In this lab, future power consumption will be predicted based on time series data uploaded to S3 in form of csv file.

Services Covered

  • Forecast Forecast

Lab description

In this lab, future power consumption will be predicted based on time series data uploaded to S3 in form of csv file.

Learning Objectives

Lab date

19-12-2021

Prerequisites

  • AWS account

Lab steps

  1. Upload the electricityusagedata.csv file to S3 bucket.
  2. Navigate to Amazon Forecast dashboard and create new dataset group. Give it a name and choose Custom forecasting domain. On the next tab specify the frequency of data to 1 hour and for data schema use the following JSON: 
    {
   "Attributes": [
       {
           "AttributeName": "timestamp",
           "AttributeType": "timestamp"
       },
       {
           "AttributeName": "target_value",
           "AttributeType": "float"
       },
       {
           "AttributeName": "item_id",
           "AttributeType": "string"
       }
   ]
}

    For the data location input the S2 buckets URI.

  3. After the import completes start the predictor training.

    Provide Forecast horizon: Enter 36 and Forecast frequency: Select hour from the drop-down menu. Turn off AutoPredictor and choose ETS algorithm. Training will take some time to complete

    It took 18 minutes to finish.

  4. Back in the dashboard click Start to generate forecast. Use the predictor created in previous step. This will also take a while (9 minutes in my case).
  5. When forecast finally get created you can go and query it. Specify start date as 2014/12/31 and end date: 2015/01/02 and key to client_1.

    Result explanation: The actual energy usage is shown in grey on the left and has no quantiles shown because there is no uncertainty. The forecast predictions are shown by three lines. The numbers P10, P50, and P90 correspond to the 10%, 50%, and 90% quantiles, respectively. The actual value has an 80% chance of being between the P90 and P10 range (90 – 10 = 80). The larger the number of PXX, the higher the probability that the value will fall within that range. The value of P50 line is the midpoint of the range.

Lab files

Acknowledgements

By : LU Tags: