Hybrid Optimization Boosts Simulated Benchmark Power Grid Loadability to 165 Percent

An artificial intelligence framework that couples self-adaptive metaheuristic algorithms with surrogate neural networks has successfully increased simulated power transmission loadability to 165 percent while significantly suppressing line losses during heavy wind generation, according to research published in Discover Artificial Intelligence. Developed by an engineering team including Bindu Vadlamudi, Indira Damarla, Subhojit Dawn, Hassan Abdurrahman Shuaibu, and Taha Selim Ustun, the computational scheme optimizes FACTS device allocation under varying levels of wind energy. Voltage constraints and stability issues often limit the full utilization of transmission line capacity. System stability, power transmission capacity, and power flow improvement through reduced line losses can be achieved by altering transmission parameters using FACTS devices. However, existing optimization procedures suffer from a lack of coordination between optimization and intelligent decision-making techniques and early convergence in determining the optimal placement of FACTS.

To overcome these limitations, the research team designed the Optimal FACTS Device Allocation with Wind Energy framework, denoted as OFDA-WE, creating a tightly coupled dual-stage architecture that combines the Self-Adaptive Bat Optimization algorithm with a specialized Chaotic Fuzzy Multi-Model Neural Network.

The first structural pillar of the OFDA-WE framework is the Self-Adaptive Bat Optimization algorithm, a nature-inspired metaheuristic grounded in the echolocation dynamics utilized by microbats to navigate obstacles and capture prey. In the mathematical formulation of the algorithm, each individual bat represents a candidate vector of power system decision variables, encompassing generator active power outputs, bus voltages, transformer tap ratios, and specific FACTS device compensation settings. As candidate bats navigate the multidimensional search domain, their velocities and positions are dynamically adjusted according to frequency vectors that control step size. Unlike standard bat algorithms that rely on rigid, predetermined tuning parameters, the self-adaptive formulation continuously updates pulse emission rates and acoustic loudness through embedded feedback mechanisms. As prospective solutions demonstrate superior objective values, pulse emission frequencies adaptively rise while loudness values systematically decay, gracefully transitioning the search profile from wide global exploration to focused local exploitation. By incorporating self-adaptive control parameters with learning rates set at 0.1, the algorithm ensures that candidate solutions maintain sufficient diversity to avoid stagnation while preserving optimal configuration profiles across successive iterations.

To bypass the massive computational burden of executing full numerical optimal power flow evaluations for thousands of candidate bat positions, the researchers integrated a Chaotic Fuzzy Multi-Model Neural Network to function as a fast surrogate fitness evaluator. The neural network features a single hidden-layer feed-forward topology comprising input neurons, hidden fuzzy nodes, and output neurons. Input neurons receive crisp system state parameters, including bus voltage magnitudes, transmission line loading percentages, active and reactive power generation figures, wind power output, FACTS device locations, and aggregate system load. Each hidden neuron maps onto a distinct fuzzy set in multidimensional space, utilizing a Gaussian membership function governed by tunable center and spread parameters to calculate intermediate membership values. The neural network integrates an angle-based chaotic search mechanism to improve search capability and reduce premature convergence to local optima. By optimizing parameters via the scaled conjugate gradient algorithm, the network learns the nonlinear relationship between system operating conditions and objective values, rapidly predicting candidate fitness without complete optimal power flow calculations.

The surrogate neural network was trained using simulation data generated from repeated power flow analyses of the IEEE 30-bus and IEEE 118-bus systems. The researchers assembled 5,000 distinct simulation samples generated through systematic, repeated power flow analyses performed on benchmark power systems under divergent loading conditions, varied renewable generation outputs, and contrasting device settings. This comprehensive dataset was divided into 70 percent for network training, 15 percent for cross-validation, and 15 percent for unseen holdout testing. Using these simulation samples, the fuzzy neural model learned to rapidly predict the fitness of candidate FACTS allocations. When embedded within the larger OFDA-WE pipeline, the bat algorithm generates prospective device coordinates and settings, while the chaotic fuzzy network immediately estimates the candidate’s fitness score. The bat population uses these instantaneous surrogate rankings to update positions, drastically reducing overall computational overhead and preventing premature convergence during multidimensional optimization searches.

To establish realistic renewable operating conditions, the research team developed a dynamic wind power generation model directly within the MATLAB and Simulink software environments, bypassing the need for external static weather datasets. The modeled wind generation facility comprises three paired sets of 1.5-megawatt Squirrel Cage Induction Generator wind turbines, delivering a combined 9 megawatts of active generation capacity. This facility interfaces with a 25-kilovolt distribution network and feeds power into a 120-kilovolt transmission grid across a 25-kilometer overhead line. Under dynamic simulation at a rated wind velocity of 11 meters per second, each turbine pair reaches its steady-state rating of 3 megawatts within approximately 8 seconds, while mechanical pitch control angles adjust between 0 and 8 degrees and turbine rotor speeds stabilize between 1.0028 and 1.0047 per-unit. At the primary interconnection bus, the simulated facility delivers 9 megawatts of real power while maintaining a bus voltage of 0.984 per-unit with 1.62 megavolt-amperes reactive of local support under a network operating base load of 283.4 megawatts and 126.2 megavolt-amperes reactive on a 100-megavolt-ampere base.

The team first evaluated the optimization framework on the standard IEEE 30-bus benchmark system, which consists of 30 electrical buses, 41 transmission branches, 6 conventional generating units, and 24 individual load buses. To represent moderate renewable penetration, two wind generation farms providing a combined nameplate capacity of 250 megawatts were integrated at bus 5 and bus 13. The multi-objective optimization problem sought to simultaneously minimize real power losses, maintain voltage profiles within strict boundaries, maximize system loadability, and maximize wind energy utilization. Under uncompensated base conditions, several heavily loaded buses in the network experienced significant voltage drops, threatening system operational security. Applying the OFDA-WE scheme successfully elevated minimum bus voltages to 0.984 per-unit, keeping the entire network profile tightly clustered around the nominal 1.0 per-unit benchmark. In comparative evaluations against Dynamic Line Rating protocols and alternative heuristic formulations like FDAWE-HCI, the proposed method achieved a maximum system loadability of 160 percent, outperforming the 150 percent recorded for Dynamic Line Rating and 155 percent for FDAWE-HCI, while reducing active power losses to 3.21 megawatts and boosting wind energy utilization to 97 percent.

Following successful validation on the medium-scale benchmark, the researchers scaled their evaluation to the substantially larger and more complex IEEE 118-bus transmission system. This extensive benchmark network contains 185 transmission branches and 19 conventional synchronous generating units. To test operational resilience under severe stress, peak system electrical demand was scaled upward to 1.2 times baseline reference values, and three major wind farms with an aggregate generation capacity of 1,600 megawatts were introduced near buses 5, 26, and 91. The optimization formulation considered ten distinct candidate transmission lines for the placement of series and shunt FACTS devices, specifically Thyristor-Controlled Series Capacitors and Phase Shifting Transformers. Permissible operational reactances for the series capacitors were constrained within negative 0.7 to positive 0.2 times the natural line reactance, while phase shifting transformer angles were limited to plus or minus 15 degrees. Operational limits required all bus voltages to remain strictly between 0.95 and 1.05 per-unit, with transformer tap positions confined within 0.90 to 1.30 per-unit.

Simulation runs across seven comprehensive operating scenarios on the 118-bus network demonstrated consistent technical advantages for the OFDA-WE methodology across differing engineering objectives. When parameterized specifically to prioritize Maximum System Loadability, the proposed algorithm identified a solution that pushed total system loadability to 170 percent with aggregate transmission losses of 5.12 megawatts, substantially eclipsing the maximum loadability achieved by Dynamic Line Rating at 144 percent with 2.93 megawatts of loss and FDAWE-HCI at 152 percent with 7.05 megawatts of loss. When evaluated under a Best Compromise Solution designed to strike an optimal balance between load growth and power dissipation, OFDA-WE achieved 165 percent loadability while keeping total transmission losses at 4.86 megawatts. In a third operational regime focused purely on loss minimization, termed the Best Loss solution, the proposed method lowered network transmission losses to 2.41 megawatts, proving superior to both Dynamic Line Rating at 2.93 megawatts and FDAWE-HCI at 7.05 megawatts under equivalent high-load stress conditions.

A critical operational factor driving these performance gains was the algorithm’s ability to coordinate intelligent active power redispatch across conventional generation units while simultaneously allocating reactive power compensation. Examining dispatch schedules showed that the OFDA-WE framework allocated higher active power outputs to selected generator buses while satisfying system operating constraints. Active power production at generator bus P2 increased from 80 megawatts under Dynamic Line Rating and 89 megawatts under FDAWE-HCI up to 95.32 megawatts under the proposed optimization. Similarly, generation outputs at buses P5, P8, P11, and P13 were rescheduled to 61.08 megawatts, 49.76 megawatts, 41.07 megawatts, and 29.15 megawatts, respectively. Furthermore, comparative hardware evaluations revealed that installing Static Var Compensators yielded superior reactive voltage support compared to alternative configurations using Thyristor-Controlled Series Capacitors or Static Synchronous Compensators alone, achieving up to 88 percent wind energy utilization efficiency under intense penetration scenarios due to fast, localized voltage stabilization.

Despite these documented technical achievements, the authors noted specific economic trade-offs and structural boundaries that frame the interpretation of their findings. The primary trade-off identified in the study involves capital expenditure: the overall investment cost for FACTS hardware under the OFDA-WE approach is higher than under simpler allocation strategies, because supporting loadability levels up to 165 and 170 percent necessitates installing devices with substantially larger operating capacities. The authors noted that the higher investment cost is justified by improvements in transmission capacity, voltage regulation, and overall system performance. Methodologically, the entire research framework was executed within MATLAB and Simulink simulation environments without integration of empirical telemetry from operational power grids or field measurement campaigns. Consequently, the reported efficiencies reflect deterministic mathematical models rather than physically measured field deployments.

Subject of Research: AI-driven FACTS device placement and wind energy integration in electrical power grids

Article Title: AI-driven multi-objective optimization of FACTS device placement using self-adaptive metaheuristics and neural networks

Article References: Vadlamudi, B., Damarla, I., Dawn, S., Shuaibu, H. A., & Ustun, T. S. (2026). AI-driven multi-objective optimization of FACTS device placement using self-adaptive metaheuristics and neural networks. Discover Artificial Intelligence, 6(1), Article 1831. https://doi.org/10.1007/s44163-026-02224-w

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02224-w

Keywords: FACTS device allocation, Self-Adaptive Bat Optimization, Chaotic Fuzzy Neural Network, wind energy integration, transmission system loadability, power flow optimization, voltage stability, transmission loss reduction, IEEE 30-bus system, IEEE 118-bus system, Static Var Compensators, Thyristor-Controlled Series Capacitors, active power dispatch, metaheuristic optimization, dynamic line rating, surrogate fitness modeling, renewable energy curtailment