge graph and rule
vector library. In order to alleviate the problem of lack of supervision
library, this method introduces manually extracted telemetry data parsing rule
documents and manually implemented satellite telemetry data parsing software
code as supervision information. This article is based on a total of 12026
telemetry parameters and 120 remote control commands from three types of
satellites, and adds satellite names, subsystems, and multi-level nodes for
single machine construction. The total number of nodes in the knowledge graph is
about 13800. After 10 iterations, the development time was shortened from 3
months to 3 weeks, and a total telemetry parameter coverage rate of 94.3% for
formulas D0001, F0010, and F0403 was successfully achieved.