Table 4 Secure control operation techniques
Type of control operation | Data-driven techniques | Applied system | Robustness | Advantages | Problems and challenges |
|---|---|---|---|---|---|
Frequency control [101] | IRL (Internal Reinforcement Learning) | Single Area System | Yes | Tackles unknown dynamical environments, and acute load deviations | Weak results may occur due to overloading |
Frequency control [102] | Genetic algorithm | New England 10 generators (applied for 1st test), 39-bus system (2nd test) | Yes | (1) Simple design, (2) Flexibility of performance, (3) Reaches steady state faster, handles discontinuities, and noisy functions | (1) Mutation operator may cause legal issues, (2) Repair function has to be applied individually |
Frequency control [103] | Neuro Hybrid Fuzzy Logic | Four area (Inter connected Hydro Thermal) power system | N/A | (1) Fastly controls the non-linearity, (2) Lessens FPD (frequency peak deviation), time imprecision, and tie-line power | Needs regular update for proper functioning |
Frequency control [104] | RADP (Robust Approximate Dynamic programming) | Multi machine power System (New England 10-machine 39-bus system) | N/A | Recovers system’s frequency | System dynamics is unknown at the time of training datasets |
Preventive [105] | MLP (Multi-layer perceptron) | (IEEE 6, IEEE 118), Real 33-bus Bodaibo subsystem | No | Prevents voltage instability and longer overload, and has smart preventive security | Independent variables are affected by dependent variables to an unknown extent |
Restorative [106] | Q learning | IEEE 57-bus, IEEE 118-bus and IEEE 300-bus systems | Yes | (1) Avoids risk of new failures (cascading), (2) Restores with better power grid topology | (1) Works on one individual power grid at a time, (2) Takes high computational time when recovery sequence increases |
General control [107] | Q-learning, WoLF-PHC WoLF (win or learn fast) PHC (policy hill climbing) | Secondary users (SUs) | No | Improves anti-jamming performance | Works better with only discrete and finite datasets |
General control [108] | Wolf-Pack | Multi-agent systems | Yes | (1) Efficient coordinated control, (2) Virtual tribes control optimizes electricity generation as wind, solar, and electric vehicles increase | Time delay occurs while obtaining optimal strategy |
General control [109] | Lazy Learning | Parallel cyber physical–social energy systems | No | (1) Highest control performance, (2) Reduced frequency deviation, (3) Predicts next state | Lacks optimization algorithm |
Voltage control [110] | Teaching–learning, Sugeno fuzzy logic | AVR system | Yes | Delivers adequate robust performance and satisfactory dynamic responses over parametric fluctuations of system | Cannot overcome OPD (optimal power dispatch) problems |
Voltage control [111] | DQN/DDPG | 200-bus test system | No | Automatically and efficiently control voltage according to situation | Lacks multiple control operation |