Artificial Neural Network Forecast Research Articles

Pre-Processing of Energy Demand Disaggregation Based Data Mining Techniques for Household Load Demand Forecasting

Demand side management has a vital role in supporting the demand response in smart grid infrastructure, in the decision-making of energy management, in household applications is significantly affected by the load-forecasting accuracy. This paper introduces an innovative methodology to enhance household demand forecasting based on energy disaggregation for Short Term Load Forecasting. This approach is constructed from Feed-Forward Artificial Neural Network forecaster and a pre-processing stage of energy disaggregation. This disaggregation technique extracts the individual appliances’ load demand profile from the aggregated household load demand to increase the training data window for the proposed forecaster. These proposed algorithms include two benchmark disaggregation algorithms; Factorial Hidden Markov Model (FHMM), Combinatorial Optimization in addition to three adopted Deep Neural Network; long short- term memory (LSTM), Denoising Autoencoder, and a network which regress start time, end time, and average power. The proposed load forecasting approach outperformed the currently available state-of-the-art techniques; namely root mean square error (RMSE), normalized root mean square error (NRMSE), and mean absolute error (MAE).

Optimal scheduling of a microgrid in a volatile electricity market environment: Portfolio optimization approach

This paper proposes an optimal scheduling strategy for a microgrid participating in a volatile electricity market. The microgrid system includes photovoltaic generators, a wind turbine, a load, grid connection, and a battery storage system. An optimal microgrid operation is achieved by maximizing the utility function represented by the exponential rate of growth of the electricity market value through electricity transactions between the microgrid and main grid, on the premise of satisfying the power balance and generation limit of system components. The uncertainties occurring during the microgrid operation are represented by generator output, load demand, and electricity price fluctuation. The proposed strategy utilizes the Kelly Criterion, an optimal strategy that maximizes the growth rate of an asset’s net worth over repeated investments, coupled with an artificial neural network forecast of electricity price to deal with the volatile energy market. The proposed algorithm provides significant improvements in microgrid scheduling by eliminating the reliance on renewable generation and load forecasts, which makes it computationally inexpensive and thus feasible for real-time implementation. In representative case scenarios, using real-world tracers, we show that the algorithm has no dependency on meteorological forecasts and performs optimally in a volatile electricity market.

A Hybrid Model of Artificial Neural Network and Genetic Algorithm in Forecasting Gold Price

The goal of this study is to compare the forecasting performance of classical artificial neural network and the hybrid model of artificial neural network and genetic algorithm. The time series data used is the monthly gold price per troy ounce in USD from year 1987 to 2016. A conventional artificial neural network trained by back propagation algorithm and the hybrid forecasting model of artificial neural network and genetic algorithms are proposed. Genetic algorithm is used to optimize the of artificial neural network neurons. Three forecasting accuracy measures which are mean absolute error, root mean squared error and mean absolute percentage error are used to compare the accuracy of artificial neural network forecasting and hybrid of artificial neural network and genetic algorithm forecasting model. Fitness of the model is compared by using coefficient of determination. The hybrid model of artificial neural network is suggested to be used as it is outperformed the classical artificial neural network in the sense of forecasting accuracy because its coefficient of determination is higher than conventional artificial neural network by 1.14%. The hybrid model of artificial neural network and genetic algorithms has better forecasting accuracy as the mean absolute error, root mean squared error and mean absolute percentage error is lower than the artificial neural network forecasting model.

A Hybrid Model of Artificial Neural Network and Genetic Algorithm in Forecasting Gold Price

The goal of this study is to compare the forecasting performance of classical artificial neural network and the hybrid model of artificial neural network and genetic algorithm. The time series data used is the monthly gold price per troy ounce in USD from year 1987 to 2016. A conventional artificial neural network trained by back propagation algorithm and the hybrid forecasting model of artificial neural network and genetic algorithms are proposed. Genetic algorithm is used to optimize the of artificial neural network neurons. Three forecasting accuracy measures which are mean absolute error, root mean squared error and mean absolute percentage error are used to compare the accuracy of artificial neural network forecasting and hybrid of artificial neural network and genetic algorithm forecasting model. Fitness of the model is compared by using coefficient of determination. The hybrid model of artificial neural network is suggested to be used as it is outperformed the classical artificial neural network in the sense of forecasting accuracy because its coefficient of determination is higher than conventional artificial neural network by 1.14%. The hybrid model of artificial neural network and genetic algorithms has better forecasting accuracy as the mean absolute error, root mean squared error and mean absolute percentage error is lower than the artificial neural network forecasting model.

Short-Term Forecasting of Coastal Surface Currents Using High Frequency Radar Data and Artificial Neural Networks

Accurate and timely information of surface currents is crucial for various operations such as search and rescue, marine renewable energy extraction and oil spill treatment. Conventional approaches to study coastal surface currents are numerical models and observation platforms such as radars and satellites. However, both have limits. To efficiently obtain high accuracy short-term forecasting states of oceanic parameters of interest, a robust soft computing approach—Artificial Neural Networks (ANN)—was applied to predict surface currents in a tide- and wind-dominated coastal area. Hourly observed surface currents from a Coastal Ocean Dynamic Application Radar (CODAR) system, and tide and wind data from forecasting models were used to establish ANN models for Galway Bay area. One of the fastest algorithms, resilient back propagation, was used to adapt all weights and biases. This study focused on investigating the sensitivity of an ANN model to a series of different input datasets. Results indicate that correlation between ANN forecasts and observation was greater than 0.9 for both surface velocity components with one-hour lead time. Strong correlation ( ≥ 0.75) was obtained between predicted results and radar data for both surface velocity components with three-hour lead time at best. However, forecasting accuracy deteriorated rapidly with longer lead time. By comparison with previous data assimilation models, in this research, best performance was achieved from ANN model’s peak times of the tidally dominant surface velocity component. The forecasts presented in this research show clear improvements over previous attempts at short-term forecasting of wind- and tide-dominated currents using ANN.

Artificial Neural Network (Ann) Forecast of University Growth: A Focus on College of Technology Education, Kumasi, University of Education, Winneba Admissions

Artificial Neural Network (Ann) Forecast of University Growth: A Focus on College of Technology Education, Kumasi, University of Education, Winneba Admissions

Solar Forecasting using ANN with Fuzzy Logic Pre-processing

Lack of predictability of solar power remains one major hindrance to the introduction of large-scale solar energy production. Comprehensive solar forecasting technologies are required to manage the intermittent nature of solar energy supply. However, one of the most challenging aspects of solar forecasting is the requirement for very short-term forecasting (in terms of minutes ahead) due to cloud movement, ambient temperature variation and humidity levels, which result in rapid ramp up and ramp down rates. This paper proposes an improved solar forecasting algorithm based on artificial neural network (ANN) model with fuzzy logic pre-processing. A three-layer (input layer, hidden layer and output layer) feed forward with back-propagation model is proposed as the neural network training algorithm. In this paper, an error correction factor based on the previous 5-min forecasted output is included to the input layer to minimize the forecast error. This is expected to produce a more accurate forecast. Weather information is a key input for the ANN model. In the case of rapid changes in solar irradiation or temperature on the forecast day, produced solar power changes greatly and forecast error would sensibly increase. In traditional prediction methods, the ANN uses all similar days’ data to learn the trend of similarity. However, learning all similar days’ data is quite complex, and it does not help if weather conditions change suddenly during the same day. Therefore, it is necessary to integrate into the ANN, a pre-processing stage that could perform real time analysis of weather information and factor it into the forecasting output. A fuzzy pre-processing toolbox is introduced into the proposed ANN model to find data correlation between cloud cover, temperature, wind speed, and wind direction with irradiance value. The accuracy of the proposed forecasting algorithm is compared with other ANN forecast algorithms.

Feature Selection and ANN Solar Power Prediction

A novel method of solar power forecasting for individuals and small businesses is developed in this paper based on machine learning, image processing, and acoustic classification techniques. Increases in the production of solar power at the consumer level require automated forecasting systems to minimize loss, cost, and environmental impact for homes and businesses that produce and consume power (prosumers). These new participants in the energy market, prosumers, require new artificial neural network (ANN) performance tuning techniques to create accurate ANN forecasts. Input masking, an ANN tuning technique developed for acoustic signal classification and image edge detection, is applied to prosumer solar data to improve prosumer forecast accuracy over traditional macrogrid ANN performance tuning techniques. ANN inputs tailor time-of-day masking based on error clustering in the time domain. Results show an improvement in prediction to target correlation, the R2 value, lowering inaccuracy of sample predictions by 14.4%, with corresponding drops in mean average error of 5.37% and root mean squared error of 6.83%.

Impact of climate variability on the salinization of the coastal wetland-aquifer system of the Po Delta, Italy

Deltaic coastal areas are constituted by a patchwork of brackish lagoons and freshwater bodies; these coastal wetland-aquifer systems are fragile ecosystems that usually respond quickly to climate changes. To understand the hydrological processes occurring within the lagoons and the groundwater system of the Po River Delta (Italy), the contribution of both evaporation and anthropogenic factors on groundwater salinization was assessed. A time series (2002–2015) of monthly average climatic data and a temperature-salinity dataset were used in three adjacent saline-brackish lagoons with the aim to identify the actual evaporation patterns and predict future trends using artificial neural networks (ANN). Moreover, the use of groundwater and surface water equivalent freshwater heads, along with the geological architecture, allowed linking the fluctuation of lagoon salinities with the degree of hydraulic connection between wetland and aquifer system. Results show that the less a lagoon is hydraulically connected with the aquifer, the higher is the salinity peak that could be reached at the end of the summer period. ANN forecasts highlight that in the near future this behaviour would be the rule rather than the exception. The increase in salinity of surface waters could be of serious concern, especially for aquaculture, sensitive to sharp salinity increases.

A multi-stage methodology for selecting input variables in ANN forecasting of river flows

<p>The scientific community has recognized the necessity for more efficiently selected inputs in artificial neural network models (ANNs) in river flows and has worked on this despite some shortcomings. Moreover, there is none or limited inclusion of ANN inputs coupled with atmospheric circulation under various patterns arising from the need of data downscaling for climate change predictions in hydrology domain. This paper presents the results of a novel multi-stage methodology for selecting input variables used in artificial neural network (ANN) models for river flow forecasting. The proposed methodology makes use of data correlations together with a set of crucial statistical indices for optimizing model performance, both in terms of ANN structure (e.g. neurons, momentum rate, learning rate, activation functions, etc), but also in terms of inputs selection. The latter include various previous time steps of daily areal precipitation and temperature data coupled with atmospheric circulation in the form of circulation patterns, observed river flow data and time expressed via functions of sine and cosine. Additionally, the no-linear behavior between river flow and the respective inputs is investigated by the ANN configuration itself and not only by correlation indices (or other equivalent contingency tools). The proposed methodology revealed the river flow of past four days, the precipitation of past three days and the seasonality as robust input variables. However, the temperature of three past days should be considered as an alternative against the seasonality. The produced models forecasting ability was validated by comparing its one-step ahead flow prediction ability to two other approaches (an auto regressive model and a genetic algorithm (GA)-optimized single input ANN). </p>

Does Artificial Neural Network Forecast Better for Excessively Volatile Currency Pairs?

This study predicts the exchange rates for three currency pairs (USD-INR, GBP-INR, and EUR-INR). We have used multi-layer perceptron (MLP) neural network architecture based on feed-forward with back-propagation learning method. The sample of the study covers daily data for the period from January 2009 to January 2016. The findings of the study confirm that the neural network predicts better for more volatile currency pairs (GBP-INR and EUR-INR) as compared to a less volatile currency pair (USD-INR). The study further observes that the optimal forecast horizon for the neural network model should be equal to the optimal lag length used in the construction of the model. This study aims to contribute in the area of foreign exchange forecasting. Exchange rate plays a crucial role in the macro-economy of a country. Hence, prediction of currency exchange rate becomes imperative for various stakeholders such as government, the central bank, and investors to maximize the returns and minimize the risk in their decision-making.

A multiperiodal management method at user level for storage systems using artificial neural network forecasts

The increase of renewable non-programmable production and the necessity to locally self-consume the produced energy led to utilize ever more storage systems. To correctly utilize storage systems, an opportune management method has to be utilized. This paper implements a multi-period management method for storage systems, using different management strategies. The method aims to minimize the total absorbed and supplied energy or the peak power exchanged with the grid. The results show the effectiveness of the method in diminishing the energy exchanged with the grid and also the possibility to optimize the performance of the storage systems

TIME SERIES FORECASTING BY USING A NEURAL ARIMA MODEL BASED ON WAVELET DECOMPOSITION

In the prediction of (stochastic) time series, it has been common to suppose that an individual predictive method – for instance, an Auto-Regressive Integrated Moving Average (ARIMA) model – produces residuals like a white noise process. However, mainly due to the structures of auto-dependence not mapped by a given individual predictive method, this assumption may easily be violated, in practice, as pointed out in Firmino et al. (2015). In order to correct it (and accordingly to produce more forecasts with more accuracy power), this paper puts forward a Wavelet Hybrid Forecaster (WHF) that integrates the following numerical techniques: wavelet decomposition; ARIMA models; Artificial Neural Networks (ANNs); and linear combination of forecasts. Basically, the proposed WHF can map simultaneously linear – by means of a linear combination of ARIMA forecasts – and non-linear – through a linear combination of ANN forecasts – auto-dependence structures exhibited by a given time series. Differently of other hybrid methodologies existing in literature, the WHF forecasts are produced carrying into account implicitly the information from the frequency presenting in the underlying time series by means of the Wavelet Components (WCs) obtained by the wavelet decomposition approach. All numerical results show that WHF method has achieved remarkable accuracy gains, when comparing with other competitive forecasting methods already published in specialized literature, in the prediction of a well-known annual time series of sunspot.

Seasonal rainfall forecasting by adaptive network-based fuzzy inference system (ANFIS) using large scale climate signals

Accurate seasonal rainfall forecasting is an important step in the development of reliable runoff forecast models. The large scale climate modes affecting rainfall in Australia have recently been proven useful in rainfall prediction problems. In this study, adaptive network-based fuzzy inference systems (ANFIS) models are developed for the first time for southeast Australia in order to forecast spring rainfall. The models are applied in east, center and west Victoria as case studies. Large scale climate signals comprising El Nino Southern Oscillation (ENSO), Indian Ocean Dipole (IOD) and Inter-decadal Pacific Ocean (IPO) are selected as rainfall predictors. Eight models are developed based on single climate modes (ENSO, IOD, and IPO) and combined climate modes (ENSO-IPO and ENSO-IOD). Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Pearson correlation coefficient (r) and root mean square error in probability (RMSEP) skill score are used to evaluate the performance of the proposed models. The predictions demonstrate that ANFIS models based on individual IOD index perform superior in terms of RMSE, MAE and r to the models based on individual ENSO indices. It is further discovered that IPO is not an effective predictor for the region and the combined ENSO-IOD and ENSO-IPO predictors did not improve the predictions. In order to evaluate the effectiveness of the proposed models a comparison is conducted between ANFIS models and the conventional Artificial Neural Network (ANN), the Predictive Ocean Atmosphere Model for Australia (POAMA) and climatology forecasts. POAMA is the official dynamic model used by the Australian Bureau of Meteorology. The ANFIS predictions certify a superior performance for most of the region compared to ANN and climatology forecasts. POAMA performs better in regards to RMSE and MAE in east and part of central Victoria, however, compared to ANFIS it shows weaker results in west Victoria in terms of prediction errors and RMSEP skill score. In general, ANFIS models show superior results in terms of correlation coefficient for the overall case study. As a pioneer study, it is proposed that ANFIS is a promising tool for the purpose of seasonal predictions in Australia as they produce comparable accuracy using minimal inputs, require less development time and they are less complex compared to dynamic models.

Medium Range Daily Reference Evapotranspiration Forecasting by Using ANN and Public Weather Forecasts

Medium range daily reference evapotranspiration (ETo) forecasts are very helpful for farmers or irrigation system operators for improving their irrigation scheduling. We tested four artificial neural networks (ANNs) for ETo forecasting using forecasted temperatures data retrieved from public weather forecasts. Daily meteorological data were collected to train and validate the ANNs against the Penman–Monteith (PM) model. And then, the temperature forecasts for 7-day ahead were entered into the validated ANNs to produce ETo forecast outputs. The forecasting performances of models were evaluated through comparisons between the ETo forecasted by ANNs and ETo calculated by PM from the observed meteorological data. The correlation coefficients between observed and forecasted temperatures for all stations were all greater than 0.91, and the accuracy of the minimum temperature forecast (error within ± 2 °C) ranged from 68.34 to 91.61 %, whereas for the maximum temperature it ranged from 51.78 to 57.44 %. The accuracy of the ETo forecast (error within ± 1.5 mm day−1) ranged from 75.53 to 78.14 %, the average values of the mean absolute error ranged from 0.99 to 1.09 mm day−1, the average values of the root mean square error ranged from 0.87 to 1.36 mm day−1, and the average values of the correlation coefficient ranged from 0.70 to 0.75. The results suggested that ANNs can be considered as a promising ETo forecasting tool. The forecasting performance can be improved by promoting temperature forecast accuracy.

A river flash flood forecasting model coupled with ensemble Kalman filter

AbstractA flash flood forecasting model including a state‐of‐the‐art data assimilation method was developed to provide a precise water stage forecast for flood emergency response. The model integrates a flash flood routing model (FFRM) coupled with an ensemble Kalman filter (EnKF) and an artificial neural network (ANN) submodel. In the model, the ANN forecasts river water stages at gauge stations first. Then, these are used as the initial and boundary conditions of the FFRM. The water stages, simulated from the FFRM, are then corrected by the EnKF for lead time. The model was applied to the Tanshui River watershed in northern Taiwan during past typhoons. The model forecasts almost covered the data observed during a typhoon period to within 95% confidence intervals. Compared with the use of FFRM without EnKF, the forecast water stages from the EnKF improved the accuracy at the conjunctions between upstream and downstream channels and the steep slope location.

Radial basis function neural networks for reliably forecasting rainfall

Rainfall forecasting is an interesting task especially in a modern city facing the problem of global warming; in addition rainfall is a necessary input for the analysis and design of hydrologic systems. Most rainfall real-time forecasting models are based on conceptual models simulating the complex hydrological process under climate variability. As there are a lot of variables and parameters with uncertainties and non-linear relationships, the calibration of conceptual or physically based models is often a difficult and time-consuming procedure. Simpler artificial neural network (ANN) forecasts may therefore seem attractive as an alternative model. The present research demonstrates the application of the radial basis function neural network (RBFNN) to rainfall forecasting for Alexandria City, Egypt. A significant feature of the input construction of the RBF network is based on the use of the average 10 year rainfall in each decade to forecast the next year. The results show the capability of the RBF network in forecasting the yearly rainfall and two highest rainfall monsoon months, January and December, compared with other statistical models. Based on these results, the use of the RBF model can be recommended as a viable alternative for forecasting the rainfall based on historical rainfall recorded data.

A feasibility of six-hourly rainfall forecast over central India using model output and remote sensing data

Rainfall forecast has prime importance in an agrarian country like India, wherein the agricultural production is solely dependent on monsoon rainfall. In this paper, an artificial neural network (ANN) technique is used to construct a non-linear mapping between output data from global forecast system (GFS) and rainfall from tropical rainfall measuring mission (TRMM) satellite measurements. The objective of the present study is to generate region-specific six-hourly quantitative rainfall forecast over central India using ANN and resilient propagation learning algorithm. Meteorological variables from the GFS model and precipitation product from TRMM multisatellite precipitation analysis (TMPA) are used as input data for training the network, which generate rainfall forecast for the next time step. The test was performed for central India during the summer monsoon period of 2010. In order to evaluate the potential of rainfall forecast skill over the studied region, the forecast precipitation has been intercompared with TMPA-3B42, and Kalpana-1 derived precipitation products and a statistical analysis was performed. The linear correlation between ANN forecast and TMPA-3B42 rainfall was 0.58, whereas it was 0.52 with Kalpana-1 derived precipitation estimates. The results show that the predicted precipitation by the present technique performs better than GFS model precipitation forecast, and the system indicates a potential for more accurate rainfall forecasting.

Assessment of Arias Intensity of historical earthquakes using modified Mercalli intensities and artificial neural networks

Abstract. This paper presents the development of a non-parametric forecast model based on artificial neural networks for the direct assessment of Arias Intensity corresponding to a historic earthquake using seismic intensity data. The neural models allow complex and nonlinear behaviour to be tracked. Application of this methodology on earthquakes with known instrumental data from Greece, showed that the artificial neural network forecast model have excellent data synthesis capability.

Improved one day-ahead price forecasting using combined time series and artificial neural network models for the electricity market

The price forecasts embody crucial information for generators when planning bidding strategies to maximise profits. Therefore, generation companies need accurate price forecasting tools. Comparison of neural network and auto regressive integrated moving average (ARIMA) models to forecast commodity prices in previous researches showed that the artificial neural network (ANN) forecasts were considerably more accurate than traditional ARIMA models. This paper provides an accurate and efficient tool for short-term price forecasting based on the combination of ANN and ARIMA. Firstly, input variables for ANN are determined by time series analysis. This model relates the current prices to the values of past prices. Secondly, ANN is used for one day-ahead price forecasting. A three-layered feed-forward neural network algorithm is used for forecasting next-day electricity prices. The ANN model is then trained and tested using data from electricity market of Iran. According to previous studies, in the case of neural networks and ARIMA models, historical demand data do not significantly improve predictions. The results show that the combined ANN?ARIMA forecasts prices with high accuracy for short-term periods. Also, it is shown that policy-making strategies would be enhanced due to increased precision and reliability.