Transformers are one of the most important part in a power system and, especially in key-facilities, they should be closely and continuously monitored. In this context, methods based on the dissolved gas ratios allow to associate values of gas concentrations with the occurrence of some faults, such as partial discharges and thermal faults. So, an accurate prediction of oil-dissolved gas concentrations is a valuable tool to monitor the transformer condition and to develop a fault diagnosis system. This study proposes a nonlinear autoregressive neural network model coupled with the discrete wavelet transform for predicting transformer oil-dissolved gas concentrations. The data fitting and accurate prediction ability of the proposed model is evaluated in a real world example, showing better results in relation to current prediction models and common time series techniques.
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.
In Brazil, the electric power distributors must buy electricity on auctions one, three and five years ahead. If there is inefficiency in the contracting of electric energy, the chamber of Commercialization of Electric Energy, which enables the commercialization, can apply penalties. Thus, this paper proposes a computational approach to forecasting electricity by the class of the consumer using a multi-layer perceptron artificial neural network with a backpropagation algorithm and a prediction using time series techniques through the Bayesian and Akaike selection criteria. The forecast of electricity consumption can serve as support in the purchase of electricity in auctions in the regulated contracting environment and in the process of settlement of differences and for energy management, customer service, and distributor billing. The results show that a multilayer network with a backpropagation algorithm is able to learn the behavior of the data that influences the electric energy consumed by consumption class and can be used to follow the evolution in the demand of each class of consumption and, consequently, to help distributors in the process of contracting of electricity, reduce losses like fines, and reduce the costs of the energy distributor.
This work is aimed at demonstrating the advantages that AI can bring to dam management and which parameters and calculations are important to make the simulations more realistic. To this end, a computational approach that combines a Multilayer Perceptron Artificial Neural Network (MLP-ANN) and Monte Carlo Simulation (MCS) method was developed and tested in simulations of floodgate operation using data collected from one of the biggest sanitation companies in the world. The conducted systematic review and simulations allowed to demonstrate the contributions of this study to the scientific literature and organizational practice, mainly because it shows that the application of the proposed approach can eliminates the need for manual operations in dams, including those aimed at preventing disasters and water wastage.
The prediction of partial discharges in hydrogenerators depends on data collected by sensors and prediction models based on artificial intelligence. However, forecasting models are trained with a set of historical data that is not automatically updated due to the high cost to collect sensors’ data and insufficient real-time data analysis. This article proposes a method to update the forecasting model, aiming to improve its accuracy. The method is based on a distributed data platform with the lambda architecture, which combines real-time and batch processing techniques. The results show that the proposed system enables real-time updates to be made to the forecasting model, allowing partial discharge forecasts to be improved with each update with increasing accuracy.
Purpose
This paper aims to propose an approach based upon the principal component analysis (PCA) to define a contribution rate for each variable and then select the main variables as inputs to a neural network for energy load forecasting in the region southeastern Brazil.
Design/methodology/approach
The proposed approach defines a contribution rate of each variable as a weighted sum of the inner product between the variable and each principal component. So, the contribution rate is used for selecting the most important features of 27 variables and 6,815 electricity data for a multilayer perceptron network backpropagation prediction model. Several tests, starting from the most significant variable as input, and adding the next most significant variable and so on, are accomplished to predict energy load (GWh). The Kaiser–Meyer–Olkin and Bartlett sphericity tests were used to verify the overall consistency of the data for factor analysis.
Findings
Although energy load forecasting is an area for which databases with tens or hundreds of variables are available, the approach could select only six variables that contribute more than 85% for the model. While the contribution rates of the variables of the plants, plus energy exchange added, have only 14.14% of contribution, the variable the stored energy has a contribution rate of 26.31% being fundamental for the prediction accuracy.
Originality/value
Besides improving the forecasting accuracy and providing a faster predictor, the proposed PCA-based approach for calculating the contribution rate of input variables providing a better understanding of the underlying process that generated the data, which is fundamental to the Brazilian reality due to the accentuated climatic and economic variations.
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.