32 B ULK D ISTRIBUTOR
Asset Management
May/June 2019
CarLo ‘revŽůƵƟ onises’ transport planning
M
ovies such as Terminator, Matrix or I, Robot shaped our
idea of artifi cial intelligence (AI).
But what possibilities does AI really offer? And what is it anyway?
Who does the thinking and what does this have to do with transport
planning?
The average human brain has about 100 billion neurons that are
interconnected. Electrical impulses help transmit information between
those neurons. This enables humans to learn, to draw conclusions
and to think abstractly. As for artifi cial intelligence, artifi cial neurons
trained by algorithms are used. However, the goal is not to reproduce
human intelligence. Instead, machine learning should enable systems
to learn pattern recognition based on a large amount of data.
The idea behind machine learning is that, based on training data,
systems automatically learn specifi c models, such as sets of rules.
Thanks to machine learning, companies no longer have to create
models manually, which means that they do not have to spend time
on defi ning rules, checks and interpretations anymore. The quality of
the training data is crucial for machine learning to deliver the
expected results.
The machine learning process in Soloplan’s CarLo is as follows:
transport planning data, such as shipment modes, dates, start and
end points, loading items, loading weights and dangerous goods, is
There is no longer a fi xed standard procedure in CarLo, instead, the process is adapted to every
company’s individual requirements
fed into the system and processed by an algorithm.
The algorithm enables CarLo to ‘learn’ the dispatcher’s behaviour
and to create a model based on which future tours are planned
automatically in accordance with the rules learned. In other words,
there is no longer a fi xed standard procedure in CarLo. Instead, the
process is adapted to every company’s individual requirements.
Soloplan had to overcome unique challenges for which no standard
solutions existed. How can a machine learning model be tailored to
Tank Management
Software
+ORTQXGVJGGHƒEKGPE[CPFXKUKDKNKV[QH[QWTVCPMHNGGVVQTGFWEG
EQUVUCPFKPETGCUGRTQƒVU
1RGTCVKPI5QHVYCTG
www.soloplan.com
Asset Track & Trace • Quotations/Orders • Documentation •
Fleet Control/Planning • Purchase Invoices • Sales Invoices •
Dgl_lag_j-Npm‹r
Amlrpmj
• Maintenance & Repairs •
Customer & Vendor Tariffs • EDI • Agent Access •
Enquiries & Reports • KPI Dashboard Reporting
Complete logisƟ cs
intelligence
L
.GCUKPI5QHVYCTG
Asset Management & Budgeting • Work Orders • Billing •
Contract Management • On-Hire & Off-Hire •
Investor Reporting • Web Portal • EDI •
Maintenance & Repairs • Fleet Cost Management •
Equipment Sales • KPI Dashboard Reporting
Find out more about our
Tank Connect app
Visit us at Transport Logistic
Munich 4-7 June 2019
ITCO Tank Village, Hall B4,
Stand 217/318
RAM Intermodal
@RAMIntermodal
[email protected]
each customer’s individual needs? The large number of functions
available in the CarLo TMS makes the customers’ data records, which
are required for learning, very heterogeneous. That is a key challenge:
an approach that works well for one customer may not be just as
good for another customer. That problem is usually addressed by
machine learning engineers, who can adjust the model manually by
considering various statistics.
Two tasks are particularly challenging when it comes to the
development of a machine learning model. The fi rst is the so-called
feature selection, which is the process of selecting a subset of a data
record’s relevant features (eg, selecting the destination, weight,
transport type, etc, from the numerous features of past transport
orders).
The second is overfi tting/underfi tting, which often poses a problem
for machine learning engineers. From a mathematical point of view,
the model must be complex enough to learn human behaviour.
However, it should not just memorise that behaviour. The aspired
solution is referred to as a ‘generalisation model’ by machine learning
engineers.
However, since Soloplan supports more than 1,000 customers
worldwide, it is impossible to provide each customer with a
customised machine learning model. Therefore, the machine learning
algorithm must be able to perform all of the above-mentioned
manual tasks automatically without human intervention. That is why
Soloplan is developing a self-optimising pipeline, which can train a
machine learning model autonomously.
The latest version of the CarLo transport management system
comes with this newly developed programme, which will revolutionise
transport planning! The advantages of using machine learning for
transport planning are obvious: it will save dispatchers a lot of time,
help avoid mistakes and increase effi ciency considerably. Another
important advantage is that knowledge is no longer lost when there
are personnel changes. Since CarLo has learned the required
behaviour based on training data, a new dispatcher, for example, will
be able to plan tours in the same way as a long-term employee. All
data remains with the customer at all times. No data needs to be
transferred to Soloplan. Furthermore, the pipeline will adapt to
changing business requirements as the model is further trained with
new transport orders.
Machine learning simplifi es daily work in many ways. It provides the
information required to complete tasks – faster and more comfortably
than ever before.
Soloplan is at transport logistic (Booth 505/606 Hall A3).
www.ramintermodal.com
ogistics software provider lbase says it will present
trailblazing solutions for logistics in the digital era at
transport logistic. These are designed to enable logistics
service providers to boost their process effi ciency and
streamline their operations.
Two special highlights featured will be innovations for route
optimisation and for goods fl ow analysis using neural networks.
The new route optimisation feature facilitates automatic
optimisation of combined general cargo transport. The system
calculates the best possible routes for full and partial loads (FTL and
LTL). Automatic optimisation is also an option in hub-and-spoke
networks. Various mathematical methods are used to plan the best
possible route for complex combinations of scheduled services,
radial transhipment structures, and local transport.
Right from the fi rst time of use, potential savings of more than 10
percent of the freight costs are possible by optimising hub-and-
spoke structures alone, the company says.
Thanks to lbase, artifi cial neural networks can now be used to
forecast goods fl ows. To date, data regarding present goods fl ows
has primarily been used for such analysis. The new solution from
lbase now considers past patterns in order to determine more
accurately current and expected shipment levels. Historical shipment
data is recorded for this with attributes, such as product
classifi cation, season, day of the week, volumes, etc. Output
neurons then represent the expected deliveries per destination.
The forecasted goods fl ows can be visualised in different ways,
such as with heat maps or chord diagrams. “The neural network
keeps learning every time it receives new information, which makes
the forecasts more and more useful. Our new solution signifi cantly
increases operational shipment and transport planning effi ciency,
and the capacity required can be predicted with greater probability,”
explained lbase chief strategy offi cer Marcus Eiser.
www.lbase.software