Load forecasting is of crucial importance for smart grids and the electricity market in terms of the meeting the demand for and distribution of electrical energy. This research proposes a hybrid algorithm for improving the forecasting accuracy where a non-dominated sorting genetic algorithm II (NSGA II) is employed for selecting the input vector, where its fitness function is a multi-layer perceptron neural network (MLPNN). Thus, the output of the NSGA II is the output of the best-trained MLPNN which has the best combination of inputs. The result of NSGA II is fed to the Adaptive Neuro-Fuzzy Inference System (ANFIS) as its input and the results demonstrate an improved forecasting accuracy of the MLPNN-ANFIS compared to the MLPNN and ANFIS models. In addition, genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization (ACO), differential evolution (DE), and imperialistic competitive algorithm (ICA) are used for optimized design of the ANFIS. Electricity demand data for Bonneville, Oregon are used to test the model and among the different tested models, NSGA II-ANFIS-GA provides better accuracy. Obtained values of error indicators for one-hour-ahead demand forecasting are 107.2644, 1.5063, 65.4250, 1.0570, and 0.9940 for RMSE, RMSE%, MAE, MAPE, and R, respectively.
The use of photovoltaics is still considered to be challenging because of certain reliability issues and high dependence on the global horizontal irradiance (GHI). GHI forecasting has a wide application from grid safety to supply–demand balance and economic load dispatching. Given a data set, a multi-layer perceptron neural network (MLPNN) is a strong tool for solving the forecasting problems. Furthermore, noise detection and feature selection in a data set with numerous variables including meteorological parameters and previous values of GHI are of crucial importance to obtain the desired results. This paper employs density-based spatial clustering of applications with noise (DBSCAN) and non-dominated sorting genetic algorithm II (NSGA II) algorithms for noise detection and feature selection, respectively. Tuning the neural network is another important issue that includes choosing the hidden layer size and activation functions between the layers of the network. Previous studies have utilized a combination of different parameters based on trial and error, which seems to be inefficient in terms of accurate selection of the desired features and also tuning of the neural network. In this research, two different methods—namely, particle swarm optimization (PSO) algorithm and genetic algorithm (GA)—are utilized in order to tune the MLPNN, and the results of one-hour-ahead forecasting of the GHI are subsequently compared. The methodology is validated using the hourly data for Elizabeth City located in North Carolina, USA, and the results demonstrated a better performance of GA in comparison with PSO. The GA-tuned MLPNN reported a normalized root mean square error (nRMSE) of 0.0458 and a normalized mean absolute error (nMAE) of 0.0238.
The discrete-time neural network proposed by Hopfield can be used for storing and recognizing binary patterns. Here, we investigate how the performance of this network on pattern recognition task is altered when neurons are removed and the weights of the synapses corresponding to these deleted neurons are divided among the remaining synapses. Five distinct ways of distributing such weights are evaluated. We speculate how this numerical work about synaptic compensation may help to guide experimental studies on memory rehabilitation interventions.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.