You are Home   »   News   »   View Article

Simudyne - how simulation and modelling can help predict

Friday, July 21, 2017

Simudyne is helping oil and gas companies use simulation and network modelling, along with machine learning, to make better predictions and improve decision making.

UK company Simudyne is bringing together machine learning, simulation and network modelling to help oil and gas company executives make better decisions.

Together, these three elements provide 'incredibly powerful techniques for predictive analytics,' said Justin Lyon, founder and director of Simudyne.

'One of them is really powerful, but when you combine all three, you combine statistical techniques with techniques for understanding organised complexity.'

Decision makers in businesses today are trying to forecast, manage risk and achieve results, all at the same time. 'It requires brilliant decision making at pace and at scale,' he said.

To help them to do this, decision makers should have sophisticated models available to them about how the real world works, he says. These models should also run on real data.

Decision makers also need tools to provide them with better answers built very quickly. 'If it takes longer than three months, it's an R+D project, and no-one going to pay for it except R+D,' he said.

The models need to be sophisticated, because the real world is very complex. In the past, many decision makers, including banks, have made decisions based on standard linear regressions (an understanding of how one parameter will change if another parameter changes).

But these tools do not provide any understanding of the deeper cause and effect relationships, he said. As a result, they do not have a clear idea of what is driving what. 'When the crisis hits, they wonder what happened,' he said.

Mr Lyon sees executives in two types, the gut instinct driven and data driven.

'We don't work with the [gut instinct driven] people,' he said. 'We're interested in people who want to have a fact based data driven approach to making decisions. They [are happy for their] assumptions to be scrutinised and made explicit by computer models.'

One challenge with trying to build interest in these sorts of tools is that no-one wants to talk about their successes, in case other people copy them.

But, 'we need to talk about successes so people can know there's gold in the hills and get the investors to go and dig it out,' he said.

Also, no-one wants to talk about their failures either. 'But we need to, because only thorough our failures can we learn to make successes possible,' he said.

For building the simulation models, typically you have a multi-disciplinary team to build the solutions. This can include a number of data scientists, for example one specialising in machine learning, one specialising in computational simulation, one on network modelling.

Sometimes you can re-use some of the work, for example if you build a tool to help make bidding decisions that could be used in other scenarios, the same user interface but different data. 'We're also finding that some of the models are surprisingly also re-usable,' he said.

Mr Lyon is a former information and physical security contractor with the Bank of England, developing policies for information and physical security (2011-2015). He has also worked on a simulation for US Health Care (2012), and a project on retail banking technology (2012).

Simulating bidding in Mexico

Simudyne was involved in a project for a (name undislosed) oil and gas company which was bidding for oil and gas rights in Mexico.

The company wanted to use simulations to try to work out the best bid strategy, based on a prediction of the bid strategies other companies might follow.

Simudyne put together a large scale computational simulation. The simulation includes 'agent based simulation' (simulating how individual 'agents' might behave), as well as traditional data and calculation models.

It basically builds 'a virtual world of the whole landscape,' he said.

The simulation took into consideration the business environment and where the hydrocarbons were thought to be.

Simudyne acquired a range of data, and built a range of simulation models, including geotechnical issues, price outlook, returns expectations, capital expenditure.

By using the tool, the project team 'predicted the formation of a consortium which they [otherwise] thought was never going to happen,' he said. 'In the real world it did happen.'

Simudyne wanted to provide the company decision makers with tools to try out different decisions well in advance of the deadline, he said.

The decision makers can take the simulated results, analyse them and use them to adjust decision making, before they make decisions in the real world, he said.

They could adjust the parameters, and see what might happen as a result of various bids they could make. They could look at the perspective from different companies.

Every time they suggest a bid, the software would run millions of different scenarios around it.

The simulation could be run to look at what might happen on individual blocks or companies. Some of the parameters was preset, and some were adjustable.

Altogether the company could put together a chart of what each company might be likely to bid.

The project needed to be put together within a tight deadline, because the license blocks were going to be awarded in four months time.

Executive communication

The 'visualisation' part of the software, building the thin client 'apps' which company decision makers would actually work with, should be done separately to developing the actual simulation, he said.

Left to themselves, 'the data scientists produce something that's elegant for data scientists and not understandable for the executives,' he said.

It is very important that the decision makers can play around with the simulation easily, he said.

'We had brilliant mathematicians, but if their insights aren't communicated in a fashion that executives can get their heads around, the insights are missed and lost.'

'The idea of giving this to a decision maker on a Surface Pro or iPad and letting them explore with parameters and fail safely [is important],' he said. 'They can run an infinite log of scenarios themselves.'

The skill of building the visualisation layer is quite similar to traditional website design, he said. You need to have someone who will have conversations with the decision makers who will use the software, about what they want to work with.

The software needs to be able to support multiple users signing in simultaneously, with different teams make decisions about different parts of the system, eh said.

If you want to optimise a complex chain, there will be many different decisions, and a decision in one place will have an impact somewhere else.


A challenge with developing simulations is that companies often keep their data in different data silos, and a lot of the value is only generated when they can be brought together, he said.

Studies show that 99 per cent of the data generated by offshore oil platforms is never used.

'McKinsey is doing lots of reports saying, as you combine the data, that's when you generate the real value,' he said. ''If you don't bring the data together, you leave so much money on the table.'

Other business applications

One audience member asked whether the software could have predicted the outcome of the US presidential election.

Mr Lyon replied that it might be possible if you could build a large scale model of the entire US population, and gather as much information as you could about them, including from their Facebook profiles (although the privacy issues could be an obstacle).

The software could handle 300m independent agents, he said. You would create a model of how each person is likely to behave.

So technically, it could be done, legally maybe it would not be so easy, he said.

Simudyne is also considering using the software for modelling London house prices, trying to predict when prices will burst and the reasons that might cause it.

Associated Companies
» Simudyne
comments powered by Disqus


To attend our free events, receive our newsletter, and receive the free colour Digital Energy Journal.


Latest Edition June July 2018
Jun 2018

Download latest and back issues


Learn more about supporting Digital Energy Journal