A memristor compact model, which can both capture state volatility and describe short-term and long-term memory transitions, is introduced. The model is based on an energy landscape acting as a pseudo-potential which generates the driving forces for configurational changes. A stable conductance change in this model is implemented through a sequence of transitions between states of plasticity occurring over different time-scales. Such transitions also modify the detail of the pseudopotential landscape, this way altering the probability distribution of subsequent state-change events. This approach departs from the usual method of applying perfectly non-volatile increments on the state variable. The model has been coded in Verilog-A, so that it can be used in many popular SPICE engines. The proposed model is semi-quantitatively fitted to measurements taken on Pt/TiO2/Pt stack memristor devices.