Все выпуски

Данная работа предлагает новый подход к построению высокоточных маршрутов на основе данных от транспортных детекторов в пакете моделирования трафика SUMO. Существующие инструменты, такие как flowrouter и routeSampler, имеют ряд недостатков, таких как отсутствие взаимодействия с сетью в процессе построения маршрутов. Наш rlRouter использует мультиагентное обучение с подкреплением (MARL), где агенты — это входящие полосы движения, а окружающая среда — дорожная сеть. Добавляя в сеть транспортные средства с определенными маршрутами, агенты получают вознаграждение за сопоставление данных с детекторами транспорта. В качестве алгоритма мультиагентного обучения с подкреплением использовался DQN с разделением параметров между агентами и LSTM-слоем для обработки последовательных данных.

Поскольку rlRouter обучается внутри симуляции SUMO, он может лучше восстанавливать маршруты, принимая во внимание взаимодействие транспортных средств внутри сети друг с другом и с сетевой инфраструктурой. Мы смоделировали различные дорожные ситуации на трех разных перекрестках, чтобы сравнить производительность маршрутизаторов SUMO с rlRouter. Мы использовали среднюю абсолютную ошибку (MAE) в качестве меры отклонения кумулятивных данных детекторов и от данных маршрутов. rlRouter позволил добиться высокого соответствия данным с детекторов. Мы также обнаружили, что, максимизируя вознаграждение за соответствие детекторам, результирующие маршруты также становятся ближе к реальным. Несмотря на то, что маршруты, восстановленные с помощью rlRouter, превосходят маршруты, полученные с помощью инструментов SUMO, они не полностью соответствуют реальным из-за естественных ограничений петлевых детекторов. Чтобы обеспечить более правдоподобные маршруты, необходимо оборудовать перекрестки другими видами транспортных счетчиков, например, детекторами-камерами.

This work provides a new approach for constructing high-precision routes based on data from transport detectors inside the SUMO traffic modeling package. Existing tools such as flowrouter and routeSampler have a number of disadvantages, such as the lack of interaction with the network in the process of building routes. Our rlRouter uses multi-agent reinforcement learning (MARL), where the agents are incoming lanes and the environment is the road network. By performing actions to launch vehicles, agents receive a reward for matching data from transport detectors. Parameter Sharing DQN with the LSTM backbone of the Q-function was used as an algorithm for multi-agent reinforcement learning.

Since the rlRouter is trained inside the SUMO simulation, it can restore routes better by taking into account the interaction of vehicles within the network with each other and with the network infrastructure. We have modeled diverse traffic situations on three different junctions in order to compare the performance of SUMO’s routers with the rlRouter. We used Mean Absoluter Error (MAE) as the measure of the deviation from both cumulative detectors and routes data. The rlRouter achieved the highest compliance with the data from the detectors. We also found that by maximizing the reward for matching detectors, the resulting routes also get closer to the real ones. Despite the fact that the routes recovered using rlRouter are superior to the routes obtained using SUMO tools, they do not fully correspond to the real ones, due to the natural limitations of induction-loop detectors. To achieve more plausible routes, it is necessary to equip junctions with other types of transport counters, for example, camera detectors.

Modeling banking systems using a network approach has received growing attention in recent years. One of the notable models is that developed by Iori et al, who proposed a banking system model for analyzing systemic risks in interbank networks. The model is built based on the simple dynamics of several bank balance sheet variables such as deposit, equity, loan, liquid asset, and interbank lending (or borrowing) in the form of difference equations. Each bank faces random shocks in deposits and loans. The balance sheet is updated at the beginning or end of each period. In the model, banks are grouped into either potential lenders or borrowers. The potential borrowers are those that have lack of liquidity and the potential lenders are those which have excess liquids after dividend payment and channeling new investment. The borrowers and the lenders are connected through the interbank market. Those borrowers have some percentage of linkage to random potential lenders for borrowing funds to maintain their safety net of the liquidity. If the demand for borrowing funds can meet the supply of excess liquids, then the borrower bank survives. If not, they are deemed to be in default and will be removed from the banking system. However, in their paper, most part of the interbank borrowing-lending mechanism is described qualitatively rather than by detailed mathematical or computational analysis. Therefore, in this paper, we enhance the mathematical parts of borrowing-lending in the interbank market and present an algorithm for simulating the model. We also perform some simulations to analyze the effects of the model’s parameters on banking stability using the number of surviving banks as the measure. We apply this technique to analyze the effects of a macroprudential policy called loan-to-deposit ratio based reserve requirement for banking stability.

Modeling banking systems using a network approach has received growing attention in recent years. One of the notable models is that developed by Iori et al, who proposed a banking system model for analyzing systemic risks in interbank networks. The model is built based on the simple dynamics of several bank balance sheet variables such as deposit, equity, loan, liquid asset, and interbank lending (or borrowing) in the form of difference equations. Each bank faces random shocks in deposits and loans. The balance sheet is updated at the beginning or end of each period. In the model, banks are grouped into either potential lenders or borrowers. The potential borrowers are those that have lack of liquidity and the potential lenders are those which have excess liquids after dividend payment and channeling new investment. The borrowers and the lenders are connected through the interbank market. Those borrowers have some percentage of linkage to random potential lenders for borrowing funds to maintain their safety net of the liquidity. If the demand for borrowing funds can meet the supply of excess liquids, then the borrower bank survives. If not, they are deemed to be in default and will be removed from the banking system. However, in their paper, most part of the interbank borrowing-lending mechanism is described qualitatively rather than by detailed mathematical or computational analysis. Therefore, in this paper, we enhance the mathematical parts of borrowing-lending in the interbank market and present an algorithm for simulating the model. We also perform some simulations to analyze the effects of the model’s parameters on banking stability using the number of surviving banks as the measure. We apply this technique to analyze the effects of a macroprudential policy called loan-to-deposit ratio based reserve requirement for banking stability.

Важнейшим компонентом распределенной вычислительной системы является сетевая инфраструктура. Несмотря на то что сеть составляет основу такого рода систем, она часто является незаметным партнером для систем хранения и вычислительных ресурсов. Мы предлагаем интегрировать сетевой элемент напрямую в распределенные системы через уровень управления нагрузками. Для такого подхода имеется достаточно предпосылок. Так как сложность и требования к распределенным системам растут, очень важно использовать имеющуюся инфраструктуру эффективно. Например, одни могут использовать измерения качества сетевых соединений в механизмах принятия решений в системе управления задачами. Кроме того, новейшие технологии позволяют другим задавать сетевую конфигурацию программно, например используя ПКС — программно-конфигурируемые сети. Мы опишем, как эти методы используются в системе управления задачами PanDA, применяемой коллаборацией ATLAS.

A crucial component of distributed computing systems is network infrastructure. While networking forms the backbone of such systems, it is often the invisible partner to storage and computing resources. We propose to integrate Network Elements directly into distributed systems through the workload management layer. There are many reasons for this approach. As the complexity and demand for distributed systems grow, it is important to use existing infrastructure efficiently. For example, one could use network performance measurements in the decision making mechanisms of workload management systems. New advanced technologies allow one to programmatically define network configuration, for example SDN — Software Defined Networks. We will describe how these methods are being used within the PanDA workload management system of the ATLAS collaboration.