摘要：This paper presents a framework for the collection, management and mining of sensor data in large cyber-physical systems. Particular emphasis has been placed on mathematical methods, data structures and implementations which enable the real-time solution of inverse problems associated with the system in question. That is, given a system model, to obtain an estimate for the phenomenological cause of the sensor observation. This enables the use of causality, rather than mere correlation, when computing measures of significance during machine learning and knowledge discovery in very large data sets. The model is an abstract representation of a real physical system establishing the relationships between cause and effects. The pertinent behaviour of the model is captured in the form of equations, e.g., differential equations. The inverse solution of these model-equations, within certain constraints, permit us to establish the semantic reference between the sensor observation and its cause. Without this semantic reference there can be no physically based knowledge discovery. Embrechts pyramid of knowledge is addressed and shown that it will not suffice for future developments. The issue of information content is addressed more formally than in most data mining literature. Additionally the Epistemology for the emergent-perceptive portion of speech is presented and a prototype implementation with experimental results in data mining are presented. A lexical symbolic analysis of sensor data is implemented.
