I have a multivariate multi-step time series prediction problem, where I input waterA, waterB, waterC, medicineA, medicineB, and output turbidity. enter image description here Among them, medicineA and medicineB are controllable, while the three types of water are uncontrollable.
I use the LSTM model to predict the next 15 data points using the first 15 data points (assuming 15 data points represent 1 hour). Here are some results on Test set. enter image description here enter image description here If I predict the turbidity for the next hour based on past data at the current moment. But at this moment, I plan to change the amount of medication and predict the future turbidity further than an hour (mainly the impact of changing medication on future turbidity), but I do not know the future values of the three types of water. I once tried to use three LSTMs to predict three types of future water instead of future water inputs, but this would result in worse prediction results. Is there any other method?
