Short-term Electricity Consumption and Demand Prediction (AI4Demand)
The AI4Demand project presents a short-term (daily or hourly) energy demand and consumption forecasting module using novel AI-based approaches where the cause-and-effect relations between energy data and externally influencing factors (energy and weather-related indicators) are considered and analyzed.
The architecture of the proposed demand forecasting module follows the hybrid-type methodology integrating different components or layers based on data mining, machine learning, and signal processing techniques.
The whole project is implemented in Python programming language using different data science and deep learning libraries.
The developed prediction module can be used for exploratory data analysis and forecasting purposes.
The provided codes are fully configurable and integrable.
The prediction module seeks to promote the following services and activities:
- Optimize the work of the building energy management systems (BEMS).
- Enable energy suppliers to improve the production procedure and anticipate peak demands.
- Allow end-users to have a better understanding of the insights of their consumption habits and scenarios.
- Help consumers to reduce the energy consumption and lower energy bills.
- Aim supplier companies to manage energy waste which leads in reducing the carbon footprint and financial loses.
The developed AI4Demand energy forecasting service is a Jupyter notebook code which integrates 3 main components:
-
The Data layer is used to collect and store real-time data through the API services and IoT sensors. The collected data types include electricity consumption data and HVAC data derived from various IoT devices and sensors, as well as meteorological data obtained from weather stations. For the data layer to collect an energy datasets, we provide a small separate code.
-
The Analysis layer processes input data with different data analytics tools for data normalization, noise reduction, and pattern extraction purposes. Raw datasets are standardized and normalized into one scale and then different learning and signal processing techniques are applied for the statistical analysis of the input datasets.
-
The Prediction layer applies AI-based learning algorithms to train the relevant data and generate forecasts for future energy consumption and demand.
As a first stage, all necessary datasets are collected and then converted to the .csv standardized format.
Raw datasets are checked for missing values and normalized for the data pre-processing procedures.
Next, a feature selection algorithm based on the ensemble learning method called Random Forest is implemented to filter out irrelevant and redundant features and keep only important ones for the learning algorithms.
Further, the signal processing technique called Discrete Wavelet Transform is used for cleaning the selected features from the existing noise.
The prediction stage is performed using a neural network architecture based on long short-term memory (LSTM) combined with Auto-Encoder networks.
The evaluation procedure is done with the help of two precision metrics, namely, a mean square error (MSE) and the prediction of change in direction (POCID)
The dataset part includes two .csv files at hourly and daily resolution.
-
The first dataset contains the hourly electricity and heating consumption values from the office building located in Malmi area together with 9 meteorological features obtained from the nearest meteorological station called Helsinki Malmi lentokenttä. The weather-related variables include cloud amount, pressure, relative humidity, air temperature, dew-point temperature, horizontal visibility, wind direction, gust speed, and wind speed. The observation period is 01/01/2020 – 17/03/2020.
-
The second dataset contains daily time series data from the period of 01/01/2019 – 31/12/2021. The data variables include the daily electricity and heating consumption values from the school building located in Kaisaniemi area, as well as 5 meteorological variables (precipitation amount, snow depth, air temperature, maximum temperature, minimum temperature) collected from the nearest weather observation station called Helsinki Kaisaniemis.