The distributed photovoltaic is connected to the distribution network dispersedly

whose anti-regulation characteristic of output and load can easily cause over-voltage at points of common coupling. Existing methods optimize the voltage profile to some extent through reasonable energy storage configuration

but the mobilization of stored energy is highly coupled with other reactive power sources in the time domain

increasing the complexity of combined voltage regulation modeling and affecting the voltage regulation precision. The rapid and frequent charging and discharging

as well as overcharging and over-discharging will lead to lifespan degradation of energy storage system

thereby reducing the economic and safety performance of the system. Aiming at these issues

this paper proposes a voltage optimization strategy for the photovoltaic and energy storage (PV-ES) distribution network with multi-timescale deep reinforcement learning (DRL) method. Integrating with DRL's model-free feature

this strategy considers the action characteristics in time domain of various discrete/continuous voltage regulation equipment

photovoltaic inverters as well as energy storage devices

whose action and the state of charge (SOC) are restricted strictly based on the event-triggering mechanism. The voltage optimization scheme for distribution network is designed from a multi-timescale perspective

which aims to minimize voltage deviation and network loss. Finally

the proposed method is simulated and verified based on an modified IEEE 33-bus system with distributed photovoltaic

energy storage

and other voltage regulation devices. The results show that the proposed method exhibits smaller voltage deviations and average network losses compared with the results in reference

measuring 0.016 6 pu and 141.2 kW respectively

and more suitable for application scenarios with relatively high uncertainty in photovoltaic output. At the same time

it effectively limits the cutting action of on-load tap changers and capacitor banks to 6 and 3 times

respectively. Moreover

it also restricts SOC variation of energy storage devices within 15 minutes to below 30%

significantly reducing the action frequency to less than 10% of continuous situation

and maintaining the SOC within a safe range of 0.2~0.8.