Devising a method for constructing the optimal model of time series forecasting based on the principles of competition

Abstract

This paper reports the analysis of a forecasting problem based on time series. It is noted that the forecasting stage itself is preceded by the stages of selection of forecasting methods, determining the criterion for the forecast quality, and setting the optimal prehistory step. As one of the criteria for a forecast quality, its volatility has been considered. Improving the volatility of the forecast could ensure a decrease in the absolute value of the deviation of forecast values from actual data. Such a criterion should be used in forecasting in medicine and other socially important sectors. To implement the principle of competition between forecasting methods, it is proposed to categorize them based on the values of deviations in the forecast results from the exact values of the elements of the time series. The concept of dominance among forecasting methods has been introduced; rules for the selection of dominant and accurate enough predictive models have been defined. Applying the devised rules could make it possible, at the preceding stages of the analysis of the time series, to reject in advance the models that would surely fail from the list of predictive models available to use. In accordance with the devised method, after applying those rules, a system of functions is built. The functions differ in the sets of predictive models whose forecasting results are taken into consideration. Variables in the functions are the weight coefficients with which predictive models are included. Optimal values for the variables, as well as the optimal model, are selected as a result of minimizing the functions built. The devised method was experimentally verified. It has been shown that the constructed method made it possible to reduce the forecast error from 0.477 and 0.427 for basic models to 0.395 and to improve the volatility of the forecast from 1969.489 and 1974.002 to 1607.065

Building an adaptive hybrid model for short-term prediction of power consumption using a neural network

“…Work [11] describes a procedure for determining the optimal predictive model by selecting the dominant models and combining them to obtain a final model, taking into consideration the weight of each forecast. The main results of the application of the methodology are a decrease in the average absolute error and volatility of the forecast.…”

“…According to [11], simple basic models can be combined by the selective or hybrid method. The selective method involves the continuous selection of the model with the best indicator for the selected criterion and switches to the appropriate model.…”

Integrating Bio Principles in Assembly Task Design to Enhance Symbiotic Human-Robot Collaboration

“…The work ofMulesa et al (2023) shed light on the project planning phase by emphasizing the construction of a task hierarchy and its subsequent distribution among executors. Their approach is marked by the introduction of e ciency indicators and the proposition of iterative methods aimed at evaluating the minimal duration and cost associated with the task.The study byGarcia et al (2023) introduces a novel framework for HRC, leveraging deep learning models to manage assembly processes conducted either individually or collaboratively by humans and robots.Further contributing to the ergonomic aspect of collaborative task design, the research conducted by Navas-Reascos et al(2022) presents a prototype and simulation to integrate a collaborative system in a wire harness assembly process.…”