Today in a modern society the pace is high and the need for flexibility and mobilityis big. Despite the increasing access and great digital solutions for online-meetings,the need for short distance transportation remains. With an increasing number ofeasy-to-use mobile applications, the many offered solutions for personal transporta-tion have in the last few years expanded. Besides the effect of an increasing pressureof road space, the pressure on the transportation solutions has escalated. This is the current situation of the company we came in contact with. Their currenttravel service provides on-demand transportation in small electric, emission free andchauffeur driven vehicles they call pods. As a step of claiming their space in thetransportation sector they are developing their application letting customers planand book a transport. For the ability to test their new booking solution this project took form. With asimulation it would be possible to test the various probabilities of a travel outcome.The aim was to develop a simulation software, simulating the events of multipletravels taking place in the system. With the simulation it would be possible to eval-uate the configuration for the booking feature, analyse how it works with the rest ofthe system and to help the company to predict how the intensity of customers in re-lations to the number of available drivers would affect the outcome of a travel request. The project covered the study of how to interpret the factors, that a transportationdirectly depends on, into events in the system and how to use historical events tocreate probabilities of the simulation outcome. The simulator software was not suc-cessful with simulating multiple rides as intended, but parts of the software could beevaluated in relation to historical events. Statistical models were built using aggre-gated events from the system. When comparing the average result of the statisticalmodels with the historical event count of the system, it resulted in an outcomewithin an acceptable range. This shows that it was possible to use aggregated his-torical events to create probabilities and that these probabilities were reliable.

BackgroundLKAB has struggled for years with measuring the amount of water that fines are containing when it leaves the filters. This is desirable because if the watercontent could be measured it would be easier to keep it stable on a suitable level and the waste in the process could be reduced. MethodThis paper explores the possibility to through predictions based on surrounding signals predict the water content of the fines with desirable precision. These signals are processed with statstical methods and models which will be compared. The models used are Linear-, LASSO, Random Forest- and Additive Models. ResultThe investigations suggests that interaction terms between predictors are not significant and neither are non-linear interactions between predictors and the response. The more complex models which includes more predictors does notperform better than the simpler ones. ConclusionThe resulting best model is a Simple Linear Model which includes the predictor Negative Pressure. This model has a measuring error which is estimated to be around +- 0.5 percentage points which can be compared to a water content percentage of about 7.5-10.5%. There is a LASSO-Model that generates a smaller estimated measuring error but the difference is not considered to be large enough to outweigh the increased complexity of the model. / BakgrundAtt mäta fukthalt i slig i direkt anslutning till de filter som torkar upp slig från slurry på ett till förlitligt sätt har i många år varit en utmaning för LKAB. Att kunna göra detta är önskvärt för att i förlängningen hålla fukten på en stabil nivå och därmed effektivisera pelletiseringsprocessen. MetodDenna uppsats undersöker möjligheten att genom prediktion baserat på kringliggande signaler prediktera fukthalten i sliget med önskvärd precision. Dessa signaler hanteras med hjälp av statistiska metoder och modeller som jämförs med varandra. De modeller som kommer att användas är Linjära-,LASSO-, Random Forest- samt Additiva Modeller. ResultatUndersökningen visar på att samspel mellan prediktorer ej är signifikanta och inte heller icke-linjära samspel mellan prediktorer och fukthalt. Vidare visar undersökningen att de mer komplexa modellerna som tar in fler prediktorer inte presterar bättre än de enklare. SlutsatsDen resulterande bästa modellen är en Enkel Linjär Modell som bygger på prediktorn Undertryck och har ett mätfel som i snitt över filtren skattas till +-0.5 procentenheter, att jämföra med en fuktalt på 7.5-10.5%. En LASSO-modell genererar ett mindre skattat mätfel men det anses ej vara nog mycket mindreför att väga upp för dess mer komplexa uppbyggnad.