Currently, systems with PV panels and heat pumps (HPs) of various types are the most dynamically developing sector in environmentally friendly technologies. In particular, PV energy is gaining interest internationally and in the EU as a provider of low-cost, energy-efficient and clean energy. In 2021, 18.7% of the world’s total PV capacity, which corresponds to 158 GW, was installed in EU. Germany leads the way with PV capacity installed of 58.5 GW, second is Italy with 22.7 MW and third is France with 14.7 GW. The expansion of the PV installations market is particularly visible in Poland, which ranked 10th in the world regarding investment made in new PV power capacity installed in 2021. It is expected that there will be further growth of PV capacity in Poland, reaching approximately 7.3 GW in 2030 and 16 GW in 2040.

Over the past few years, HPs technology has also made great inroads into the EU market by increasing energy efficiency, reducing greenhouse gas emissions and promoting RES. HPs are now particularly seen as devices that can significantly reduce energy consumption in buildings for both heating and cooling purposes. According to the European Heat Pump Association, the HPs market in Europe has been growing steadily over the last decade and in 2021 have exceeded 34%, surpassing two million units sold per year for the first time [8]. This big growth has been driven by several factors, including increasing energy prices, energy efficiency regulations and the push towards decarbonization. ASHPs have the largest share in the HPs market in 2021, which is equal to 94% with the remaining 6% being ground- or water-based. ASHPs are considered less expensive compared to other existing HP-based technologies.

There is a need to develop solutions that will help maximize the use of energy generated from RES and reduce or even limit existing development barriers. Such issues include the problem of insufficient use of generated electricity in PV on-grid microinstallations (

To improve interval-based solar energy forecasting for residential self-consumption, it is essential to combine high-resolution local weather data (irradiance, temperature, cloud cover) with real-time measurements of photovoltaic production and household energy demand. The integration of smart sensors, sky cameras, and satellite data helps anticipate rapid variations caused by cloud movements, which is especially valuable for short-term forecasts (5 to 30 minutes). These data streams feed hybrid models that combine physical approaches (e.g., PV performance models) with machine learning techniques such as recurrent neural networks (LSTM, GRU) or decision tree ensembles (e.g., XGBoost), enhancing both accuracy and robustness particularly when applied within hybrid renewable energy systems.

Moreover, accounting for uncertainty is crucial for optimal residential energy management. Using quantile regression or Bayesian models allows the generation of both point forecasts and confidence intervals, helping home energy management systems (HEMS) make more informed decisions regarding battery charging, appliance usage, or grid interaction. Integrating these forecasts into a real-time HEMS platform enables the optimization of self-consumption, cost reduction, and improved energy independence especially through the intelligent coordination of renewable sources (solar, wind), battery storage, and, if necessary, backup generators within a hybrid energy system.

Around the world, the full deployment of solar energy is being facilitated by several factors including, but not limited to, the reduced price of solar panels; environmental, political and social concerns; and solar energy undercutting utility prices, inter alia. According to [1] global installed capacity will double every two years; however, significant factors have been identified which impede the speed at which solar dominance can be achieved: lack of investments in efficiency, insufficient government incentives, and regulatory constraints. Small-scale Photovoltaic (PV) installations such those in the residential sector benefit from self-consumption by shifting a load from hours when electricity prices are high to hours when the PV energy is being generated, thereby achieving electricity bill savings. Going one step further, the aggregation and coordination of several PV installations in the shape of a Virtual Power Plant (VPP) with the accurate forecasting of global production facilitates its integration into the network. Consequently, the increasing PV penetration can lead to the increasing aggregation of PV systems into VPPs. However, these new business models are difficult to implement due to the regulatory constraints.

Power forecasting along with load demand and energy prices, for different time horizons and resolutions, are factored into the equation. For VPPs, spatial horizons should also be considered. Forecasting methods can be classified according to different factors, such as: the forecasted parameter (irradiance or power), the time horizon and resolution, the lead time, the model approach, and the nature of the forecasting statistic. Regarding the forecasted parameter, two different alternatives exist: direct [3] and indirect. The direct method predicts the solar power through historical datasets of PV power generation and weather conditions. Indirect forecasting differs from the direct method in that it firstly predicts the solar irradiance and then, the solar power is calculated by using a performance model of the PV plant. As far as the time horizon is concerned, four categories can be found: nowcasting (from 1 min to several minutes) which is used for real-time optimization in Energy Management Systems (EMSs); short-term forecast (from 1 h to several hours) used for intra-day market participation and for day-ahead operation optimization; medium term forecast (from 1 month to 1 year); and long-term forecast (up to several years). Time resolutions may range from 1 minute for real-time market operations, and 15-minute periods for load-shifting strategies and for optimizing Battery Energy Storage Systems (BESSs), to 1 hour for longer time horizons used by consumption monitoring, and a 1-week resolution for 1-year time horizons which can be used to identify consumption trends. The lead time can be defined as the time difference between the instant when the forecast is launched and the occurrence of the forecasted value, considering the forecast horizon as the maximum forecast lead time. Forecast errors increase with forecast lead time due to the atmospheric motion. As for the model, the optimal method for solar irradiance prediction depends on the forecast lead time [8]. In this regard, four approaches have been widely used [9]: (a) time-series-based statistical models whose aim is to identify patterns between historical datasets and the output parameters; (b) machine learning (ML) models mainly based on artificial neural networks (ANNs), which use historical datasets to learn the dependency between the past and the future; (c) physical strategies which utilize Numerical Weather Prediction (NWP) and PV models for solar power forecasting; and (d) hybrid models which explore different algorithm combinations with the aim of improving forecast accuracy and reducing the computational burden of online forecasting applications.