How to know whether or not to produce an oil field, and how to know how? The oil industry is used to make a lot of simulations to satisfy the SEC rules, and to know how much oil they can drill today and in the future. This book goes further than just the usual how much from a field under specific condition, by adding the development plan as well as the oil price in the equation.

Content and opinions

The book seems to be based on things Petrobras, the Brazilian national oil company, does, so it’s fair to assume they actually do far more than what is presented in the book.

There are four chapters that make the core of the book, exposing the different algorithms, explaining bits of them (you will still need to dig further to understand what they actually are, and obviously even more to duplicate the results), but also giving lots of examples.

Let’s start with the first two chapters, they are not in the “core” of the book. The first one explains why you need to do uncertainties in the oil world. Main reason: why would you spend billions of dollars when you are not sure you will get them back? The second chapters explains some financial theory. Usual uncertainty analysis (the one for the SEC for instance) don’t involve financial studies. They just require you to give your P1/proven oil and gas reserves (10% chance that you get this oil out of the field). Then for your management, you want to get the P2 (50%). Here, you go one step further, because these 2 values don’t show how much money you will get. So the financial framework is important to understand how you can get to the final step of the simulation.

The third chapter deals with the Machine Learning algorithms that will be used through the book. Several evolutionary algorithms (genetic algorithms and genetic programming) are now mainstream, but some of them are not (cultural algorithms or quantum-inspired GA). It’s always interesting to see new algorithms, especially as they are applied to a concrete problem in one of the biggest industries. Then the authors tackle neural networks, with a simple approach (compared to my previous review), fuzzy logic (I hadn’t heard about it for quite some time) and the mix between the two of them, neuro-fuzzy models.

The fourth chapter is about the exploration process and mainly uses evolutionary algorithms to optimize the development plan, that is the number of wells and their type to optimize the number of oil barrels at the end of the life of the field. It is interesting to see that they don’t simulate an actual complete development plan (when are the wells drilled and then started). Injection of prior knowledge is addressed, and it isn’t always making the result better, interesting to know!

Fifth chapter starts introducing the market behavior by adding the option to activate new parts of the development plan, depending on several parameters. Here, the authors use genetic algorithms and fuzzy models, adding also the simulation of the oil price during the life of the field, ending up with Monte Carlo methods to define the “option” price. I have to say that I didn’t understand everything in their examples, but I understood the global ideas, without knowing much of the financial side. The explanation of how the fuzzy numbers and statistics differ in the end is lacking, although they are quite close. It’s a different approach, with a different theory anyway.

The sixth chapter and the last one in the “core” takes everything together and tries to make it even further. It introduced additional mathematical tools but wasn’t as thorough as the last three. It was still interesting, but obviously taking advantage of this one requires mastering the other ones before.

The last chapter present the tools used in production and the architecture decisions behind. It has less value as you can’t get the tool…

Conclusion

I picked up the book because I’m currently in the oil industry, and I have interests in intelligent systems. Even though I can’t use the tools now, it gives some good leads to build such a system, and hopefully I may work on such systems some day (OK, I’ve already designed the core framework around the reservoir simulator, before I even read the book).