You should always start with an extremely simple forecast, like the historical average. This can already be surprisingly hard to beat.
After that, as user 2974951 writes, you could apply Croston's method, which is tailored for intermittent-demand, i.e., demand with many zeros. It's not sound statistically, but it's a frequent benchmark (and implemented in most software packages).
You probably have seasonality - if only an intra-daily seasonality, with people probably renting more at certain times of day. Take a look at Seasonal Exponential Smoothing.
However, this simple seasonality is probably intermixed with an intra-weekly seasonality, with weekends probably differing from weekdays. This is a case of multiple-seasonalities. The tag wiki contains pointers to special models for this, like TBATS.
As seanv507 writes, you may also have effects of holidays, which you could model with a regression on holiday dummies. You could either use weekday and hour dummies (potentiall interaction terms between them) to capture the seasonalities, or run a regression on holidays and model residuals using a (multiply) seasonal time series method.
Note that these are in increasing order of complexity. Per above, in forecasting simple methods often work surprisingly well.
Finally, note that your choice of error measure will influence what the best forecast is. If you use the MAE (which elicits the conditional median), a forecast that is biased low will look better than one which is an unbiased expectation forecast - which all the methods above aim for. So I would very much recommend using the MSE, or RMSE. More information on this effect, with pointers to literature, can be found in section 2.12.2 in Petropoulos et al. (2021), "Forecasting: theory and practice".
Finally, I recommend this excellent free online forecasting textbook: Forecasting: Principles and Practice by Athanasopoulos & Hyndman, 2nd ed. or 3rd ed.
Best Answer
You can use generalized ARMA (GLARMA) models. See, for example, Kedem and Fokianos (2002), Regression Models for Time Series Analysis.
See also R package glarma (on CRAN)