In this dissertation, we present a cost-effective, neural network-based technique for data sanity checking and small system parameter monitoring which utilizes the contextual information in which data is collected to avoid the need for multiple metering. Multiple metering is not always a feasible nor an optimal solution to the problem. In an environment where it is necessary to monitor a large number of different physical variables, the mere installation and maintenance of multiple metering equipment can prove to be very costly. Moreover, multiple measurements of the same quantity result in a phenomenon known as data explosion. Context-based sensoyvalidation is achieved through cross sensor redundancy, which is not to be confused with metering redundancy. Neural networks are used to model the relationships among the various parameters and to provide context-based estimates which help in identifying sensor (versus system) malfunction. Slow tracking of the relationships among the parameters as they change over time is made possible through on-line training of the neural networks on the most recent data. This helps to account for the dependency of the relationships among system parameters on the range of external variables such as ambient temperature.

A prototypical system titled DASANEX is implemented to illustrate the validity of the technique. The system is used to monitor and filter real-time transformer and ambient temperature data. A proof-of-concept is established using field data from the city of Martinsville Electric Department. Results prove the superior ability of the technique to identify sensor malfunction and to provide real-time adequate replacement values during short downtimes of the sensors even when some sensor data are missing or contaminated.