In the recent period, the share of distributed renewable sources of electricity (RES) in the medium-voltage (SN) and low-voltage (NN) distribution network is experiencing an exceptional increase. RES and its variable and uncertain production, especially solar photovoltaic (PV) and wind power plants, which cannot be dispatched, have a significant impact on the reliability of the electricity grid. Distribution network operators (DSOs) have traditionally been dealing with low voltage problems in the peak hours of consumption. Due to the expansion of RES, they have recently begun to deal with overvoltages during the peak production hours. These problems have traditionally already been solved in the phase of planning and dimensioning of the power system and during the operational phase by investing in the network reinforcements. Traditional methods are expensive, and in some cases almost impossible to implement due to physical constraints (e.g. in densely populated cities, shopping centres, etc.). The need to reduce these costs and to minimize the deviation between the produced and the consumed energy in the distribution network increases the need for the accelerated development of demand response (DR) programs, designed to encourage consumers to change their patterns of active energy consumption.
Demand Response Units (DR) feature adjustable consumption and are also referred to as flexibility units. Their control strategies offer new possibilities for more advanced control and operation of the distribution network and for improving the technical conditions in the network. The new possibilities also present new opportunities for new trading strategies of the trader or supplier in the electricity market. Suppliers can act as an aggregator and combine active customers and their DR units, e.g. owners of the heat pumps, electric vehicles and other DR units into a group that adjusts its consumption as needed. In a completely deregulated electricity market, the operation of the distribution network is a regulated activity, separated from the market operations. The DR units are usually privately owned, but their owners want to increase profits so the state-level incentives are being developed to stimulate their participation in commercial adaptation programs to provide flexibility. Aggregators are the actors in the electricity market, and the role is most often assumed by the suppliers who manage aggregation platforms to create a flexible energy portfolio. By combining DR in a portfolio across several medium voltage and low voltage networks, the aggregators can offer
flexible energy products on the electricity wholesale and reserve markets or provide ancillary services to network operators.
Examples of these flexible energy products are the 15-minute flexibility products on the balancing market and one-hour products on the intraday wholesale electricity market, while examples of frequency-related system services include the provision of manual frequency restoration reserve (rRPF) or automatic frequency restoration reserve (aRPF). DR can also provide other types of system services such as voltage and current congestion control for DSOs and regulation of phase imbalance (e.g. with advanced inverters). The aggregator plans to operate the DR in its flexible energy portfolio in order to increase its profits, which are then shared with the participating units in accordance with the agreed business model.
From the society point of view, it is desirable to increase the activity of customers, thereby increasing their participation in the DR programs and co-operating with flexibility schemes. Traditionally, the technical effect of DR operation in low voltage networks is viewed as positive, since it activates the flexible energy that can help fill the growing need for flexibility in the electricity system. DR can be used to reduce the peak demand, to increase consumer’s own consumption or provide system services.
If flexible-load units change their consumption at the wrong time, however, these actions can negatively affect local network conditions and the operational reliability of the distribution network, which would lead to violations of voltage constraints or line congestions. In such cases, the distribution network operator must intervene and ensure that the normal consumption schedules, together with the forecast of local RES and scheduled actions of DR, do not lead to a breach of the operational limitations of the network. The DSO must therefore retain the ultimate control over all the elements that can be controlled on the network.
One of the effective ways to ensure the full control of the DSO over the actions of active customers is a traffic light system (TLS), which was developed as part of the doctoral research and is presented in the INCREASE project. The INCREASE project was co-financed by the European Commission within the EU's Seventh Framework Program for EU Research and Technological Development (EU FP7) which focused on the management of RES and DR in the MV and LV networks with the aim of providing system services to the DSOs, such as voltage control and provision of a reserve. The project also examined the various options for providing
different services to aggregators with new advanced control strategies, along with new business models for the aggregators and for the owners of RES and DR. TLS allows the DSO to approve or reject the scheduled DR actions proposed by aggregators if their actions lead to network problems.
There are other solutions that enable the DSO to keep the network conditions within the prescribed limits and help to alleviate the problems in the distribution network, such as: (i) installation of the On-Load Tap Changer (OLTC), (ii) installation of FACTS devices and (iii) the traditional strengthening of cables and power lines. To alleviate network problems, DSOs currently do not have any other solutions or systems that would allow them to control and operate exclusively only by checking the status of the network and the influence of the proposed DR schedules. A number of studies have focused on improving the network conditions with the control of DR units in the direction of the optimum functioning for the distribution network, but these studies did not take into account the economic opportunity cost for the aggregator and the DR unit owner that could be avoided by different control strategy. Therefore, the latter is the main topic of the doctoral research.
In the future, a significant increase in the number and hence in the total power of DR is expected. Along with it, the importance of the role of aggregators which will bring together many DR and control them in order to increase efficiency and profit, is expected to increase significantly. For the control of many DRs, the aggregators will require fully automated or at least semi-automated scheduling processes for DR activations. The most appropriate methods for automating the scheduling process are the machine learning methods and artificial intelligence (AI) methods, including the intelligent agent approach. The alternative to these methods is the use of linear optimization techniques that theoretically provide the best result in an unconstrained network, but in the case of a-priori unknown limitations and rejection of timetables by the DSO, their result is significantly worsened. The aim of the doctoral research was to develop the scheduling method for DR based on intelligent agent control that would achieve a better result than economic optimization with rejections in a constrained network.
In addition to the concept of TLS, we have also conceived the new concept of flexibility trading. The concept defines the roles of all stakeholders and the exchange of information between them. The newly developed concept covers all stages of trading, from long-term reservation of flexibility for the provision of system services, to the balancing market. The concept also
includes the preventive and corrective actions of the DSO, depending on the time of energy delivery and TLS implementation.
We have also included an active scheduling agent in the trading concept, who uses the newly developed method of Generalized Q-learning (PQL) for scheduling DR. By using this method, an agent learns to avoid scheduling rejection and steers toward achieving greater profit by considering / predicting network conditions. Two novel ways of using Generalized Q-Learning method are proposed in the thesis, the one-step- and the two-step PQL, which differ in the scheduling process and the criteria function used for predicting the network constrains.
The results of the newly developed PQL method show that it is possible to reduce the risk of DR schedule rejection, thereby avoiding possible penalties for unsuccessful activation of scheduled flexibility.
The operation and efficiency of various methods for scheduling DR were checked on a test system based on the real network and measurements of consumption and production.