Authors

Hybrid digital model based on Neural ODE in the task of increasing the economic efficiency of processing small-ore raw materials

The results of a study are presented, the purpose of which was to develop the structure of a hybrid digital model for managing the processes of processing small-ore raw materials, as well as an algorithm for converting technological data in accordance with this structure, ensuring improved management quality and, as a consequence, the economic efficiency of processing. The original idea underlying the hybrid digital model is the use of neural ordinary differential equations (Neural ODE) to calculate the dynamics of technological objects and the processes implemented in them. Neural ODEs are a type of physics-motivated neural networks that use physical laws during their learning process. The resulting digital intelligent machine learning system is capable of highly accurate reconstruction of the dynamics function using observational data of a technological object or process. The proposed hybrid model provides for the joint use of Neural ODE and Simulink simulation models of technological processes for processing fine ore raw materials when calculating control actions. This allows you to quickly model and analyze the reaction of dynamic objects to control inputs and quickly make the necessary changes without waiting for the reaction of the physical original. Numerical experiments have shown that the use of Neural ODE as part of a hybrid digital model accurately reproduces the dynamics of technological objects under various initial conditions. For comparison, experiments were carried out with a model in which an LSTM recurrent neural network was used instead of Neural ODE. Experiments demonstrated that in the latter case, the dynamics were simulated with high accuracy only under the original initial conditions, and when they changed, it was severely degraded. At the same time, the use of Neural ODE instead of LSTM has shown consistently high accuracy in displaying dynamics under these changes, which will help improve the quality of control of technological processes for processing fine ore raw materials and their economic efficiency. Read more...

Neuro-fuzzy model of resource provision of innovative activity of an industrial enterprise

A neuro-fuzzy model of resource provision of innovative activity of an industrial enterprise is proposed. The model implements a two-stage procedure for describing and managing innovative activity of an industrial enterprise: at the first stage, interaction resources are classified based on the supplemented VRIO analysis of the interaction profile; at the second stage, an innovative activity strategy is selected. The neuro-fuzzy model of resource provision is based on stacking of private machine learning models, such as the k-nearest neighbors method, random forest, and multilayer perceptron. The classification results of private models are combined using a trained tree of fuzzy inference systems that performs the final classification, which ensures an increase in its accuracy compared to individual private models. A distinctive feature of the model is the use of a fuzzy logical inference system to assess the probability of resource availability used in planning the need for it, which allows taking into account expert judgments as input data. Testing of the neuro-fuzzy model, carried out in the MatLab software system using the example of solving the problem of assessing the resource provision of an innovation process during the interaction of a regional instrument-making enterprise with one of the counterparties, demonstrated the model’s performance and high accuracy of classifying the resources of innovative interaction. Read more...

Forecasting irregular time series based on LSTM networks and accounting for sampling interval correlation

A method for forecasting a nonequidistant (irregular) time series with an irregular sampling interval is presented. Data presented as irregular time series are often encountered in various fields, such as healthcare, biomechanics, economics, climatology, and others. Forecasting irregular time series is in demand in these fields for early warning and proactive decision-making, but there is no universal method for taking into account the unevenness of sampling in the forecast, which determines the relevance of research in this area. The purpose of the study was to develop a method for forecasting a nonequidistant series based on deep neural networks, which allows for good forecast accuracy with a relatively lightweight network architecture. The novelty of the research results lies in the developed method for forecasting nonequidistant time series, the architecture of the deep neural network, and the algorithm that implements the proposed forecast method. The method uses a closed loop, in which the forecast results at the current step are used at the following steps. The original feature of the proposed forecasting method is the use of a multilayer perceptron to forecast the duration of the next irregular sampling interval. This interval is calculated taking into account the correlation time calculated based on the autocovariance function of the durations of irregular sampling intervals. A distinctive feature of the proposed architecture is the presence of a separate input channel of neural network data for analyzing the values of sampling intervals, which allows forecasting the next value of the series taking into account the duration of the forecasted sampling interval. The method is developed for a one-dimensional series, but it can be extended to multidimensional series if the synchronicity of the sampling of the components of the series is observed. The computational experiments showed that with low requirements for computing resources, the accuracy of the forecast based on the proposed method is comparable to modern forecast models within the correlation interval. Read more...