Freelancer : freelanmohan7
Partager :
Signaler la proposition

Electrical load forecasting using fuzzy logic

This is an example forecasting problem, which I solved previously for hourly electrical load prediction carried out for one week. The forecasting algorithm uses a fuzzy logic model, whose parameters have been trained through optimization. It considers the hour, week day and previous loads as inputs and predicts the load for the next few hours. The figure attached shows the weekly prediction that has been made by the fuzzy network, trained using one months, data. If this prediction accuracy is satisfactory, we can discuss to modify/improve it to suit to your case.


                                                                                                                        Proposition n°                                            30
                                         du concours                                             Forecast hourly energy consumption for a certain period of time
Proposition n°30

Tableau de clarification publique

  • freelanmohan7
    freelanmohan7
    • il y a 7 ans

    If you want for 3 months, I can do it by considering all the previous data. That is fine. I would like to know how much is the accuracy you are expecting? What is the acceptable mean percentage error for your project? Depending on your accuracy requirements, I am considering to test other models like neural networks and neuro-fuzzy networks.

    • il y a 7 ans
  • scringhelteamsrl
    Titulaire du concours
    • il y a 7 ans

    By 01 - 03.2014 I meant January - March 2014 full month predictions

    • il y a 7 ans
  • freelanmohan7
    freelanmohan7
    • il y a 7 ans

    I can carryout prediction using your data. How much data should I use for training? If I use all the data at the preliminary testing level, it may be too much. can I use previous one month data for training?

    • il y a 7 ans
    1. scringhelteamsrl
      Titulaire du concours
      • il y a 7 ans

      some have only one month, and even thought 1 month is not relevant it should be used. If a consumption point has more than 1 month of data, using just 1 month it will not be relevant.

      • il y a 7 ans
  • freelanmohan7
    freelanmohan7
    • il y a 7 ans

    I could not upload the images here. This dropbox link has the forecast for 01-03-2014. It is obtained by training the network using previous one month's data. The accuracy can be improved further by training it for longer times. https://www.dropbox.com/s/4qw530zqm7oo9w9/forecast_01_03_2014.jpg?dl=0

    • il y a 7 ans
  • freelanmohan7
    freelanmohan7
    • il y a 7 ans

    Hi, I don't see an option to send new results here. Could you please initiate a chat so that I can send you the forecast for 1-3-2014?

    • il y a 7 ans
  • freelanmohan7
    freelanmohan7
    • il y a 7 ans

    hi, the training has been carried out in matlab using a hybrid optimization technique. I am not clear about your second question. I think the answer for this is desktop side. I have received the data in a txt file, which I used for training.

    • il y a 7 ans
  • scringhelteamsrl
    Titulaire du concours
    • il y a 7 ans

    can you provide a forecast for 01 - 03.2014, in a excel file in the same format as the provided consumption history, so we can compare with the actual

    • il y a 7 ans
  • scringhelteamsrl
    Titulaire du concours
    • il y a 7 ans

    3. Can you provide a forecast based on the date we provided, using your solution?

    • il y a 7 ans
  • scringhelteamsrl
    Titulaire du concours
    • il y a 7 ans

    hi, thx for your entry. A few questions: 1. how do you make the training? 2. Does your solution is DataBase side or Desktop side? (I ask this because it will be a great advantage if it was Database side because we have another similar project where we could use the solution) - you could use CLR (code in C#/Vb and import as CLR in database)

    • il y a 7 ans

Montrer plus de commentaires Traitement en cours...