CNROpt Charging Stations Forecasting
AI Model based on LSTM forecasting the Utilization and Energy Consumption of Charging Stations. We forecast the future activity/consumption/utilization/Customer-experience on EV Charging stations using LSTM timeseries forecasting yielding highly accurate predictions for period ranging from a few days to one year. These forecasts are used on an interactive GeoAnalytics powered planning tool that assists the technical and business users to select the best candidate locations for building the Network Expansions, ensuring optimal exploitation of the available CAPEX investment maximizing RoI and Customer Experience.
This model was trained on timeseries data from 3 year charging session raw data from Croatian Charging Station Networks from Croatia, funded by the i-nergy project by cascade funding in the 2nd Open Call. There have been extensive tests for many alternative timeseries forecasting models and configurations, such as ARIMA, Trasformers, etc. and this particular configuration of LSTM model that was trained on the historical data of 257 charging stations of Croatia has given us the best accuracy. We are confident that it can be used also for adhoc forecasting of any Charging Stations provided that it is fed with adequate hourly historical data (2 years minimum recommended history length)
CNROpt Project
An AI based Digital Platform for Intelligent Planning of Electric Vehicle Charging Network expansion.
Country: Cyprus, Greece, Croatia
Areas of Experimentation: Timeseries forecasting using both off-the-shelf and custom models, Geospatial models for location-based optimization.
Can you describe your project in a few words?
CNROpt is an ambitious project to develop an innovative tool for power companies and the operators of Charging Station Networks for Electric Vehicles. It will help facilitate the efficient rollout of new Charging Station Locations and the expansion of the existing ones, by suggesting the optimal sites for expansion, taking into account both business and customer experience requirements.
Powered by AI, we develop new AI Algorithms based on the recent research in the field of AI and trained the models with a rich set of time series data. These data are results of anonymous charging session raw data of several hundred of thousand sessions from more than 250 locations in Croatia.
Who will help implement the AI solution?
SLOA Local AI is a technology startup in Kalamata, Greece founded 2022 for AI R&D, branded as Local AI, developing innovative solutions for sustainability utilizing the power of AI for local and global stakeholders, such as Municipalities, Regional Government, Private Energy Stakeholders, etc. AI innovation is at the core of its business strategy, and it actively participates in research and innovation projects to jump-start the development of new functionalities and new commercial solutions. We envision a future where AI is ubiquitous and assisting local, regional, and global green transition initiatives. We work together with the University of Zagreb, with their deep domain knowledge of the EV charging network technologies to provide a disruptive way to develop Charging Networks in Europe.
What is the AI solution the project plans to implement?
For the Timeseries Forecasting models, we plan to try 3 algorithms:
- ARIMAoff-the-self algorithm, a widely used procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects. It works best with time series that have strong seasonal effects and several seasons of historical data. ARIMA is robust to missing data and shifts in the trend, and typically handles outliers well. The ARIMA procedure includes many possibilities for users to tweak and adjust forecasts. We will use human-interpretable parameters to improve our forecast by adding your domain knowledge.
- Transformer Architecture for Timeseries. Transformers constitute a class of Deep Learning Models that are suited to handle temporal dependencies on data. They are the backbone of modern Language Models (GPT. We will develop our custom Transformer-based prediction model to perform the congestion-utilization prediction.
- LSTM. Long short-term memory (LSTM) is an artificial neural network used in the fields of artificial intelligence and deep learning. Unlike standard feedforward neural networks, LSTM has feedback connections. Such a recurrent neural network (RNN) can process not only single data points (such as images), but also entire sequences of data (such as speech or video). This characteristic makes LSTM networks ideal for processing and predicting data.
For the Candidate Stations Scoring – Selection modelling, we will approach this problem by trying to geographically cluster the existing Charging Stations with the K-means Clustering in Geographical Dimensions based on various KPIs (RoI, Utilization, Congestion, waiting times, etc.) as they are produced by the previous step, the Timeseries Forecasting. This enables our tool to detect the geographical areas indicating hot spots for the Charging Station Network, prioritizing the relevant candidate sites for expansion.
Mitigation Actions: The tool that enables the interaction with the end user will provide comprehensive utilisation analytics both for the current configuration and the proposed new station additions. This way we have sanity checks in the tool showing the specific split of the forecasted charging traffic on the old and the new stations. The end user gets a clear picture of the target performance ensuring the soundness of results.
REQUIREMENT #2 Technical Robustness and Safety: The solution based on AI need to be resilient and secure
Mitigation Actions: The overall solution will be cloud native, thus ensuring high availability. Moreover, the data repository is the PostGIS database platform, which has a comprehensive security and access authorization governance implemented.
REQUIREMENT #3 Accountability: What if something goes wrong in the algorithm results?
Mitigation Actions: Our team members, designers, developers, and deployers comply with standards and legislation to ensure the proper functioning of AIs during their lifecycle. The tool that enables the interaction with the end user provides comprehensive utilisation analytics both for the current configuration and the proposed new station additions. This way we have sanity checks in the tool showing the specific split of the forecasted charging traffic on the old and the new stations. The end user gets a clear picture of the target performance ensuring the soundness of results.
Based on the information provided by the beneficiary, the project has potential ethical issues related to Personal data (protection). The potential ethical issues are related to the optimization of the charging network expansion for electric vehicles based on AI. The beneficiary should clarify the process of how this data will be anonymized and by whom.
• Description of the anonymization/pseudonymization techniques that will be implemented must be submitted as a deliverable.
• The beneficiary must explain how all the data they intend to process is relevant and limited to the purposes of the research project (in accordance with the ‘data minimization ‘principle). This must be submitted as a deliverable.
Proposed actions / mitigation:
The data that are uploaded on AIOD will be aggregated, in a cumulative form and not the sessions themselves.
We confirm that all research and innovation activities funded by the H2020 (I-NERGY) adheres to the EC Ethics By Design and Ethics of Use Approaches for Artificial Intelligence (https://ec.europa.eu/info/funding tenders/opportunities/docs/2021- 2027/horizon/guidance/ethics-by-design-and-ethics-of-use approaches-for-artificial-intelligence_he_en.pdf ) and Ethics Guidelines for Trustworthy Artificial Intelligence (AI) available at https://digital strategy.ec.europa.eu/en/library/ethics-guidelines trustworthy-ai
More specifically, we have implemented complete Anonymisation in the source data, which technically means that all the columns that had any identifiers of the end users have been removed from the raw data. The charging sessions have only the charging station identified and the time – no activity of the user remains there. Moreover, since the end user id has been removed and not pseudonymised, there is no indirect way to detect end user patterns that could potentially reveal indirectly a person’s identity.
Our target is fully served by the anonymised data since the individual patterns are of no concern to us. The remaining information is minimised to fit our needs including only the EV charging locations and the time dimensions. Of course, these dimensions are essential to perform spatiotemporal predictions and new Station locations optimisation.
GDPR Compliance Declaration report will be included as part of the MVP technical deliverable.