On the 20th and 21st of September 2023, R&D Nester attended the 5^th I-NERGY plenary meeting, held in Riga, Latvia. The assembly, which was the first physical plenary meeting of the European project, aimed to presents the main results of the demonstration pilots and of the other projects' components before the end of the project, foreseen in December 2023.

I-NERGY is an EU funded project aiming to support, develop and demonstrate innovative AI-as-a-Service (AIaaS) Energy Analytics Applications and Digital Twins services validated along 9 pilots across 8 countries. The services focus on three different sectors: Energy Commodity Networks (in which R&D Nester is involved), Distributed Energy Resources, and Energy Efficiency. The large geographical coverage of the demo sites, illustrated in the figure below, aims to support the EU-wide replicability and market take-up of AI-driven solutions in different socio-economical contexts to maximize the impact of I-NERGY services across Europe. I-NERGY pilot's approach permits to comprehensively test the analytics devised to cover the initially detected interests of relevant Electrical Power & Energy Systems (EPES) stakeholders within the energy value chain, covering their whole energy market: from the operation and maintenance to the society, as well as cross-cutting interests, such as policy making and research.

Under the scope of the project, R&D Nester has implemented a demonstration pilot in Portugal, focusing on the two following Use Cases (UCs):

• UC1:  AI for enhanced network assets predictive maintenance, integrating off-grid data with condition-based Monitoring;
• UC2: AI for network loads and demand forecasting towards efficient operational planning.

UC1

AI for enhanced network assets predictive maintenance, integrating off-grid condition-bases Monitoring

On top of presenting the current progress and results for both use cases, the meeting in Riga was an opportunity to showcase a live demo of one of the services developed for UC1, namely the Circuit Breaker Asset Management. Nicolò Italiano, an energy systems research engineer from R&D Nester, pitched a step-by-step demonstration on the functioning of the service, which is currently available as web-based application on the AIoD Catalogue, at the following LINK.

The service, which can be tested after requesting the credentials to R&D Nester, features a fully automated fault data processing, analysis, reporting system, and circuit breakers' Remaining Useful Life (RUL) estimation, for events occurring in any electrical system. Specifically, it includes the following submodules, depicted in the workflow below:

• Fault detection, which leverages signal processing techniques to detect whether an event is a fault or not;
• Fault classification, identifying the type of fault occurred (e.g. single-phase-to-ground, three-phase-to-ground), using Machine Learning (ML);
• Incident reporting, which extracts incident statistics that are key to assets management divisions of system operators;
• RUL prediction, predicting the Remaining Useful Life of a given circuit breaker, leveraging failure probability functions fitted on historical dataset of the events occurred, and identifying the corresponding maintenance recommendation.

Moreover, the service proposes an automatic and prompt visualization of the events processed, plotting all current's and voltage's signals segmented into the different phases characterising the faults, as per the graph below. Finally, the web-based application includes an asset management database, in which all processed faults are stored and grouped by assets. The database leverages a traffic light label system for fast and easy-to-read visualization of assets' status, with the colour depending on the predicted failure probability.

 UC2

AI for network loads and demand forecasting towards efficient operational planning

The second use case is considering the forecasting of the Portuguese national load and of the net load of a specific TSO/DSO interface for the day-ahead operational planning. The most recent results presented in Riga concerns the Net Load Forecasting Model, applied to a specific case in which the net load is subject to a substantial change in the time-series pattern, occurring within a short timeframe. This is commonly referred as concept drift, and is illustrated below:

Concept drift observed in the net load (y-axis), during a given time-period (x-axis)

Such events may occur for different reasons, among which, for instance, the installation of a new Photovoltaic (PV) power plant at the distribution level, self-consumption or Electric Vehicles (EV) chargers, in large-scale, or the connection of a new large-scale consumer. The first case could even lead to reverse power flows, represented by the negative values in the figure, which may be quite challenging to predict by a model that has always been trained exclusively upon positive values. 

For this reason, it is critical to develop and use models that, in addition to feature high forecast accuracy, are also able to adapt to new patterns and changes in the data. Under the scope of I-NERGY, the three approaches listed below have been considered:

• Long Short-Term Memory neural network-based model (LSTM); 
• Ensemble models (XGB, DT, Lasso, SVR) based on Iterative Learning;
• Adaptive Random Forest based on incremental learning (Adaptive RF).

The three models were at first trained upon a large dataset that included significant drifts (Scenario 1), and then trained on a smaller dataset prior to the occurrence of the drifts (Scenario 2), with the aim to study the difference of the performance between the two cases. The second scenario was designed to analyse how the models react to new patterns and changes in the data that are not experienced during the training. The scenarios considered the same test dataset, which included one year that was not used for training the models.

The results showed that the Adaptive RF is the model that better adapts to the concept drifts, maintaining Normalised Root Mean Squared Error (NRMSE) of 8.6%, for both scenarios. The NRMSE corresponds to the ratio of the root mean squared error to the maximum of the absolute values within the test dataset. On the other hand, the ensemble model presents the worst performance, with an NRMSE of 9.1% in Scenario 1 and 14.1% in Scenario 2. The performance's drop from the first to the second scenario of the ensemble model indicates that it does not adapt well to new patterns unseen in the training dataset, contrary to the Adaptive RF. These results, described in detail in the scientific paper "Short term net load forecasting using computational intelligence techniques" presented at the 7th E-Mobility Power System Integration Symposium, in Copenhagen, highlights the importance of using adaptive forecasting techniques when facing net load time-series characterised by several concept drifts.

The figures below depicts the forecasting results of the three different models compared with the actual net load, for Scenario 1 and 2, respectively, and for a given timespan. Once again, it is possible to observe the Adaptive RF (in blue) is the model that better follows the actual net load patterns, especially for the second scenario.

Forecasting Results of Scenario 1

Forecasting Results of Scenario 2

For a full discussion on the results, please refer to "Laouali I., et al, SHORT TERM NET LOAD FORECASTING USING COMPUTATIONAL INTELLIGENCE TECHNIQUES, 7th E-Mobility Power System Integration Symposium, 2023", in which the dataset is described in detail, as well as the techniques used to prevent overfitting and the additional evaluation metrics considered.

For more information:

I-NERGY Project @ R&D Nester website

I-NERGY Project website