Nonlinear Autoregressive Neural Network Models for Prediction of Transformer Oil-Dissolved Gas Concentrations

In this paper five neural network models were developed using NARX-SP neural network type in order to predict air pollutants concentrations (SO2, PM10, NO2, O3 and CO ) for the 72nd hour ahead for Sarajevo. Hourly values of air pollutants concentrations and meteorological parameters (air temperature, pressure and humidity, wind speed and direction) for Sarajevo were used. Optimal model was selected based on the values of R2, MSE and the complexity of models. Optimal neural network model can predict air pollutants concentrations for the 72nd hour ahead with high accuracy, as well as for all hours up to 72nd hour.

Download Full-text

Wind speed forecasting using neural networks

Wind Engineering ◽

10.1177/0309524x19849846 ◽

2019 ◽

Vol 44 (1) ◽

pp. 33-48 ◽

Cited By ~ 6

Author(s):

Tyler Blanchard ◽

Biswanath Samanta

Keyword(s):

Neural Network ◽

Neural Networks ◽

Artificial Neural Networks ◽

Wind Speed ◽

The United States ◽

Weather Data ◽

Neural Network Models ◽

Artificial Neural ◽

Nonlinear Autoregressive ◽

Autoregressive Neural Network

The prediction of wind speed is critical in the assessment of feasibility of a potential wind turbine site. This work presents a study on prediction of wind speed using artificial neural networks. Two variations of artificial neural networks, namely, nonlinear autoregressive neural network and nonlinear autoregressive neural network with exogenous inputs, were used to predict wind speed utilizing 1 year of hourly weather data from four locations around the United States to train, validate, and test these networks. This study optimized both neural network configurations and it demonstrated that both models were suitable for wind speed prediction. Both models outperformed persistence model (with a factor of about 2 to 10 in root mean square error ratio). Both artificial neural network models were implemented for single-step and multi-step-ahead prediction of wind speed for all four locations and results were compared. Nonlinear autoregressive neural network with exogenous inputs model gave better prediction performance than nonlinear autoregressive model and the difference was statistically significant.

Download Full-text

Wavelet-Like Transform to Optimize the Order of an Autoregressive Neural Network Model to Predict the Dissolved Gas Concentration in Power Transformer Oil from Sensor Data

Sensors ◽

10.3390/s20092730 ◽

2020 ◽

Vol 20 (9) ◽

pp. 2730 ◽

Cited By ~ 1

Author(s):

Francisco Elânio Bezerra ◽

Fernando André Zemuner Garcia ◽

Silvio Ikuyo Nabeta ◽

Gilberto Francisco Martha de Souza ◽

Ivan Eduardo Chabu ◽

...

Keyword(s):

Neural Network ◽

Autoregressive Model ◽

Time Delays ◽

Transformer Oil ◽

Sensor Data ◽

Percentage Error ◽

Optimal Time ◽

Dissolved Gas ◽

Gas Concentration ◽

Autoregressive Neural Network

Dissolved gas analysis (DGA) is one of the most important methods to analyze fault in power transformers. In general, DGA is applied in monitoring systems based upon an autoregressive model; the current value of a time series is regressed on past values of the same series, as well as present and past values of some exogenous variables. The main difficulty is to decide the order of the autoregressive model; this means determining the number of past values to be used. This study proposes a wavelet-like transform to optimize the order of the variables in a nonlinear autoregressive neural network to predict the in oil dissolved gas concentration (DGC) from sensor data. Daubechies wavelets of different lengths are used to create representations with different time delays of ten DGC, which are then subjected to a procedure based on principal components analysis (PCA) and Pearson’s correlation to find out the order of an autoregressive model. The representations with optimal time delays for each DGC are applied as input in a multi-layer perceptron (MLP) network with backpropagation algorithm to predict the gas at the present and future times. This approach produces better results than choosing the same time delay for all inputs, as usual. The forecasts reached an average mean absolute percentage error (MAPE) of 5.763%, 1.525%, 1.831%, 2.869%, and 5.069% for C2H2, C2H6, C2H4, CH4, and H2, respectively.

Download Full-text