Decentralized Voltage and Var Control of Active Distribution Network Based on Parameter-Sharing Deep Reinforcement Learning

This work introduces a Parameter Sharing - twin-delay deep deterministic policy gradient (PS-TD3) method for carrying out decentralized voltage and var control in active distribution networks. The more renewable energies are integrated, the more Active Distribution Networks (ADNs) are at risk of instability. Most optimization techniques depend on accurate uncertainty models and take a long time to produce results. By allowing agents to share parts of their neural network, the proposed Parameter Sharing - twin-delay deep deterministic policy gradient algorithm improves the stability and efficiency of voltage regulation. Importantly, the framework makes it possible to add new actions, such as reducing Photovoltaic (PV) output, when traditional ways (adjusting reactive power and turning off wind power) are not enough. Also, when looking at the financial side of energy trading under net metering, Energy Storage System (ESS) operations are organized to even out the costs of charging and discharging. The results from IEEE 33-node simulations show that the approach reduces voltage fluctuations and lowers power loss better than other methods. The findings indicate that parameter-sharing helps agents work together, so large-scale Active Distribution Networks can use decentralized voltage control more efficiently. The framework allows the action space to change by including other control variables, such as reducing photovoltaic output when regular actions such as reactive power increase and wind curtailment do not support voltage stability.