smart-meter

Domestic energy experiments involving EDF and smart meters. Note that in order to use these scripts, in addition to having an EDF account, you will need to have a valid OpenAI API key set in an OPEN_API_KEY environment variable, a valid Anthropic key in ANTHROPIC_API_KEY and a valid Groq key in GROQ_API_KEY for full LLM capabilities. You will also need to know your MPAN number and IHD number and you will need to set MPAN and IHD_MAC environment variables for them. You will also need to set environment variables for HOUSE_NUMBER and POSTCODE.

• smart-meter-playbook - a Jupyter notebook walking through how to access your smart meter data from the DCC using your supply point MPAN number and your IHD number.

• edf-bill-playbook - a Jupyter notebook walking through how to process the energy data you can download from the EDF customer portal per the screenshot below. The csv which has rows corresponding to months and five columns as follows:

Timestamp,Electricity consumption (kWh),Electricity cost (£),Gas consumption (kWh),Gas cost (£)
08/2024,38.788,27.39,135.1189,19.95

• energy-advisor.py - a command line agent entirely developed using ChatGPT Pro and Cursor which is able to autogenerate an HTML insights report plus graph from the input csv. Various models are supported from OpenAI (gpt-4 and gpt-3.5-turbo), Groq (mixtral-8x7b-32768 and llama2-70b-4096), Claude (claude-3-5-sonnet-20241022) and Ollama (llama3.2). The command line help looks like this:

Usage:
    energy-advisor.py [-v] [-m MODEL] [-i CONTEXT] <csv_file>
    energy-advisor.py (-h | --help)
    energy-advisor.py (-V | --version)

Options:
    -h --help           Show this help message
    -v --verbose        Enable verbose logging output
    -V --version        Show version and author information
    -m --model MODEL    Model to use for analysis [default: gpt-4]
                        - OpenAI: 'gpt-4', 'gpt-3.5-turbo'
                        - Groq: 'mixtral', 'llama2'
                        - Claude: 'claude'
                        - Ollama: 'llama3.2'
    -i --input CONTEXT  Path to text file containing user context for personalization

Arguments:
    csv_file            Path to CSV file containing energy data

Of these options, Groq mixtral is the fastest and Ollama llama3.2 is the least expensive in terms of carbon cost. Here is an example invoking the Groq mixtral-8x7b-32768 LLM using freeform homeowner input context supplied in the context.txt and bill data held in combined-consumption-2024-01-01-2024-12-31.csv file:

$ python energy-advisor.py -m mixtral combined-consumption-2024-01-01-2024-12-31.csv -i context.txt

Generated report from that command line looks like this:

• smart-meter-advisor.py - a command line agent entirely developed using ChatGPT Pro and Cursor which is able to autogenerate an HTML insights report of smart meter data using MPAN and IHD per the recipe outlined above in the smart-meter-playbook. The same models are supported as in energy-advisor.py. There are three commands:
  • dump-meter: this dumps information about your smart meter make and model.
  • get-smart-meter-data: this retrieves all your smart meter electricity data both daily and monthly in two separate csvs.
  • generate-report: this generates the HTML insights, usage graph and recommendations report.

The command line help looks like this:

Usage:
    smart-meter-advisor.py [-v] [-m MODEL] [-i CONTEXT] --command CMD
    smart-meter-advisor.py (-h | --help)
    smart-meter-advisor.py (-V | --version)

Commands:
    --command CMD       Command to execute
                        - dump-meter
                        - get-smart-meter-data
                        - generate-report

Options:
    -h --help           Show this help message
    -v --verbose        Enable verbose logging output
    -V --version        Show version and author information
    -m --model MODEL    Model to use for analysis [default: gpt-4]
                        - OpenAI: 'gpt-4', 'gpt-3.5-turbo'
                        - Groq: 'mixtral', 'llama2'
                        - Claude: 'claude'
                        - Ollama: 'llama3.2'
    -i --input CONTEXT  Path to text file containing user context for personalization