Real-time decision-making in industrial mining complexes

An industrial mining complex or mineral value chain is an engineering system managing the extraction of materials from a group of mines and the treatment of the extracted materials through different interconnected processing facilities and waste/tailings locations, all leading to sellable products delivered to various customers and/or the spot market. The inherent uncertainty in the supply of materials from the mineral deposits mined, the market demand, as well as the various related operational aspects is the core element critically affecting decision-making. These sources of uncertainty are the central facets of this integrated business and the related planning, production scheduling, performance forecasting and product delivery.

Research by GERAD members Roussos Dimitrakopoulos and Michel Gamache contributes new advances towards an all-inclusive supply-meets-demand simultaneous stochastic optimization framework. This framework includes the development of new stochastic, intelligent and self-learning approaches, enhancing the capability of a mining complex optimization to respond to new unveiling information in operating environments in real-time. It also supports related operational decision-making in addition to adapting both strategic and operational policies, including environmental aspects. The so-termed “self-learning mining complex” is viewed as a future research outcome incorporating in real-time new unveiling production data, including sensor data, and market information, to dynamically adapt operational, production and scheduling decisions. At present, this is approached via self-play reinforcement learning including a multiple agent setting that integrates uncertainties and manages related risk. New developments are tested and evaluated in actual case studies.

Examples of related publications:

• Kumar, A., Dimitrakopoulos, R., Maulen, M., Adaptive self-learning mechanisms for updating short-term production decisions in an industrial mining complex. Journal of Intelligent Manufacturing, 31(7), 1795–1811 2020.
• Rimélé, A., Dimitrakopoulos, R., Gamache, M., A dynamic stochastic programming approach for open-pit mine planning with geological and commodity price uncertainty, Resources Policy, 65, 101570, 2020.
• de Carvalho, J.P., Dimitrakopoulos, R., Integrating production planning with truck-dispatching decisions through reinforcement learning while managing uncertainty, Minerals, 11(6), 587, 2021.

The Case of Telecommunications Infrastructure

As the world becomes "smart", with "smart cities", "smart grids", "smart buildings", etc., it is also becoming increasingly reliant on its telecommunications infrastructure. To answer the needs of smart applications, the underlying infrastructure must be extremely reliable, sustainable and highly adaptable to the system's changing conditions. Current and forthcoming smart systems increasingly require telecommunication networks that are ubiquitous, all pervasive, highly performant and transparent to users. This smoothness in network operation requires a complex, large scale and highly heterogeneous telecommunications infrastructure. Wireless, optical, satellite, aerial, computing and storage technology must efficiently work together to provide the availability and the response time that is required by smart applications. Networks need to rearrange their resource offering in the nick of the time, and that is only possible with the increased use of virtualization to define network functions, alongside a form of centralized network intelligence based on the concept of Software Defined Networking (SDN). Thus, telecommunications infrastructure itself must be the smartest, for all the other smart systems relying on it to function properly in real time.

An example of such infrastructure is currently provided by the deployment of 5G in large cities that may set virtual networks, called "slices", to different smart systems, such as intelligent transportation, connected vehicles or telemedicine, each having applications with different real-time performance requirements. The challenge for operators is to ensure that those systems are always available, and that the response time provided by the infrastructure is suitable for each smart application. To help in this task, a citywide 4G/5G simulator has been developed at GERAD by Professor Brunilde Sansò's research group. The large-scale discrete event simulator recreates equipment specifications normalized by the telecommunications standards entities. It also uses real telecommunications and city infrastructure data to assess the response time of key applications and to detect problems in the network. One of the major challenges tackled by the research group is the large scale modelling of the system that may contain hundreds of base stations operating at the millisecond level in a citywide networking mode. Another challenge is the mapping of the citywide application into the simulated infrastructure. From the time-scale standpoint, the mapping implies assessing the application performance, in minutes or hours while simulating milliseconds, which involves advanced statistical and machine learning modelling. Another challenge arises from mobility applications, such as V2V or ITS that, in their path towards their destination, make use of different telecommunications equipment. Thus, the mapping needs advanced algorithms for distributed computations. Finally, machine learning methods are put in place to assess the performance of the applications and to identify network problems, such as dormant cells or denial of service attacks. An on-line version of the simulator can be accessed here.

A final thought on real-time decision-making and telecommunication infrastructure is that the "smartness", reliability and high availability of such infrastructure comes with a hidden price: increased energy consumption and environmental impact of data centres and network components. For years, Professor Sansò's group has been interested in ways of guaranteeing network reliability and availability while reducing energy consumption and environmental impact. Among others, the intelligent real-time operation of data centres, the design and operation of wireless access with energy constraints and the study of networks fed by solar energy. The next step currently being developed as an extension of the above-mentioned citywide simulator is how to integrate unreliable energy sources to insure real-time response of smart applications even in challenged and catastrophic conditions.

References:

• Manzanilla-Salazar, O.G., Malandra, F., Mellah, H., Wette, C., Sansò, B., A Machine Learning Framework for Sleeping Cell Detection in a Smart-City IoT Telecommunications Infrastructure. IEEE Access, 8, 61213-61225, 2020.

• Seyedi, Y., Karimi, H., Wette, C., Sansò, B., A New Approach to Reliability Assessment and Improvement of Synchrophasor Communications in Smart Grids. IEEE Transactions on Smart Grid, 11(5), 4415-4426, 2020.

• D'Amours, M., Girard, A., Sansò, B., Planning Solar in Energy-managed Cellular Networks. IEEE Access, 6, 65212-65226, 2018.

• Malandra, F., Chiquette, L.O., Lafontaine-Bedard, L.P., Sansò, B., Traffic characterization and LTE performance analysis for M2M communications in smart cities. Pervasive and Mobile Computing, 48, 59-68, 2018.

• Larumbe, F., Sansò, B., A Tabu Search Algorithm for the Location of Data Centers and Software Components in Green Cloud Computing Networks. IEEE Transactions of Cloud Computing, 1(1), 22-35, 2013.

• Boiardi, S., Capone, A., Sansò, B., Radio planning of energy-aware cellular networks. Computer Networks, 57(13), 2564-2577, 2013.

Rescheduling Decisions in Logistics Applications

In many logistics applications, it is of the utmost importance to be able to take into account the latest data and make decisions to readjust existing plans on a real-time basis.

In retail operations, it is not only important to make the right inventory decisions, but personnel planning is also an essential task which contributes to customer satisfaction. This is a complex task because a personnel schedule must take into account various constraints such as union rules and legislative requirements, as well as employee preferences and cost considerations. However, the planned schedule cannot always be executed due to sudden changes in demand or workforce availability. In such a case, it is very important to be able to generate a new, feasible schedule taking the new situation into account. Ideally, this new schedule does not lead to too many modifications or a big cost increase. Since such a new schedule must be generated in real time, Guy Desaulniers and Issmail El Hallaoui developed effective and quick heuristics that can solve large rescheduling problems in a matter of seconds.

Transportation provides another application area for real-time decision-making. Antoine Legrain considers the case of ride-sharing, where requests from different customers must very frequently be updated and reoptimized. To make efficient and cost-effective real-time decisions in the context of sequential decision-making under uncertainty, Yossiri Adulyasak has proposed decomposition frameworks for Markov decision processes (MDPs) under uncertainty and demonstrated the efficiency of the resulting policies in inventory management and robot planning. This axis is also perfectly aligned with the research mission of the Canada Excellence Research Chair in Data Science for Real-Time Decision-Making led by Andrea Lodi.

References:

• Adulyasak, Y., Varakantham, P., Jaillet, P., Solving uncertain MDPs with objectives that are separable over instantiations of model uncertainty. 29th Conference on Artificial Intelligence (AAAI), 29(1), 3454-3460, 2015.

• Ahmed, A., Varakantham, P., Adulyasak, Y., Jaillet, P., Regret based robust solutions for uncertain Markov decision processes. Advances in Neural Information Processing Systems (NeurIPS) 26, 2013.

• Hassani, R., Desaulniers, G., El Hallaoui, I., Real-time personnel re-scheduling after a minor disruption in the retail industry. Computers & Operations Research, 120, 104952, 2020.

• Hassani, R., Desaulniers, G., El Hallaoui, I., Real-time bi-objective personnel re-scheduling in the retail industry. European Journal of Operational Research, 293(1), 93-108, 2021.

• Riley, C., Legrain, A., Van Hentenryck, P., Column generation for real-time ride-sharing operations. In International Conference on Integration of Constraint Programming, Artificial Intelligence, and Operations Research. Springer, Cham, 472-487, 2019.