There are a lot of forecasting methods developed till current time. All of them have their own advantages and disadvantages. To decide, which method fits into the certain situation, a comparison is required. One of the highly developed trends now is a fuzzy cognitive maps forecasting. There are a lot of new methods were recently proposed. Among them there are multi-step gradient method, method which is using real-coded genetic algorithm and method which is using structure optimization genetic algorithm. In this paper a comparison of several fuzzy cognitive maps multidimensional forecasting methods is provided. Main purpose of current paper is to compare fuzzy cognitive maps forecasting methods to each other and to compare all of them to popular and widely spread forecasting methods: linear regression and random forest.
One of the prospective directions to improving the quality of decision-making in the energy system with renewable energy sources (RES) is development and implementation of automation systems based on intelligent algorithms using historical data about energy consumption and power generation. Similar solutions in addition to reducing the load on the ecology, allow to obtain a significant economic effect due to optimum switching between energy sources, taking into account the current value of energy produced and the tariff plan. Decision trees (DT) have been recognized as interpretable, efficient, problem independent and scalable architectures. In case of fuzzy representation, there is no procedure of automation tree building. In other words, existing approaches of building DT and fuzzy DT (FDT) cannot provide automatically generate fuzzy sets and fuzzy knowledge bases to build FDT. Paper presents: i) an optimal algorithm for switching among power sources based on the forecasted data and well known hybrid renewable energy systems (HRES) using genetic algorithm and multiagent technologies approach, ii) a new method of building fuzzy decision trees called decision trees based on fuzzy rules (DTFR). This method combines tree growing and pruning, to determine the structure of the FDT, to improve its generalization capabilities. This paper proposes a method (DTDR) considered as a variant of decision tree inductive using fuzzy set theory.
21 May 2018 in Journal of Physics: Conference Series
The article describes a method for forecasting sales of medicines in conditions of data sampling, which is insufficient for building a model based on historical data alone. The developed method is applicable mainly to new drugs that are already licensed and released for sale but do not yet have stable sales performance in the market. The purpose of this study is to prove the effectiveness of the suggested method forecasting drug sales, taking into account the selected factors of influence, revealed during the review of existing solutions and analysis of the specificity of the area under study. Three experiments were performed on samples of different volumes, which showed an improvement in the accuracy of forecasting sales in small samples.
Research is focused on prevention of overheating in office buildings and shopping malls during mid-season season (saving potential is about 25% of total annual building gas consumption in Benelux countries). The majority of control algorithms of heating ventilation and air condition systems (HVAC) don't take into consideration future weather conditions. We pair existing HVAC systems with a cloud server to use more efficient algorithms on existing systems, develop supervisory on/off weather compensated control (Sup-WCC) and supervisory model predictive control (Sup-MPC) algorithms. Sup-WCC was implemented in production and Sup-MPC was tested on data from 3080 sq. m. building, located in The Netherlands.
The main purpose of this paper is to identify factors that affect sales volumes of sowing crops and develop a method for the most accurate forecasting of their sales to support decision making and improve the efficiency of business processes of agro-industrial companies. This article describes the developed approach to the forecasting of sales volumes of sowing crops, which includes the identification of factors that affect sales, the formation of a training sample, and a comparison of methods for constructing mathematical models. For the construction of forecasts, linear regression methods, random forests and a neural network are used. Also, the article describes a software platform that builds forecasts of sales of crops, using R and ShinyApps.