Modelling an agent to trade on behalf of V2G drivers

Abstract

Due to the limited availability of fuel resources, there is an urgent need for using renewable sources effectively. To achieve this, power consumers should participate actively in the consumption and production of power. Consumers nowadays can produce power and consume a portion of this locally, and they could offer the rest of the power to the grid. This new feature allows for new decisions for the power consumers. Specifically, vehicle-to-grid (V2G), which is one of the most effective sustainable solutions, could provide these opportunities due to its power storage capability. V2G is where an electric vehicle (EV) offers electric power to the grid when parked. Moreover, V2G could use solar and wind power and significantly decrease the amount of primary power that is utilized for transportation. Furthermore, it offers a potential for reducing the consumers' power cost if used effectively.

In this thesis, the specific problems that we discuss can be categorized into three levels of complexity. At the simplest level is the problem of understanding the power market price behavior in the context of V2G, where we have complete information about the vehicle usage behavior and we assume there is one trip a day. At the next level, the problem of uncertainty in power market price is considered, while we keep the same assumption for the vehicle usage behavior. One of the real-life examples of this model is the bus timetable trip, where there is complete information about the trip times and the uncertainty is only on the power market prices side. Lastly, in addition to the problem of the uncertainty in the power market price, the uncertainty in vehicle usage behavior for the drivers is included for possible multiple trips in a day. The real-life example of this model is the normal vehicle drivers, where there is a chance that they will use their vehicle at any time, and so there are two types of uncertainty, in vehicle usage behavior and in the power market prices. For each of these subjects, we proposed a model and also conducted two surveys in order to attain our study aims.

In more detail, initially, we develop an agent to trade on behalf of V2G users in terms of maximising their profits without uncertainty in the power market price. We then run the proposed model in three different scenarios using an optimal algorithm based on backward induction concept and we compare the results for our solution to a simple benchmark. These scenarios have been proposed to model the user behaviour for the duration of a single day where we assume that users drive their cars for a single period per day. Furthermore, these scenarios differ according to when the drivers started using their cars. We show that our solution outperformed the simple strategy in the first scenario by 49%. Moreover, in the second scenario it outperformed the simple strategy by 51%, while in the third scenario our solution outperformed the simple strategy by 10%.

Next, we develop a heuristic algorithm that can trade on behalf of the V2G users in terms of maximising their profits, considering price uncertainty. Our proposed algorithm is combining the concept of consensus algorithms and expected value with a backward induction algorithm. We then run the proposed algorithm with two types of consensus algorithms using Borda, and majority voting, and with expected value algorithm and compare the results for each algorithm. The concept of consensus can be defined as that there are several samples of feasible steps to be considered at each period of time. After solving each sample, the decision that appears most frequently at time t is selected. Simulations show that, expected value algorithm outperform the other two (Borda and majority) under all power market prices scenarios considered.

Finally, we increase the complexity for the problem by considering the uncertainty in the vehicle usage behavior in the context of V2G in addition to the uncertainty in the power market price. Furthermore, we consider the battery degradation cost, which happens because of the charging or discharging actions. To do such, we refine the second model and we use the multinomial logit model to consider the vehicle usage behaviour. We then run the proposed algorithm and the benchmark algorithms and compare the results for each algorithm. Simulation shows that, our proposed algorithm outperforms the naive algorithm for about 15 times in terms to the average profits when we start the experiments with a half amount of battery. Moreover, our proposed algorithm outperforms the naive algorithm for about 5 times in terms to the average profits when we start the experiments with a full amount of battery. On the other hand, our proposed algorithm performs 89% of the complete information algorithm in terms to the average profits when we start the experiments with a half amount of battery. Furthermore, complete information provides almost same results of our proposed algorithm in terms to the average profits when we start the experiments with a full amount of battery. Indeed, this is good result if we consider that, complete information algorithm deals with known information and the proposed algorithm deals with uncertain data. vehicle at any time, and so there are two types of uncertainty, in vehicle usage behavior and in the power market prices. For each of these subjects, we proposed a model and also conducted two surveys in order to attain our study aims.

More information

Identifiers

ORCID for Enrico Gerding: orcid.org/0000-0001-7200-552X

Catalogue record

Export record