Urban water distribution networks are crucial infrastructures for providing essential services to society, but their exorbitant costs and limited water resources make their optimization a critical research area. Optimal management and design of these networks can help to reduce costs and enhance their efficiency while meeting technical, economic, and quality standards. In particular, the management of network pressure is critical for reducing leakage in water distribution systems. Thus, this study aimed to investigate two objective functions for optimizing the water distribution network: (i) minimizing costs while considering the number of pressure-relief valves, and (ii) minimizing network pressure by observing the optimal pressure range. To achieve this, the Nondominated Sorting Differential Evolution (NSDE) multi-objective metaheuristic algorithm was employed as the optimization tool, and a computer program was written in MATLAB software for solving the optimization models. EPANET software was also used for hydraulic simulation of the water distribution network. The efficiency and capabilities of these models were tested on the case study of the third district of Mashhad in Iran. The results indicated that the installation and adjustment of pressure-relief valves in accordance with the positions and optimal settings of the output of the proposed models significantly improved the desired goals, particularly the average pressure of the network. As an example of optimization, the study achieved a 56.12% reduction in pressure compared to the case without a plan, considering five pressure-relief valves.

In the context of fuzzy logic speed control for switching reluctance motor (SRM) applications, the objective of this work is to propose a unique zone-based simplification technique. Using the procedure that has been outlined, it is made easier to reduce membership functions as well as rule sets in a logical manner. This is accomplished by splitting the error–change-of-error plane into discrete decision zones. This method is separate from heuristic or adaptive reduction strategies since it employs a systematic framework that reduces the number of rules from 49 in the standard design to 9 and 5 without compromising the accuracy of the control. This is accomplished without adversely affecting the performance of the control. The simplified controller that was produced as a consequence of this study decreases the amount of overshoot, enhances the speed at which a dynamic response happens, and makes it simpler to use on digital platforms that are affordable. All of these capabilities were achieved by the controller. Based on simulations and testing carried out in the real world, it has been determined that the zone-based simplified fuzzy controller that was proposed has a superior performance to traditional PID and full-rule fuzzy systems in terms of reaction time, stability, and energy efficiency. Taking all of this into consideration, it is evident that it has the potential to be useful in real-world applications for SRM drives that demand a high level of speed while maintaining a low cost factor.

Hybrid microgrids struggle to manage electricity due to renewable source, storage, and load demand variability. This paper proposes a centralized controller employing hybrid deep learning and evolutionary optimization to overcome these issues. Solar panels, BESS, EVs, dynamic loads, steady loads, and a switching main grid make up the hybrid microgrid. To capture spatial and temporal patterns, a centralized controller uses a deep learning model with a CNN–LSTM architecture. The imperialist competitive algorithm (ICA) optimizes neural network hyperparameters for more accurate controller outputs. The controller controls grid switching, voltage source converter power, and EV reference current. R² values of 0.9602, 0.9512, and 0.9618 show reliable controller output predictions. A typical test case, low sunshine, and no EV or BESS initial charging are validation situations. Its constant power flow, uncertainty management, and adaptability make this controller better than others. Even with intermittent energy and limited storage capacity, the ICA-optimized hybrid deep learning controller stabilized smart-grids.

The rising demand for sustainable energy in off-grid regions has propelled the use of solar photovoltaic (PV) systems for water pumping, particularly for irrigation. This study investigates the design and simulation of a standalone PV-powered water pumping system employing a three-phase induction motor (IM) for remote applications. The research aims to optimize energy extraction and ensure system stability under varying solar irradiance. The methodology utilizes MATLAB/Simulink to model the system, integrating a Perturb and Observe (P&O) Maximum Power Point Tracking (MPPT) algorithm, a DC-DC boost converter, and a Voltage Source Inverter (VSI) to drive the IM. A standalone solar PV array rated at 5.84 kW was designed to drive a 4-kW induction motor water pump. Under simulated dynamic irradiance conditions, the motor load temporarily dropped to 2 kW, demonstrating the system’s ability to maintain stable operation which reduces to 2.92 kW (half the rated capacity) during an irradiance drop from 1000 to 500 W/m² at t=5 seconds. The motor maintains a steady speed of 1430 rpm at a 4-kW load, with the P&O algorithm effectively adapting to dynamic conditions. The model acknowledges simplifications, noting that actual output may decrease by 10-15% due to temperature, inverter, and wiring losses. These findings affirm the system’s potential for off-grid irrigation, though further refinements are recommended to address real-world inefficiencies and enhance reliability.

This paper analyzes 3 different deep learning models including Gated Recurrent Unit (GRU), Nonlinear Autoregressive model with Exogenous Inputs (NARX) and one-dimensional Convolutional Neural Network (1D-CNN), used to predict monthly energy consumption in the USA between 1973 and 2022 years, with realistic dataset containing 600 months of data. In this experiment the dataset is split into 70% training, 15% validation and 15% test sets. The time series were transformed into supervised learning sequences, each consisting of a 12-month lookback periods. In NARX model, a feature engineering process implemented using three exogenous inputs generated for each time step including a 3-month moving average, trend and volatility. The GRU and 1D-CNN models, in contrast, were trained directly on the raw time series sequences. The experimental results indicating recurrent architectures being exceptionally well-suited for this task, with GRU model achieving highest performance scores with R² of 0.85 with slight overfitting gap of 0.124 R² between training and test sets, and Mean Absolute Percentage Error of 3.16. The NARX model performed nearly as well as GRU model, achieving an R² of 0.846 and MAPE of 3.18, in contrast to 1D-CNN, resulting in a substantially lower R² of 0.648.