Synaptic transmission is temperature-sensitive. However， due to lack of a temperature-dependent synaptic conductance model， it is difficult to study the temperature effects when modeling nervous systems. The memristor is widely considered to be ideal bionic synapse due to its continuously variable resistance and nanometer size. This paper proposes a new tungsten oxide memristor model by improving the memristive retention value and taking into account the effect of temperature on ion migration and diffusion， which theoretically captures the actual behaviors of the memristor. In addition， this memristor is used as a bionic synapse to couple two identical HH neurons， and this system can reflect the influence of temperature on synaptic transmission. Simulation studies demonstrate that， as the temperature rises， the rate of oxygen vacancy migration and diffusion increases， and hence accelerates the memristor conductance change rate， and further affects the amplitude of excitatory postsynaptic membrane potential and the number of postsynaptic neural firings， all of which accords well with biological experimental findings. This paper shows that the improved tungsten oxide memristor model is more suitable for simulating neural synapse in neuromorphic systems， and hence could guide design and manufacturing of memristor to improve its bionic performance， and this model also provides a new idea for studying the effect of temperature on synaptic transmission.