I am developing a solution for modelling industrial processes, where one of the goals is to predict a situation given the evolution of vectors of values collected in the production line over the time.
For example, let say that I have three modules in that production line for which I can generate State Vectors at a sampling time t (SV1t, SV2t, SV3t). I am generating a graph that connects a node ( it is a subgraph but for keeping it simple let’s say a node) representation of each module. I then connect Modules nodes to establish the process.
Let’s say that the output of that production line is also measured as another vector F (final result) .
The goal of this model is to answer: How would you predict F given the ((SV1 (t), SV1 (t-1), SV1 (t-2), SV1 (t-3), SV1 (t-4), …, SV1 (t-n)), (SV2 (t), SV2 (t-1), SV2 (t-2), SV2 (t-3), SV2 (t-4), …, SV2 (t-n)), (SV3 (t), SV3(t-1), SV3 (t-2), SV3 (t-3), …, SV3 (t-n))?
So that is the main problem for which I have few questions:
- Would you use neo4j for modelling the whole structure needed for the given problem? Or would you develop a hybrid model in which the timeseries would be in a K,V database for example, and the process on a Neo4j graph? Or would you avoid graph at all?
- If it was a full neo4j database, what recipe of algorithms would you use for timeseries prediction? I do not find any clear recommendation on how to do it in graphs, which might be part of the answer . But I prefered to ask the experts first.
Thanks you so much for taking the time to read this and hopefully provide me guidance.