Can this time series data be preprossed or not,if

Question

My mean purpose is to clean the data for forecasting (LSTM), and the data Daily bus speed data. However, when I do a lot of denoised methods, it seems the data can't be used, here is the original data:

---

And this is the first result I got by wavelet denoised method (do not drop those missing values):

---

But we can see in the original data, there have lots of periods the bus do not on the road (Those horizontal lines). I want got a more reliable data

So I clean and fill those missing value by use pd.loc and denoised by wavelet again, but the result I got do not have the valleys, it is not the result I want to see.

---

My question is: Can the data be used in this way or not? If it can, should I do the pd.loc or do I just ignore it use wavelet only. Thanks a lot.