Designing a resource demand forecasting system

Question

Overfilled and overloaded parts are becoming a problem for one of my customers. Purchase orders are submitted ~ every two weeks. Delivery takes 1-2 weeks. My client tracks inventory records using the .NET system I built.

I want to improve the inventory management system to constantly advise on inventory levels, predicting inventory demand and offering optimal supply orders. I read a lot of what I could find in forecasting algorithms, but I need help developing a good algorithm (or weighing a few algorithms) for an accurate forecast.

Substituted items carry a higher value than overloaded items for this customer, since most parts will eventually be used, but cash could be used for something else.

It would be nice to imagine the demand with an increase in risk, which relies on the initial implementation of Monte Carlo.

How do I approach this problem? What good / standard algorithms are used for this type of problem? Are there any good, free libraries to solve the problem of inventory demand?

Answer 1

This is a pretty well understood area. It also falls under the name "demand forecasting", "inventory management" or "replenishment."

A simple technique that works well is called exponential smoothing. He emerged from a study conducted in the 1940s for the Navy - looking at how to keep shipyards reserved for the construction of warships.

Charles C. Holt's original article, “Predicting Seasons and Trends Using Exponentially Weighted Moving Averages,” was republished in 2004 in the “ International Prediction Magazine ,” but I'm sure you can find other textbooks and articles — this will become standard technique.

There is stackexchange that includes this section: http://stats.stackexchange.com . Prediction and anti-aliasing are good tags to use. Rob Hindman gave many good answers and wrote at least one book on this subject: " Forecasting: Methods and Applications ."

Answer 2

Before you go and try to guess the future, you can present to your users something that will be just as useful. Can they currently get past orders with different time periods to get an idea of ​​the cyclical nature of your business?

It seems to me that this is not a problem that the system cannot predict future demand, but rather represents a problem that represents current and past data trends for the user. Even a printed chart or graph that has additional space in the "future", allowing them to ask themselves in the future with a pencil, can be a huge blessing.

In the end, I would be cautious about too much confidence in risk management algorithms, since these poorly understood algorithms (on the part of users) led to the death of more than one business.

I understand that this does not answer your question, but perhaps the answer is found elsewhere? If you want, I will delete this answer, so that others will consider this question still unanswered.