OpenTalks.AI - Дмитрий Спандерашвили, Искусственный интеллект для задач управления предприятием в реальном времени

Transcript

1. Artificial Intelligence solutions for real time enterprise management Dmitry Spanderashvili Project manager, PhD
2. Development of adaptive real time management technology Release of a solution for field staff management 2016 Implementations in large companies New solutions for manufacturing 2017 Products for telecoms, logistics, oil industry 2018 Adeptik
3. Development and implementation of business applications for the effective management of enterprise resources in real time What are we doing?
4. Modern business must be flexible and capable of making quick decisions The human brain is a bad decision-making tool in real time - it makes decisions slowly and inefficiently New management paradigm
5. 1 Gradual removal of a person from making management decisions Continuous automation of the main enterprise activities Use of modern mathematics New management paradigm implies 2 3
6. • Restrictions What plan can be considered feasible in the first place? • Objective function How to compare two feasible plans? Which one is better? Operational management There is always a place of randomness The rules of the game are changing every minute Workflow management has to be adaptive – constantly changing plans depending upon the situation
7. To select the optimal simple plan from 10 steps, you need to consider 1000 options. Out of 100 steps ~ 1030 options Exact algorithm solving the problem Тask size Time “Greedy” algorithm You should not expect an optimal solution from either the “greedy” algorithm or the linear mathematical programming Operational complexity
8. Hundreds of oil trucks Thousands of gas station tanks Ten oil farms • Drain the entire volume of the tank • Drain only from the end of the composition • Compliance of petroleum products • Time windows at gas stations • Paperwork Restrictions: • Minimization of overprocessing • Minimization of gas station drainage • Cost minimization Target function: Oil industry case
9. • Far from optimal solution • Quick calculations Linear Mathematical Programming • Work with simplified models • Slower calculations with respect to "greedy" algorithms • Higher accuracy with respect to greedy algorithms • Exact result for a limited number of subject areas Greedy Algorithms • Poor multiprocessing architecture Types of algorithms
10. Search and optimization Evolutionary and Swarm Methods Logical conclusion Probabilistic methods and fuzzy reasoning Classifiers and statistical learning methods Artificial Neural Networks Knowledge Representation Models Artificial Intelligence Technologies
11. Based on modeling the behavior of social, biological and physical systems The degree of approximation to the optimal solution Algorithm time Approximate ‘’greedy’’ algorithm Swarm algorithm Swarm intelligence and multiagency
12. Resistance to input changes Fast approximate result and its continuous improvement The ability to scale calculations - an increase in computing resources allows you to quickly find a better option Algorithmic support and computing infrastructure that differs from the classical Swarm intelligence and multiagency 1 2 3
13. Oil industry case
14. Oil industry case
15. Oil industry case
16. Oil industry case
17. Oil industry case reduction of up to 30% of the fleet of fuel trucks with a minimum of drainage Case summary:
18. • Labor productivity growth by 20-40% • Reducing the cost of spare parts and fuel to 15% • Improving customer relationships due to improved service quality (attracting new customers) • The reduction of the manufacturing cycle by 20-60% • Relevant decrease in tied capital and required inventory • The ability to always have a current production plan Service Companies Industrial manufacturing Cases for service companies and manufacturing
19.  Modern business requires more efficient management of operational processes than human can provide  Modern mathematics provides effective tools for solving such problems  Technologies are also ready: scaling computing is not a problem - neither in price nor in complexity of use Conclusion
20. sales@adeptik.com snrd.ru +7 (495) 120-19-60