Intelligent deep reinforcement learning based SMCF boundaries—An effective assessment of integrated network operating state

Abstract

In this paper, applied energy system with multiple distributed generation (DG) units like solar PV, wind firm and biomass energy has been assessed and its operating state is estimated. Deep reinforcement learning (dRFL) based sensitive modulation control factor (SMCF) is proposed for assessment of entire applied energy system operating state during the utility grid connected or grid isolated mode/islanding mode. Grid isolated/island state identification is essentially required in integrated power grid when operating with DG to avoid the hazardous to human begins, as well as power electronic components. Time interval of distorted signal due to disturbance in integrated power grid has been taken for SMCF control. Variations in voltage (VRV), variations in frequency (VRF) beyond their limits i.e. lower limit (LL) and higher limit (HL) and these are considered to obtain SMCF in-phase component and quadrature phase component of circuit. Instant of SM control, tsm for a UV or OV are calculated and it is updated by the change in VRV. Obviously VRF also updates the instant of SM control and hence SMCF control factor is obtained. In-phase control variable of SMCF circuit i.e. αs_d with boundary limits i.e. αs_dLL and αs_dHL are used for the every 10 ms interval period/range to identify data of event i.e. either in island event (ISE) or non island event (NOIE). Now the dRFL is used for αsd training with a deterministic policy gradient (DMPG) evaluation. For a HL targets, two segment limits i.e. τ1, τ2 are taken for error minimization in dRFL training. According to first target; error Eα1 is validated the test condition of τ1<τ2 in a proposed dRFL training. Also error Eα2 is validated the test condition of τ1≥τ2 for second target in a proposed dRFL training. The completion of dRFL training gives the αHL and hence minimization of EαHL as per objective will be obtained. Error minimization of reinforcement learning (RFL) has been obtained by DMPG for every target and action of energy system operating state. Effective dRFL based SMCF control algorithm introduced to validate every ISE, NOIE including lower mismatch scenarios and avoid the false tripping signal.