Myself I would think that the predictive part is either very hard or impossible.
However other than that you have described a rules engine. You persist rules and then run the rules against a set of data, where the set is defined by each rule. The analyst or even operators create, update and delete rules. How you do that depends on how much time you want to invest in the system. But ultimately you would have a GUI or even an IDE that would be used to create a rule.
Other complexities involved are
- Can you find an existing rules system or must you build your own.
- How complex do you want to allow the rules to become. The more complex they are the harder it is for a new person to learn how to use it.
- Will you and must you be prepared to build plug in rule modules perhaps either for performance or complexity needs. If you must then how will the process for that proceed.
With only thousands of alarms in a day a very fast system might not be a requirement but if it does then that also becomes a factor in the above.