This contribution proposes a novel solar time series forecasting approach based on multimodel statistical ensembles to predict global horizontal irradiance (GHI) in short-term horizons (up to 1 hour ahead). The goal of the proposed methodology is to exploit the diversity of a set of dissimilar predictors with the purpose of increasing the accuracy of the forecasting process. The performance of a specific multimodel ensemble forecast showing an improved forecast skill is demonstrated and compared to a variety of individual single models. The proposed system can be applied in two distinct ways. The first one incorporates the forecasts acquired from the different forecasting models constituting the ensemble via a linear combination (combination-based). The other one consists of a novel methodology that delivers as output the forecast provided by the specific model (involved in the ensemble) that delivers the maximum precision in the zone of the variable space connected with the considered GHI time series (selection-based approach). This forecasting model is issued from an appropriate division of the variable space. The efficiency of the proposed methodology has been evaluated using high-quality measurements carried out at 1min intervals at four radiometric sites representing widely different radiative climates (Arid, Temperate, Tropical, and High Albedo). The obtained results emphasize that, at all sites, the proposed multi-model ensemble is able to increase the accuracy of the forecasting process using the different combination approaches, with a significant performance improvement when using the classification strategy.
Nowadays, the real life constraints necessitates controlling modern machines using human intervention by means of sensorial organs. The voice is one of the human senses that can control/monitor modern interfaces. In this context, Automatic Speech Recognition is principally used to convert natural voice into computer text as well as to perform an action based on the instructions given by the human. In this paper, we propose a general framework for Arabic speech recognition that uses Long Short-Term Memory (LSTM) and Neural Network (Multi- Layer Perceptron: MLP) classifier to cope with the nonuniform sequence length of the speech utterances issued fromboth feature extraction techniques, (1)Mel Frequency Cepstral Coefficients MFCC (static and dynamic features), (2) the Filter Banks (FB) coefficients. The neural architecture can recognize the isolated Arabic speech via classification technique. The proposed system involves, first, extracting pertinent features from the natural speech signal using MFCC (static and dynamic features) and FB. Next, the extracted features are padded in order to deal with the non-uniformity of the sequences length. Then, a deep architecture represented by a recurrent LSTM or GRU (Gated Recurrent Unit) architectures are used to encode the sequences ofMFCC/FB features as a fixed size vector that will be introduced to a Multi-Layer Perceptron network (MLP) to perform the classification (recognition). The proposed system is assessed using two different databases, the first one concerns the spoken digit recognition where a comparison with other related works in the literature is performed, whereas the second one contains the spoken TV commands. The obtained results show the superiority of the proposed approach.
Accurate solar irradiance forecasts are now key to successfully integrate the (variable) production from large solar energy systems into the electricity grid. This paper describes a wrapper forecasting methodology for irradiance time series that combines mutual information and an Extreme Learning Machine (ELM), with application to short forecast horizons between 5-min and 3-hour ahead. The method is referred to as Wrapper Mutual Information Methodology (WMIM). To evaluate the proposed approach, its performance is compared to that of three dimensionality reduction scenarios: full space (latest 50 variables), partial space (latest 5 variables), and the usual Principal Component Analysis (PCA). Based on measured irradiance data from two arid sites (Madina and Tamanrasset), the present results reveal that the reduction of the historical input space increases the forecasting performance of global solar radiation. In the case of Madina and forecast horizons from 5-min to 30-min ahead, the WMIM forecasts have a better coefficient of determination (R2 between 0.927 and 0.967) than those using the next best performing strategy, PCA (R2 between 0.921 and 0.959). The Mean Absolute Percentage Error (MAP) is also better for WMIM [7.4–10.77] than for PCA [8.4–11.55]. In the case of Tamanrasset and forecasting horizons from 1-hour to 3-hours ahead, the WMIM forecasts have an R2 between 0.883 and 0.957, slightly better than the next best performing strategy (PCA) (R2 between 0.873 and 0.910). The Normalized Mean Squared Error (NMSE) is similarly better for WMIM [0.048–0.128] than for PCA [0.105–0.130]. It is also found that the ELM technique is considerably more computationally efficient than the more conventional Multi Layer Perceptron (MLP). It is concluded that the proposed mutual information-based variable selection method has the potential to outperform various other proposed techniques in terms of prediction performance.
The use of wind energy is progressively utilized to produce electrical energy. Wind energy is related to the variation of some atmospheric variables such as wind direction, wind speed, air density and atmospheric pressure. Recently, numerous methods base on Artificial Intelligence techniques to forecast wind speed have been proposed in the literature. In this paper a new artificial intelligence approach for wind speed time series forecasting is proposed, it is composed from two blocs: The first one is based on the use of a deep architecture. The Autoencoder which is a type of deep neural networks, utilized generally for Denoising, is employed to reduce the wind speed input dimensionality. In the second bloc of the proposed methodology, the Elman neural network is employed to forecast future values of wind series, it is a kind of recurrent neural networks that are very sensitive to historical variations. To evaluate our approach we used the following error indicators: Root Mean Square Error (RMSE),Mean Absolute Bias Error (MABE), Mean Absolute Percentage Error (MAPE)and the coefficient of determination (R 2 ). The obtained results are compared with those of the Extreme Learning Machine method.
Automatic Speech Recognition can be considered as a transcription of spoken utterances into text which can be used to monitor/command a specific system. In this paper, we propose a general end-to-end approach to sequence learning that uses Long Short-Term Memory (LSTM) to deal with the non-uniform sequence length of the speech utterances. The neural architecture can recognize the Arabic spoken digit spelling of an isolated Arabic word using a classification methodology, with the aim to enable natural human-machine interaction. The proposed system consists to, first, extract the relevant features from the input speech signal using Mel Frequency Cepstral Coefficients (MFCC) and then these features are processed by a deep neural network able to deal with the non uniformity of the sequences length. A recurrent LSTM or GRU architecture is used to encode sequences of MFCC features as a fixed size vector that will feed a multilayer perceptron network to perform the classification. The whole neural network classifier is trained in an end-to-end manner. The proposed system outperforms by a large gap the previous published results on the same database.
In this paper, we propose a set of times series forecasting techniques based on the combination of Support Vector Regression methods to predict global horizontal solar radiation in Algeria. The models were constructed and tested using different architectures of Support Vector Machine (SVM), namely, (RBF kernel, Polinomial kernel and Linear kernel). We use individual time series models and linear combination techniques to predict global solar radiation indifferent sites in Algeria. For this aim, the recorded data of 4 stations spread over Algeria were used to build different combination schemes for the different times series algorithms. The efficiency of the different models was calculated using a number of statistical indicators: the Mean Absolute Percentage Error (MAPE), the Mean Squared Error (RMSE), Mean Bias Error (MABE) and the Coefficient of Determination (R2). The results obtained from these models were compared with the measured data.
Renewable Energy Sources (RES) are one of the key solutions to handle the world’s future energy needs, while decreasing carbon emissions. To produce electricity, large concentrating solar power plants depend on Direct Normal Irradiance (DNI), which is rapidly variable under broken clouds conditions. To work at optimum capacity while maintaining stable grid conditions, such plants require accurate DNI forecasts for various time horizons. The main goal of this study is the forecasting of DNI over two short-term horizons: 15-min and 1-h. The proposed system is purely based on historical local data and Artificial Neural Networks (ANN). For this aim, 1-min solar irradiance measurements have been obtained from two sites in different climates. According to the forecast results, the coefficient of determination (R2) ranges between 0.500 and 0.851, the Mean Absolute Percentage Error (MAPE) between 0.500 and 0.851, the Normalized Mean Squared Error (NMSE) between 0.500 and 0.851, and the Root Mean Square Error (RMSE) between 0.065 kWh/m2 and 0.105 kWh/m2. The proposed forecasting models show a reasonably good forecasting capability, which is decisive for a good management of solar energy systems.
In this paper, the problem of time series prediction is studied. A Bayesian procedure based on Gaussian Process (GP) model combined with Linear Discriminate Analysis (LDA) as dimensionality reduction method is proposed. To evaluate the proposed approach, its performance is assessed using three scenarios: long window (latest 50 variables), short window (latest 5 variables) and persistence. To evaluate the performance of the proposed forecast model, the results of the different scenarios are compared to that of Extreme Learning Machines (ELM). Based on measured irradiance data from Tamanrasset, Algeria, the present results describe the performance of the combination of LDA with GP for forecasting hourly global solar irradiance.
Solar energy is expected to provide a major contribution to the future global energy supply, while helping the transition toward a carbon-free economy. Because of its variable character, its efficient use will necessitate trustworthy forecast information of its availability in a variety of spatial and time scales, depending on application. This study proposes a new forecasting approach for irradiance time series that combines mutual information measures and an Extreme Learning Machine (ELM). The method is referred to as Minimum Redundancy – Maximum Relevance (MRMR). To assess the proposed approach, its performance is evaluated against four scenarios: long window (latest 50 variables), short window (latest 5 variables), standard Principal Components Analysis (PCA) and clear-sky model. All these scenarios are applied to three typical forecasting horizons (15-min ahead, 1-h ahead and 24-h ahead). Based on measured irradiance data from 20 sites representing a variety of climates, the test results reveal that the selection of a good set of relevant variables positively impacts the forecasting performance of global solar radiation. The present findings show that the proposed approach is able to improve the accuracy of solar irradiance forecasting compared to other proposed scenarios.
Nowadays, wind power and precise forecasting are of great importance for the development of modern electrical grids. In this paper we propose a prediction system for time series based on Kernel Principal Component Analysis (KPCA) and Extreme Learning Machine (ELM). To compare the proposed approach, three dimensionality reduction techniques were used: full space (50 variables), part of space (last four variables) and classical Principal Components Analysis (PCA). These models were compared using three evaluation criteria: mean absolute error (MAE), root mean square error (RMSE), and normalized mean square error (NMSE). The results show that the reduction of the original input space affects positively the prediction output of the wind speed. Thus, It can be concluded that the non linear model (KPCA) model outperform the other reduction techniques in terms of prediction performance.
This paper presents a set of artificial neural network models (ANN) to estimate daily global solar radiation (GSR) on a horizontal surface using meteorological variables: (mean daily extraterrestrial solar radiation intensity G0, the maximum possible sunshine hours S0, mean daily relative humidity H, mean daily maximum air temperature T, mean daily atmospheric pressure P and wind speed Vx) for Djelfa city in Algeria. In order to consider the effect of the different meteorological parameters on daily global solar radiation prediction, four following combinations of input features are considered: 1) Day of the year, G0, S0, T and Vx. 2) Day of the year, G0, S0, T, P and Vx. 3) Day of the year, G0, S0, T, H, P and Vx. 4) Day of the year, G0, S0, T, H and Vx. These models were compared using three evaluation criteria: Mean square error (MSE), mean absolute error (MAE), and root mean square error (RMSE). The results show that the two parameters: atmospheric pressure and relative humidity affect the prediction output of global solar radiation. In addition, the results show that the relative humidity is the most important features influencing the prediction performance. It can be concluded that fourth model can be used for forecasting daily global solar radiation in other locations in Algeria.
We present three models for the estimation of hourly global solar radiation for two sites in Algeria, namely: Djelfa (Latitude 34.68°N, Longitude 3.25°E, Altitude 1126 (m)) and Ain Bessem (Latitude 36.31°N, Longitude 3.67°E, Altitude 629 (m)). The models are: the Gaussian distribution model, the model by Collares-Pereira-RabI and the H.A. model (Hourly absolute modelling approach). The experimental assessment was done using recorded values of the hourly global solar radiation on a horizontal plane during the period 2000-2004. The obtained results show a close similarity between the solar radiation values calculated by the three models and the measured values, especially for the first model. The experimental validation shows promising results for the estimation and precise prediction of the hourly global solar radiation.
Precise forecasting of renewable energies such as solar and wind is becoming one of the very important concerns in developing modern electrical grids. Hence, establishing appropriate tools of weather forecasting with satisfactory accuracy becomes an essential preoccupation in today's changing power world. In this paper, an approach based on Principal Component Analysis (PCA) and Extreme Learning Machines (ELM) is proposed for the forecasting of time series. The PCA maps the data into a smaller subspace in which the components accounts for as much of the variability in the input space as possible. The variables extracted by the PCA are then introduced to the extreme learning machines, a learning algorithm much faster than the traditional gradient-based learning algorithms. The experiments carried out on three time series lead to: (i) The PCA as variable selection method shows a positive impact on the accuracy of the forecasting process. (ii) ELM model is significantly faster than Multi-Layer Perceptron Network, Radial Basis Function Networks, and Least Squares Support Vector Machines, while preserving the same accuracy level.
The automatic analysis of data acquired with hyperdimensional sensors is rather challenging since it should be carried out in hyperdimensional feature spaces. The huge size of the feature space involves the so-called curse of dimensionality. This latter is due to the unbalancing between the number of features and the number of samples. In this book, it is proposed to exploit the whole information available in the original hyperdimensional feature space by means of the fusion of multiple regression methods. The development of the proposed multiple regression systems will include three main steps. The first one is related to the partition of the original hyperdimensional feature space into subspaces of reduced dimensionality. The second step consists in training in each of the subspaces obtained in the previous step a regression method. Finally, in the third and final step, the results provided by the different regression methods will be combined in order to produce a global estimate of the physical parameter of interest with an expected higher accuracy with respect to what can be achieved by the classical regression approach based on feature selection.
he wind has constantly been a natural partner in propelling our societies forward. The wind is often considered as one of the most complex meteorological parameters to predict. This is a consequence of the compound interactions between large scale of natural forcing phenomena such as pressure, temperature differences, earth rotation, and local characteristics of the earth surface. The predicting technique employed depends essentially on the available information and the time scale in question (horizon), and thus its application. In this book it is proposed to deal with the prediction of wind speed by two different and independent methodologies: In the first one, the proposed static system seeks to get the best prediction performance from a set of predicting algorithms, this is done by using a new approach, where the outputs yielded by the different single prediction architectures are combined by three fusion methods. In the second one, the wind speed prediction problem is formulated in the framework of time series. Several variable selection techniques were investigated to find the optimal number of historical wind speed values in order to get the best prediction performance.
A new approach based on multiple architecture system (MAS) for the prediction of wind speed is proposed. The motivation behind the proposed approach is to combine the complementary predictive powers of multiple models in order to improve the performance of the prediction process. The proposed MAS can be implemented by associating the predictions obtained from the different regression algorithms (MLR, MLP, RBF and SVM) making up the ensemble by three fusion strategies (simple, weighted and non-linear). The efficiency of the proposed approach has been assessed on a real data set recorded from seven locations in Algeria during a period of 10 years. The experimental results point out that the proposed MAS approach is capable of improving the precision of the wind speed prediction compared to the traditional prediction methods.