DIGITAL EDITION

 

The current issue of ORMS Today is now available to INFORMS members in "digital magazine" format. Log in to the Digital Editions Archive and enjoy a new online reading experience.

Software Surveys

  STATISTICAL ANALYSIS SOFTWARE SURVEY
Biennial survey of statistical analysis software...
more »

Kazakhstan National Railway

How a railway-planning model helped develop a scheduled, market-driven, customer service-oriented operation across a vast Central Asian nation.

By Yury Gorbunov, David Hunt and Marc Meketon

KZT railway train

KTZ operates a rail network of 14,800 kilometers and its freight traffic in 2013 totaled 231 billion ton-kilometers – 43 percent of the country’s total freight movement.

There are two fundamental philosophies for operating a freight railroad: tonnage-based and scheduled. In a tonnage-based operation, a train will be held in a yard until it has been loaded with a predefined number of cars. This guarantees long trains, which intuitively are more efficient than a larger number of smaller trains. However, this can lead to poor service quality and requires more deadheading of labor and equipment.

In scheduled railroading, freight trains run on a fixed schedule regardless of whether or not the train is full. A scheduled operation improves customer service, thereby improving the ability of railroads to compete with long-haul trucks for higher-valued, more time-sensitive goods. A well-tuned plan also produces long trains but needs less repositioning of locomotives and crews and is less costly. All North American railways have converted to scheduled operations in the last 20 years as software and operations research tools for designing well-tuned plans have advanced.

One of the early adopters of scheduled railroading was the Canadian Pacific Railway (CP), which worked with MultiModal Applied Systems (now part of Oliver Wyman) to explore the concept. CP was pressured by customers to improve service quality, but concerned that a fixed timetable would greatly increase costs. A schedule-based operating plan was developed to match traffic patterns with optimized blocking and routing plans, with a focus on reducing equipment miles, train miles and train hours. This work was recognized in 2003 with the prestigious Franz Edelman Award from INFORMS.

Over the past several years, the authors have been working with the Kazakhstan National Railway to design and build a network planning system based on the concepts in the Edelman Award-winning work. The Kazakh railway, part of the vast former Soviet Union rail system, was firmly entrenched in a tonnage-based operation. While many planning issues are similar with North American railroads, there also are notable exceptions that have required new models and concepts.

Kazakhstan Temir Zholy

Figure 1: Map of the KTZ rail network.

Figure 1: Map of the KTZ rail network.

Kazakhstan is located in Central Asia, sharing a northern border with Russia, a southeastern border with China, and southern borders with Turkmenistan, Uzbekistan and Kyrgyzstan. It is the ninth largest country by area in the world, and the largest landlocked country. Rich reserves of oil, coal and ores are used domestically and also exported to help spur robust economic growth. Kazakhstan is also an important trade route for goods from China to Europe; in fact, the city of Almaty was a stop on the famed Silk Road. All of these factors contribute to the need for an efficient freight rail system to support the Kazakh economy.

Kazakhstan Temir Zholy (translation, Kazakhstan Iron Road) is the national freight railroad of Kazakhstan. KTZ operates a rail network of 14,800 kilometers and its freight traffic in 2013 totaled 231 billion ton-kilometers – 43 percent of the country’s total freight movement. There are two interchange locations between China and Kazakhstan, with many railcars traveling the width of the country before interchanging with the Russian rail system. The largest yard in the KTZ rail system, located in the capital city of Astana, receives an average of 6,000 railcars a day.

The railway-planning problem for general freight traffic on KTZ can be broken down into two major issues. The first is devising an aggregation strategy, called a blocking plan. The plan specifies “blocks” of railcars that are temporarily grouped together for movement from origin to destination. Good blocking plans consider car-kilometers, car-handlings and respect physical limitations of the yards and network. The design of the blocks strongly influences the efficiency of the railway and the service received by the customers.

The second planning issue is to form the trains that carry blocks. A train is characterized by the days it operates, the route it takes, the planned arrival and departure times along the route, and which blocks are picked up and set out along the route. In general, longer trains are preferred, but the timing/frequency of trains determines transit time and hence customer service. The train plan also must balance the number of trains running in each direction to reduce repositioning of locomotive and crew assignments.

KTZ’s operating plan consisted of running single-block trains short distances, typically between two adjacent major yards, several times per day. A railcar would go from origin to destination by passing through major yard after major yard, with intermediate handlings to place the car on different trains at each yard. Occasionally KTZ would run single-block “bypass trains” that could skip over a major yard to reduce car handlings. They did so only if they could run two or more trains per day that bypassed the yard, and if there was still sufficient traffic to justify continuing to operate the existing trains between adjacent yards.

Revising the KTZ Operating Plan

KTZ began to use our MultiRail™ network planning software to assist in developing its operating plan. As was the case with CP, freight planning involved four steps:

  1. Develop a traffic forecast.
  2. Use the forecast to design the blocking plan.
  3. Design trains based on the blocking plan.
  4. Use simulation and analysis to estimate yard and train workloads and identify refinements to the plan.

Creating a traffic forecast: The initial step was to create a traffic file accurately capturing KTZ’s expected freight movements. A historical perspective on loaded and empty railcar movements was developed at first, but developing a fined-tuned plan that could be used month-to-month required two additional data inputs. First was a KTZ domestic forecast of the revenue (loaded) railcar movements for the next month, aggregated to 14 regions. Second was an external forecast of loaded and empty movements of cars going to/from Kazakhstan at the various international interchange locations. As a legacy of the former Soviet Union, such a forecast is generated for all ex-CIS railways by a dedicated inter-country center in Russia.

Figure 2: Previous KTZ plan.

Figure 2: Previous KTZ plan.

Figure 3: Revised KTZ plan.

Figure 3: Revised KTZ plan.

These components were combined into a statistical/optimization model to produce a complete forecast at a station-to-station level. The model used quadratic programming concepts to find a least-squares solution of loaded and empty flows that examined:

  • the concept of “conservation of mass,” i.e., railcars into a region should equal railcars out of the region;
  • adjustments to historical loaded and empty moves, to be consistent with the domestic forecast of loaded movements and the Russian-supplied interchange forecast of loads and empties; and,
  • the addition of empty “return” moves for completely new loaded moves, which do not have a balancing empty move in historical data.

Optimizing the blocking and train plan. The next step was to develop a new blocking plan, which required first modeling the current plan. This proved to be a significant challenge, as the railway was using a text document to describe its blocking and train formation plan. A typical entry would indicate for a specific type of train at a specific yard, and based on the destination of the traffic, the next yard for the train and also the type of train – local, road, mainline unit or export.

The devil is in the details, however, and in this case the plan for most trains included notes that indicated the purpose or special rules for blocking. For example, the notes might state that a local train should be built in a specific sequence from local yards on the line. Or, that the train should take cars for certain destinations only if these cars go through a specific interchange point. There were numerous such details that were not available in a consistent, electronic form, but had to be considered and represented accurately when coded using the MultiRail planning software.

Once an accurate representation of the current plan had been completed, KTZ then proceeded to use MultiRail to optimize the plan. One optimization concept replaced the strict rules for single-block bypass trains with multiple-block trains that used bypass blocks, as illustrated in Figure 2 and Figure 3. In this example, one block of 70 cars might be forecasted to go from Yard A to Yard B every day, and then 30 of those cars travel on to Yard C as a separate block. The old plan would use Train AB to carry the 70 cars, and a Train BC to carry the 30 cars, plus other cars picked up at B and going to C. The new plan has 30 less car-handlings, thereby reducing costs and improving transit times.

The new practice, which can be easily designed using software tools, develops multi-block trains that reduce intermediate yard activities and reduce the overall dwell time for cars improving total system velocity.

Simulation and network analysis. The ability to identify and analyze block bypass opportunities and refine the blocking plan, as well as to create a proper train plan using MultiRail, was part of the CP project, and is discussed in our previous Interfaces article [1]. But for Kazakhstan, it was necessary to develop four more operations research tools, which are described below.

Yard relaxation involves optimizing blocking rules for outbound blocks originating from a specific yard, to ensure that traffic is routed onto the most economical set of blocks. For the most part, this is based on the traffic destination of each block. This has become a useful tool, as it indicates how the planner could better optimize the plan. It also provides a “sanity check” for the rules and helps to identify incompleteness and incorrect routings. The current tool works for a single yard at a time. An algorithm to globally optimize the traffic rules for the blocking plan is currently under development.

The Astana yard is a junction of four lines and is the busiest terminal in the KTZ network. It receives about 6,000 cars daily.

The Astana yard is a junction of four lines and is the busiest terminal in the KTZ network. It receives about 6,000 cars daily.

Train frequency optimization determines the optimal frequencies for trains, given the physical limits of the trains and infrastructure and has become a significant planning enhancement. As a simple example, say a train carries a single block, that block is only carried by that train, and the estimated weekly volume of the block is 700 cars. Since the maximum train size is only 71 cars, the train must run 10 times a week. The complexity of the calculation will grow, however, as the train plan becomes more complex, due to, for example, multiple blocks per train, multiple trains carrying a block, some trains preferred over other trains and “block swaps” (where a block is carried to an intermediate point by one train, then picked up in its entirety by another train to move it to its destination).

Capacitated volume analysis is a new tool to quickly estimate typical train sizes. It considers KTZ operational rules of train capacity and traffic prioritization, such as the practice of taking limited traffic that cannot fit onto a main train and transferring it to a local train. It is based on a linear programming formulation [2]. While useful for quickly designing trains, this tool trades off day-specific accuracy for computational speed, and it does not estimate transit times.

Capacitated simulation is a customized model developed to: (1) accurately estimate train-size fluctuations and yard workloads over the course of a week; (2) estimate shipment transit times; and (3) be effective for tonnage trains that remain in the system. The concept is that traffic is released over a set of weeks, is accumulated into yards, and then runs on either tonnage or scheduled trains according to the routing plan. For trains that are scheduled, if the railcars in the traffic record cannot fit onto the train due to capacity limitations, the traffic will roll over to the next train that is allowed to carry the traffic. For tonnage trains, trains are not released until full or until a threshold of time has passed.

Results to Date

Figure 4: Traffic density map for the Astana Terminal daily cars.

Figure 4: Traffic density map for the Astana Terminal daily cars.

An example of the value of using MultiRail for planning is related to the offloading of work from the major yard at Astana. The Astana yard is a junction of four lines and is the busiest terminal in the KTZ network. It receives about 6,000 cars daily, of which 2,600 cars are classified into new trains. Astana yard operates close to its maximum operating capacity. In addition, about two dozen long-distance passenger trains pass through the yard mainline tracks every day.

To reduce work in the Astana yard and increase the velocity of railcars, a new blocking scheme was proposed (Figure 5). Previously, Astana and adjacent yards would operate point-to-point trains with single blocks. A car from Karaganda to Pavlodar would be handled in Astana and Ekibastuz. In the new scheme, Karaganda would build a Pavlodar train with a block for Astana and a block for Pavlodar. The Astana-Ekibastuz trains were scheduled so they arrive to Ekibastuz about three to four hours before the Karaganda-Pavlodar train. This gives the Ekibastuz yard enough time to process cars from Astana and prepare the block for the Karaganda-Pavlodar train. This solution, although relatively simple, was quite uncommon in KTZ operations. These opportunities require extensive traffic analysis and a great deal of inter-regional collaboration and thus are difficult to identify without software tools.

Figure 5: Astana blocking plan.

Figure 5: Astana blocking plan.

The KTZ planning team went through several iterations of this idea before it was deemed compatible with all operational hurdles and limitations at each of the yards. In the end, this solution proved to be highly efficient. Almost 400 cars per day had less handlings, and velocity was increased by 15 percent. This effect was measurable even when track maintenance slowed all the trains in the corridor. Car processing at Astana fell to 2,400 cars per day, and a similar result was achieved in Ekibastuz.

There are several interesting implications from the successful deployment of such solutions by KTZ. Prior to the use of MultiRail, KTZ capacity for inter-divisional advanced train planning was very limited by data availability and complexity. The use of new software tools allowed for complex inter-divisional planning, leading to additional efficiency gains.

Figure 6: Car velocities improvements.

Figure 6: Car velocities improvements.

Moreover, it was identified that the development of solutions similar to the Astana yard example were fundamentally incompatible with existing KTZ operational metrics and organization. Production systems would only measure velocities within the 14 divisions, not over specific routes or segments. The new software tools helped KTZ develop key performance indicators that reflect the entire network efficiency.

Just as importantly, these software tools support new planning methodologies and processes that KTZ has used to create a more scheduled, market-driven, customer service-oriented operation.

Yury Gorbunov is a senior specialist in Oliver Wyman’s Transportation Practice, where he focuses on optimizing railway operations and working with clients on strategic planning.

David Hunt is a manager in Oliver Wyman’s Transportation Practice, where his expertise includes strategic planning, operational improvements and regulatory and policy analysis, with an emphasis on projects involving advanced analytics and network modeling.

Marc Meketon is a vice president in Oliver Wyman’s Transportation Practice, where he specializes in the design, development and use of software applications for network design, and where he leads the development efforts for MultiRail, which has been used to design railroad operating plans on six continents.

References

  1. Phil Ireland, Rod Case, John Fallis, Carl Van Dyke, Jason Kuehn, Marc Meketon, “The Canadian Pacific Railway Transforms Operations by Using Models to Develop Its Operating Plans,” Interfaces, Vol. 34, No. 1, January–February 2004, pp. 5–14.
  2. B.W. Patty (ed.), “Handbook of Operations Research Applications at Railroads,” International Series in Operations Research & Management Science 222, Springer Science+Business Media, LLC, 2015.
  3. INFORMS, citation for the 2003 Franz Edelman Award winner (https://www.informs.org/Recognize-Excellence/Franz-Edelman-Award).