Deep learning model for multiple sparse time series

dI have a dataset with multiple (about 50) sparse time series (of 30 days), of 30,000 people.

The data is sparse and many of the series are like (12.3,0,0,0,…,21.5,0,0,…0) : it contains many zeros, and I want to predict a single value using these sparse time series. (not forecasting the next value of the time series, single value using all the series)

The length of sequences are kind of short, so I couldn’t just make the daily sequence to weekly or monthly.

At first I have used some usual sequential models like LSTM or GRU, but the result didn’t seem like good enough.

I also tried some event-sequence based models : treating non-zero values as occurrence of an event, but those models consider the events like 0-1 : couldn’t take use the numerical values of the event (like 12.3, 21.5 on the example above)

Is there model or some methods that can handle this kind of dataset?