Aggregator Business Processes and Optimization

There are quite a few business processes going on inside an aggregator company. One of the main process that should be done on a daily basis is buying load and generation

control from its customers. However, other business activities such as buying and selling electricity on the market are closely connected to it. When combined, they make up the core business area of an aggregator.

The aggregator deals with power. It regulates it from one side (customers) and trades it on the electricity markets on the other side. In order for the business to be profitable for both aggregator and its customers, this process requires a lot of optimization. Some of optimization aspects are described in this section.

4.10.1 Trading Optimization and DER Scheduling

The signals that sent and received from the customer are used to control DER. These signals include power control signals and tariffs. The optimization of these signals is a complex task that requires mathematical optimization methods. The resulted model must comply with the rules of a certain electricity market (i.e. balance management).

The optimization can be done through specific software that has all the necessary interfaces to the sources of input data, and the corresponding channels for the result output. The software must consider control signals for customers as well as offers for selling the load reduction on electricity markets or through bilateral contracts.

The following requirements can be applied to optimization software:

 Consideration of different market segments in synergy with bilateral contracts Different electricity market design exists in different countries. Therefore, creating unified software for optimization is a great challenge. Most probably it will be developed for specific aggregator and aimed at specific electricity markets. The same issue concerns balance mechanisms.

 Consideration of various customer contracts

The software should be able to perform probabilistic forecasts for market and imbalance prices. This is also quite challenging, as such algorithms might be

variables involved in such calculations. Therefore it is a complicated task to do.

Variables for customers may include: indoor temperature, time of the year, time of last control call, appliance type, etc.

 Calculation based on customer groups

Considering every customer in the aggregator's portfolio individually may take forever. Therefore similar customers must be grouped together and treated similarly.

 Time requirement

The speed of forecast calculation must be low enough to allow balance settlement. For example in Germany, the resolution of balance settlement is 15 minutes. Therefore, the software must provide a new set of control orders every 15 minutes.

Load forecasting is a vital component of aggregator's business. The aggregator should forecast the electricity consumption of its own customers, in case it is a retailer-aggregator. It should also forecast its own power balance.

The consumption can be forecasted within different time ranges. The most important one is short-term load forecasting. Its range is usually seven days. Not only this forecast is important for secure and economic operation of power system, but also it can be used for demand response aggregation, hydro-thermal coordination, and network status analysis.

Various mathematical methods can be used for load forecasting:

 Neural networks

 Time series models

 Linear regression models

Neural networks are machine learning tools. In this approach the user does not have to specify relationship between forecasted load and other variables. However, this method has some issues. It might be difficult to determine the best suited parameters for the neural network. Therefore, empirical determining of such parameters might be needed. This process can take a lot of time. Another issue is that a large amount of historic data (i.e. 1-2 years) is required for this method.

Time series forecasting is the use of a model to forecast future events based on known past events: to predict data points before they are measured. For example, it can be used for predicting the opening price of electricity price based on past information.

Linear regression models express load function as function of exogenous inputs (i.e.

weather, holidays, etc.). However, forecasting nonlinear dependencies with linear model can be inaccurate and can create unnecessary errors in the results.

4.10.3 Distributed Generation Forecasting

The aggregator may have to forecast distributed generation (i.e. wind power, solar power, hydro power, and CHP) in order to forecast its own imbalance position or imbalance position of deregulated parties with whom it has made bilateral contracts.

This forecast requires extensive information about DER generation portfolio.

Distributed generation forecasting is usually based on weather predictions to which mathematical methods described in Load Forecasting section are applied. Forecasting accuracy depends on the local weather variability. A typical forecast error is usually 5-15% of installed power.

4.10.4 Market Electricity Price Forecasting

Time series models and multiple regression methods are used for price forecasting on short-term electric power market. The artificial neural networks can also be used. The benefit of using neural networks is that the user does not have to set relationship between weather and price as the forecasting is mainly based on historic spot prices, power demand, wind speed, sun shine, and temperature.

4.10.5 Issues Connected with Forecasting

The usual business is that the aggregator company predicts customers' load reduction in response to control signals as a function of time. But the realized customers' load can differ from the forecast and therefore it has influence on the imbalance position of the aggregator. The penalties that a customer pays for possible deviations in load are usually not equal to the penalties that the aggregator company faces. Therefore such imbalance probabilities from the customer must be taken into account when making a balance for DSO.

4.11 Existing Aggregators