[worked on it for the last month]
predicted value (demand for heat in a district heating system)(*) depends on:
-hour of the day
-day of the week
-past pattern (of the demand)
I think that I know how to feed the Keras LSTM network with a 3D input matrix with series of data that contain past: weather,demands and days and hours and adjust with it the future output demands (2D matrix e.g for next 24h) Here is a hint how to do it
However, there is a problem while I want to add the weather forecast to the input matrix, because I cannot fit it to the matrix. To do it I would need to make room for the weather forecast. Basically I would need to leave the past demands untouched while switching the weather (with corresponding hours and days) in reference to the past demands. In other words the weather with hours and days would be lagging in reference to the past demands (by for example 24 hours)
Is it the correct way to do it? Please help. I'm a student and have nobody to help me with this.
*I have past demands with weather for the last two years with 1 hours resolution.
*I've planned to model the demand in the 4 month period during winter.