Alexander Efitorov

Comparison of the quality of solving the inverse problems of spectroscopy of multi-component solutions with neural network methods and with the method of projection to latent structures

This study provides comparative analysis of application of artificial neural networks and method of projection to latent structures (partial least squares) for simultaneous determination of types and concentrations of dissolved inorganic salts in multicomponent water solutions by Raman spectra. It is shown that the method of projection to latent structures has several advantages, such as the quality of the solution and the time of construction of a regression model, when solving problems with low level of nonlinearity.

Neural Network Approaches to Solution of the Inverse Problem of Identification and Determination of Partial Concentrations of Salts in Multi-сomponent Water Solutions

The studied inverse problem is determination of partial concentrations of inorganic salts in multi-component water solutions by their Raman spectra. The problem is naturally divided into two parts: 1) determination of the component composition of the solution, i.e. which salts are present and which not; 2) determination of the partial concentration of each of the salts present in the solution. Within the first approach, both parts of the problem are solved simultaneously, with a single neural network (perceptron) with several outputs, each of them estimating the concentration of the corresponding salt. The second approach uses data clusterization by Kohonen networks for consequent identification of component composition of the solution by the cluster, which the spectrum of this solution falls into. Both approaches and their results are discussed in this paper.

Neural Network Approaches to Solution of the Inverse Problem of Identification and Determination of the Ionic Composition of Multi-component Water Solutions

The studied inverse problem is determination of ionic composition of inorganic salts (concentrations of up to 10 ions) in multi-component water solutions by their Raman spectra. The regression problem was solved in two ways: 1) by a multilayer perceptron trained on the large dataset, composed of spectra of all possible mixing options of ions in water; 2) dividing the data set into compact clusters and creating regression models for each cluster separately. Within the first approach, we used supervised training of neural network, achieving good results. Unfortunately, this method isn’t stable enough; the results depend on data subdivision into training, test, and out-of-sample sets. In the second approach, we used algorithms of unsupervised learning for data clustering: Kohonen networks, k-means, k-medoids and hierarchical clustering, and built partial least squares regression models on the small datasets of each cluster. Both approaches and their results are discussed in this paper.

Solution of an Inverse Problem in Raman Spectroscopy of Multi-component Solutions of Inorganic Salts by Artificial Neural Networks

The paper presents a study of aspects of using single and multiple output artificial neural networks to determine concentrations of inorganic salts in multicomponent water solutions by processing their Raman spectra. The dependence of the results on complexity of the inverse problem has been demonstrated. The results are compared for two data arrays including spectra of solutions of: (1) 5 salts composed of 10 different ions, and (2) 10 salts composed of 10 different ions.

Prediction of Relativistic Electrons Flux in the Outer Radiation Belt of the Earth Using Adaptive Methods

Prediction of the time series of relativistic electrons flux in the outer radiation belt of the Earth encounters problems caused by complexity and nonlinearity of the “solar wind—the Earth’s magnetosphere” system. This study considers such prediction by the parameters of solar wind and interplanetary magnetic field and by geomagnetic indexes, using different methods, namely, Artificial Neural Network, Group Method of Data Handling and Projection to Latent Structures (also known as Partial Least Squares). Comparison of quality indexes of predictions with horizon from one to twelve hours among each other and with that of trivial model is presented.

Use of Adaptive Methods to Solve the Inverse Problem of Determination of Composition of Multi-Component Solutions

This study considers solving the inverse problem of determination of salt or ionic composition of multi-component solutions of inorganic salts by their Raman spectra using artificial neural networks. From the point of view of data analysis, one of the key problems here is high input dimensionality of the data, as the spectrum is usually recorded in 1–2 thousand channels. The two main approaches used for dimensionality reduction are feature selection and feature extraction. In this paper, three feature extraction methods are compared: channel aggregation, principal component analysis, and discrete wavelet transformation. It is demonstrated that for neural network solution of the inverse problem of determination of salt composition, the best results are provided by channel aggregation.

Prediction of relativistic electron flux in the Earth’s outer radiation belt at geostationary orbit by adaptive methods

The paper investigates the possibilities of the prediction of the time series of the flux of relativistic electrons in the Earth’s outer radiation belt by parameters of the solar wind and the interplanetary magnetic field measured at the libration point and by the values of the geomagnetic indices. Different adaptive methods are used (namely, artificial neural networks, group method of data handling, and projection to latent structures). The comparison of quality indicators of predictions with a horizon of 1–12 h between each other and with the trivial model prediction has shown that the best result is obtained for the average value of the responses of three neural networks that have been trained with different sets of initial weights. The prediction result of the group method of data handling is close to the result of neural networks, and the projection to latent structures is much worse. It is shown that an increase in the prediction horizon from 1 to 12 h reduces its quality but not dramatically, which makes it possible to use these methods for medium-term prediction.

Neural Network Solution of an Inverse Problem in Raman Spectroscopy of Multi-component Solutions of Inorganic Salts

The paper presents a study into several aspects of solution of the inverse problem on determination of concentrations of components in a multi-component water solution of inorganic salts by processing Raman spectra of the solutions by perceptron type artificial neural networks. The studied aspects are: (1) determination of the optimal architecture of a multi-layer perceptron, (2) influence of the input dimensionality reduction by aggregation of adjacent spectral channels on the error of problem solution. The results are compared for two data arrays including spectra of solutions of: (1) 5 salts composed of 10 different ions (salt determination problem), and (2) 10 salts composed of 10 different ions (ion determination problem).

A New Type of a Wavelet Neural Network

Wavelet transformation uses a special basis widely known for its unique properties, the most important of which are its compactness and multiresolution (wavelet functions are produced from the mother wavelet by transition and dilation). Wavelet neural networks (WNN) use wavelet functions to decompose the approximated function. However, for a standard wavelet basis with fixed transition and dilation coefficients, the decomposition may be not optimal. If no inverse transformation is needed, the values of transition and dilation coefficients may be determined during network training, and the windows corresponding to various wavelet functions may overlap. In this study, we suggest a new type of a WNN—Adaptive Window WNN (AWWNN), designed primarily for signal processing, in which window positions and wavelet levels are determined with a special iterative procedure. Two modifications of this new type of WNN are tested against linear model and multi-layer perceptron on Mackey-Glass benchmark problem.

Adaptive Neuro-Fuzzy Inference System Used to Classify the Measurements of Chemical Sensors

Many data processing problems are successfully solved by artificial neural networks (ANN) possessing the property of a universal approximator. However, in case when the number of data patterns available is small, ANN may tend to overtrain and not to generalize well enough. An alternative is use of such a biologically inspired cognitive architecture as fuzzy networks, or Adaptive Neuro-Fuzzy Inference Systems (ANFIS), based on the notions of fuzzy logics and often used in control systems. Like conventional ANN, ANFIS can be also trained by example with error backpropagation algorithm. In this study, we demonstrate use of neuro-fuzzy networks to solve a classification problem for high-dimensional, highly variable and noisy data of chemical sensors. The results are compared to those obtained by a multi-layer perceptron ANN and by linear regression.