Neural networks application in managing the energy efficiency of industrial enterprise

The rise of robo-advisor wealth management services, which constitute a key element of fintech revolution, unveils the question whether they can dominate human-based advice, namely how to address the client’s behavioral biases in an automated way. One approach to it would be the application of machine learning tools during client profiling. However, trained neural network is often considered as a black box, which may raise concerns from the customers and regulators in terms of model validity, transparency, and related risks. In order to address these issues and shed more light on how neurons work, especially to figure out how they perform computation at intermediate layers, this paper visualizes and estimates the neurons’ sensitivity to different input parameters. Before it, the comprehensive review of the most popular optimization algorithms is presented and based on them respective data set is generated to train convolutional neural network. It was found that selected hidden units to some extent are not only specializing in the reaction to such features as, for example, risk, return or risk-aversion level but also they are learning more complex concepts like Sharpe ratio. These findings should help to understand robo-advisor mechanics deeper, which finally will provide more room to improve and significantly innovate the automated wealth management process and make it more transparent.

Transforming the energy sector to provide universal access to reliable and modern energy services is an essential task for Ukraine, one of the Eastern Partnership countries with heavy energy dependence. It will help accelerate the achievement of the Sustainable Development Goals. The paper is devoted to studying Ukraine’s readiness to generate sustainable energy compared to the EU and other Eastern Partnership countries and the development of an information base for monitoring the achievement of SDG7.
The data from the World Energy Council (WEC), the International Energy Agency (IEA), and the State Statistics Service of Ukraine are analyzed. Thus, the study proposed to expand the list of national monitoring indicators that more fully reflect the social, economic, and environmental results of SDG7 “Affordable and clean energy” in Ukraine. The development of an information monitoring base expands the opportunities to assess the availability, sustainability, and balance of national energy policy in green economic transformation. Furthermore, the indicators of energy intensity, carbon intensity, as well as access to sustainable energy for the population and business are emphasized. The findings are aimed to raise the level of awareness of government agencies and make balanced decisions to accelerate the achievement of SDG7 in Ukraine.

The problem of calibration of local volatility model of Dupire has been formalized. It uses genetic algorithm as alternative to regularization approach with further application of gradient descent algorithm. Components that solve Dupire’s partial differential equation that represents dynamics of underlying asset’s price within Dupire model have been built. This price depends in particular on values of volatility parameters. Local volatility is parametrized in two dimensions (by Dupire model): time to maturity of the option and strike price (execution price). On maturity axis linear interpolation is used while on strike axis we use B-Splines. Genetic operators of mutation and selection are then applied to parameters of B-Splines. Resulting parameters allow us to obtain the values of local volatility both in knot points and intermediate points using interpolation techniques. Then we solve Dupire equation and calculate model values of option prices. To calculate cost function we simulate market values of option prices using classic Black-Scholes model. An experimental research to compare simulated market volatility and volatility obtained by means of calibration of Dupire model has been conducted. The goal is to estimate the precision of the approach and its usability in practice. To estimate the precision of obtained results we use a measure based on average deviation of modeled local volatility from values used to simulate market prices of the options. The research has shown that the approach to calibration using genetic algorithm of optimization requires some additional manipulations to achieve convergence. In particular it requires non-uniform discretization of the space of model parameters as well as usage of de Boor interpolation. Value 0.07 turns out to be the most efficient mutation parameter. Using this parameter leads to quicker convergence. It has been proved that the algorithm allows precise calibration of local volatility surface from option prices.