Robust Trading Decision-Making Model for Demand-Side Resource Aggregators Considering Multi-Objective Cluster Aggregation Optimization

Abstract

In the context of a high proportion of new energy grid connections, demand-side resources have become an inevitable choice for constructing new power systems due to their high flexibility and fast response speed. However, the response capability of demand-side resources is decentralized and fluctuating, which makes it difficult for them to effectively participate in power market trading. Therefore, this paper proposes a robust transaction decision model for demand-side resource aggregators considering multi-objective clustering aggregation optimization. First, a demand-side resource aggregation operation model is designed to aggregate dispersed demand-side resources into a coordinated aggregated response entity through an aggregator. Second, the demand-side resource aggregation evaluation indexes are established from three dimensions of response capacity, response reliability, and response flexibility, and the multi-objective aggregation optimization model of demand-side resources is constructed with the objective function of the larger potential market revenue and the smallest risk of deviation penalty. Finally, robust optimization theory is adopted to cope with the uncertainty of demand-side resource responsiveness, the robust transaction decision model of demand-side resource aggregator is constructed, and a community in Henan Province is selected for simulation analysis to verify the validity and applicability of the proposed model. The findings reveal that the proposed cluster aggregation optimization method reduces the bias penalty risk of the demand-side resource aggregators by about 33.12%, improves the comprehensive optimization objective by about 18.10%, and realizes the optimal aggregation of demand-side resources that takes into account both economy and risk. Moreover, the robust trading decision model can increase the expected net revenue by about 3.1% under the ‘worst’ scenario of fluctuating uncertainties, which enhances the resilience of demand-side resource aggregators to risks and effectively fosters the involvement of demand-side resources in the electricity market dynamics.

1. Introduction

In light of the “dual carbon” objectives, the significant integration of renewable energy sources has emerged as a crucial pathway for the evolution of the power system [1]. However, the strong volatility of new energy sources has proliferated the need for flexible regulation of the power system, posing a great challenge to its stable operation [2,3]. Demand-side resources have become an important means for power systems to improve resource utilization efficiency and promote the use of renewable energy by virtue of the advantages of high flexibility and fast response time [4]. The economy is growing, but it faces constraints due to heavy reliance on fossil fuels and frequent power outages. Access to diversified energy sources, especially electricity, is critical to sustaining this growth [5]. Against this background, there is an urgent need to tap the flexibility adjustment potential of demand-side resources to improve the power system’s ability to cut peaks and fill valleys and to regulate and balance supply and demand.

Demand-side resources have the characteristics of small capacity, large scale, geographically dispersed [6,7,8], and their direct participation in regulation will bring a huge burden to the scheduling organization, and their power fluctuations will also affect the accuracy of power load forecasting and the effectiveness of the deployment [9]. As a result, the scheduling agency will set certain requirements that prevent a single resource entity from independently taking part in electricity market transactions. To harness the flexible regulatory potential of demand-side resources and effectively integrate them into electricity market transactions, the concept of demand-side resource aggregator has emerged [10], aiming to aggregate demand-side resources in a certain region into a coordinated and unified market entity. Under aggregation management, the dispersed fluctuating single resource is integrated into an aggregation response body with a certain scale and relatively controllable, and then the aggregator agent participates in the power market transactions and manages and regulates the user’s electricity consumption behavior [11,12], which can significantly improve the reliability and stability of the demand-side resource regulation capability.

Current research on demand-side resources focuses on aggregation optimization methods [13], aggregator trading decision model construction [14], and so on. Regarding demand-side resource aggregation optimization methods, the optimal aggregation of resources is often achieved through game models, evaluation models, and aggregation optimization models. The literature [15] achieves optimal aggregation of demand-side resources in distribution-level markets through the Nash–Stackelberg game model. The literature [16] constructs an aggregation index system from three aspects: load management, peaking potential, and historical credit, and determines the aggregation priority of each resource using a cloud model. The literature [17] develops a multi-objective aggregation optimization model aimed at reducing both the system’s peak-to-trough variation and the associated regulatory expenses, and proposes a hierarchical partitioned aggregation regulation strategy. However, existing studies ignore the risk of penalizing deviations in the response of demand-side resources to power fluctuations, making it difficult to deal with the risk of market transactions when uncertainties fluctuate in the “worst” direction. Therefore, this paper proposes an aggregation optimization method that takes into account both economy and risk to achieve the optimal aggregation of demand-side resources and guarantee reasonable market returns for aggregators.

In terms of trading decision models for demand-side resource aggregators, existing studies have conducted more research on trading strategy development and response plan development. The literature [18] develops a two-tiered framework enabling demand-side resource aggregators to engage in electricity trading, with the upper tier aiming to maximize social benefits and the lower tier aiming to maximize aggregator revenue. The literature [19] proposes a decision model for EV aggregators to take part in the day-ahead electricity market in response to the uncertainty of EV travel behavior. The literature [20] proposes an optimization method for aggregators to participate in market trading decisions, considering user interests and price elasticity coefficients. However, none of the above literature has been able to explore the high cost of bias penalties associated with uncertainty in responsiveness. Therefore, this paper constructs a robust trading decision model for demand-side resource aggregators based on the optimization results of cluster aggregation and derives a trading strategy adapted to the “worst” scenario of fluctuating uncertainties to improve the risk-resistant ability of aggregators.

Aiming at the above problems, this paper proposes a multi-objective cluster aggregation optimization robust trading decision model for aggregators. One is to promote the participation of demand-side resource aggregation in power trading, and the other is to improve the risk-resistant ability of demand-side resource aggregators. Therefore, a multi-objective cluster aggregation optimization robust trading decision model for aggregators is proposed in this paper. First, the demand-side resource aggregation operation model is introduced, and a multi-objective cluster aggregation optimization method based on maximizing the market revenue and minimizing the risk of deviation penalty is proposed. Second, the aggregator robust trading decision model is constructed to take into account the uncertainty of demand-side resource responsiveness. Finally, a community in Henan Province is selected for arithmetic simulation, and the effectiveness and advancement of the proposed model are verified through comparative analysis.

In this paper, we propose a robust transaction decision model for demand-side resource aggregators considering multi-objective clustering aggregation optimization. Firstly, in the first part of the article, we summarize the research background and the state of the art of related studies, and then, a demand-side resource aggregation operation model is designed to aggregate dispersed demand-side resources into a coherent aggregated response entity through an aggregator. Secondly, demand-side resource aggregation assessment indexes are established from the three dimensions of response capability, response reliability, and response flexibility. A multi-objective demand-side resource aggregation optimization model with greater potential market revenue and minimum deviation penalty risk as objective functions is constructed. Finally, the robust optimization theory is adopted to cope with the uncertainty of demand-side resource responsiveness and a robust trading decision model for demand-side resource aggregators is constructed. In the arithmetic example part, we choose a community in Henan Province for simulation and analysis to verify the validity and applicability of the proposed model and put forward relevant suggestions based on the research results.

2. Demand-Side Resource Aggregation Operating Model

2.1. Aggregate Response Framework

Demand-side resources are numerous and small in capacity, and the individual response capacity of a single resource often fails to meet the conditions for access to the electricity market. In addition, demand-side resource users usually do not have the ability to receive price signals from the electricity market in real time, such as real-time communication and centralized processing of electricity consumption data [21]. Consequently, demand-side resources frequently find themselves incapable of engaging in electricity market transactions on their own. In this situation, the demand-side resource aggregation operation mode regulates a certain scale of resources through aggregator aggregation so that the whole system participates in the transaction, to address this issue of regulating and optimizing the huge amount of resources. Figure 1 shows the framework of aggregation response.

2.2. Aggregation Operating Model

Under the aggregated response framework, the aggregator, as a third party that manages the aggregation of demand-side resources, signs an agency contract with the user, acts as an agent to participate in the electricity market transactions and regulates the behavior of electricity consumption, and calculates and settles the amount of response power as a whole for all the resources that the scheduling agency regards as its agent in the market transactions. The role of aggregators can be understood from two aspects: at the physical level, as an intelligent regulation platform, it adopts advanced technologies to realize capacity aggregation and response optimization in order to bring into play the regulation potential; at the market level, as an agent, it can act as an agent for users to declare, settle, etc., and formulate a response strategy so that the households can obtain the market revenue. Figure 2 shows a schematic diagram of the aggregation operating model.

However, the demand-side resource aggregation operating model relies on the aggregator’s resource aggregation management, market transaction management, and many other complex functions [22]. Figure 3 shows the aggregator’s functional architecture and flow. In actual operation, the demand-side resource aggregator first evaluates the response performance of each resource and performs aggregation optimization, determines the set of proxy resources, and signs proxy contracts with individual users. After receiving the regulation demand information, the aggregator takes into account the resource response capacity, forecasted clearing price, and other information in the previous day’s stage, decides on the optimal bidding and trading strategy by combining the regulation period with the demand, and reports the regulation capacity and price to the scheduling organization; and in the real-time stage, it takes into account the information on the operating status of the resources to formulate the real-time response plan and the optimal response strategy.

After the response is completed, the dispatching organization settles the revenue for the aggregator according to the effective response power and collects the deviation penalty fee for the unfinished response power, and finally, the aggregator settles the response for each resource in accordance with the contractual compensation price and deviation penalty ratio.

3. A Multi-Objective Cluster Aggregation Optimization Method for Demand-Side Resources

3.1. Multidimensional Aggregated Evaluation Indicators

In this section, aggregation metrics are selected from the three dimensions of response capacity, response reliability, and response flexibility to comprehensively represent the response performance of each demand-side resource, respectively.

Table 1. Evaluation indicators for aggregation of demand-side resources.

Evaluation Dimension	Evaluation Indicators	Explicit Description
response capacity	response capacity	maximum response power available
	response fluctuation range	difference between the maximum and minimum values of the response charge
response reliability	degree of response fluctuation	standard deviation of fluctuations in response power
response flexibility	duration	length of time that the response lasts from the start of the response to the end of the response
	response rate	length of time between dispatch instruction issuance and start of response

shows the aggregated evaluation indicators for the three dimensions.

3.1.1. Response Capacity

(a)

Response capacity

Response Capacity means the maximum available response power of a Demand Side Resource during a Regulation Period, calculated as follows:

$$F_{RC,t}^i=u_{RP,t}^i\underset{k}{\max }{P}_{his,k,t}^{i,DR}$$

(1)

where $F_{RC,t}^i$ is the value of the response capacity indicator for the demand-side resource $i$ during the regulation period $t$; $u_{RP,t}^i$ is a 0–1 variable indicating whether the resource $i$ can provide response power in the specified direction during the period $t$; and $P_{his,k,t}^{i,DR}$ is the response power of the demand-side resource $i$ when it participates in the response for the first time during the period $t$ at $k$.

(b)

Response fluctuation range

Response fluctuation range means the difference between the maximum and minimum values of the historical response power of a demand-side resource, calculated as follows:

$$F_{RR,t}^i=u_{RP,t}^i\left(\underset{k}{\max }{P}_{his,k,t}^{i,DR}-\underset{k}{\min }{P}_{his,k,t}^{i,DR}\right)$$

(2)

where $F_{RR,t}^i$ is the value of the response fluctuation range indicator for the demand-side resource $i$ during the regulation period $t$, respectively.

3.1.2. Response Reliability

Response reliability is used to describe how reliably demand-side resources provide responsive power to the power system in actual operation, and is usually measured by the degree of response fluctuation.

The degree of response fluctuation refers to the standard deviation of fluctuations in historical response power from demand-side resources, calculated as follows:

$$F_{RL,t}^i=u_{RP,t}^i\sqrt{{\displaystyle \frac{{\displaystyle \sum _k{\left(P_{his,k,t}^{i,DR}-{\overline{P}}_{his,t}^{i,DR}\right)}^2}}{N_{his}}}}$$

(3)

where $F_{RL,t}^i$ is an indicator of the degree of response fluctuation of demand-side resources during the regulation period; ${\overline{P}}_{his,t}^{i,DR}$ is the average of the historical response of demand-side resources during the period; and $N_{his}$ is the number of historical responses.

3.1.3. Response Flexibility

Response flexibility is used to describe the comprehensive response performance of demand-side resources, which mainly includes 2 evaluation indicators: duration and response rate.

(1)

Duration

Duration is the length of time a demand-side resource lasts from the start of the response to the end of the response, calculated as follows:

$$F_{RD}^i=\underset{l}{\min }\left(t_{end}^{i,l}-t_{begin}^{i,l}\right)$$

(4)

where $F_{RD}^i$ is the duration metric value for the demand-side resource $i$; $t_{begin}^{i,l}$ and $t_{end}^{i,l}$ are the times at which the demand-side resource $i$ started responding and ended responding in the $l$ test, respectively.

(2)

Response rate

Response rate refers to the length of time it takes for a dispatch instruction to be issued and for a demand-side resource to start responding, and is calculated as follows:

$$F_{RS}^i={\displaystyle \frac{{\displaystyle \sum _l\left(t_{begin}^{i,l}-t_{send}^{i,l}\right)}}{N_{test}}}$$

(5)

where $F_{RS}^i$ is the response rate metric value of the demand-side resource $i$; $t_{send}^{i,l}$ is the time at which the demand-side resource aggregator sends a scheduling instruction to the user in the $l$ test; and $N_{test}$ is the number of times the aggregator conducts an online response test.

3.2. Multi-Objective Aggregate Optimization Model

In the process of demand-side resource aggregation optimization, it is necessary to consider two factors. One is the maximization of response capacity gain, and the other is the minimization of fluctuation deviation penalty risk. These considerations are essential for constructing an objective function. The objective function is ultimately used to determine the cluster aggregation results. Figure 4 shows the demand-side resource aggregation optimization framework.

3.2.1. Objective Function

(1)

Maximizing potential market revenue

The higher the response capacity of the demand-side resource aggregation response entity, the greater the potential market benefits it receives. The calculation is as follows:

$$\max B_{DR}={\displaystyle \sum _{t\in T_{ad}}\left(u_{C,t}{p}_{com,t}{\displaystyle \sum _{i\in \Omega }{u}_{in}^i{F}_{RC,t}^i}\right)}$$

(6)

$$u_{C,t}=\left\{\begin{array}{l}1,{\displaystyle \sum _{i\in \Omega }{u}_{in}^i{F}_{RC,t}^i}\ge F_{RC,\min }\\ 0,{\displaystyle \sum _{i\in \Omega }{u}_{in}^i{F}_{RC,t}^i}<F_{RC,\min }\end{array}\right.$$

(7)

where $B_{DR}$ is the objective function for maximizing potential market returns for demand-side resource aggregation responders; $u_{C,t}$ is a 0–1 variable that indicates the aggregator’s regulatory capacity in the regulatory time period $t$; $p_{com,t}$ is the market clearing price in the regulatory time period $t$, which can be an average of the historical clearing price; $u_{in}^i$ is a 0–1 variable that controls whether or not the demand-side resource $i$ is included in the aggregator’s response; $T_{ad}$ is the pool of the regulatory time period; $\Omega$ is the pool of the region’s demand-side resource aggregation. Equation (7) indicates that the aggregated responders participating in the electricity market need to meet the requirements of the dispatching agency for response capacity.

(2)

Minimizing the risk of deviation penalties

In electricity market trading, the scheduling agency will charge a high deviation penalty fee for not completing the response power, and the larger the response volatility deviation, the higher the risk of deviation penalty. The calculation formula is as follows:

$$\min V_{DR}={\lambda }_{RR}{V}_{RR}+{\lambda }_{RL}{V}_{RL}$$

(8)

$$V_{RR}={\displaystyle \sum _{t\in T_{ad}}\left(p_{pun,t}{\displaystyle \sum _{i\in \Omega }{u}_{in}^i{F}_{RR,t}^i}\right)}$$

(9)

$$V_{RL}={\displaystyle \sum _{t\in T_{ad}}\left(p_{pun,t}{\displaystyle \sum _{i\in \Omega }{u}_{in}^i{F}_{RL,t}^i}\right)}$$

(10)

where $V_{DR}$ is the objective function for minimizing the risk of deviation penalties for demand-side resource aggregation respondents; $V_{RR}$ and $V_{RL}$ are the risk of deviation penalties for aggregation respondents in terms of the range of response fluctuations and the degree of response fluctuations, respectively; ${\lambda }_{RR}$ and ${\lambda }_{RL}$ are the aggregator’s risk appetite weights for the range of response fluctuations and the degree of response fluctuations, respectively, and there exists ${\lambda }_{RR}+{\lambda }_{RL}=1$; and $p_{pun,t}$ is the price of deviation penalties for the regulating time period $t$ as stipulated by the scheduling agency.

3.2.2. Constraints

(1)

Duration constraints

In order to ensure the security and stability of the power system, the aggregator is required to ensure that each resource subject of the agent is able to meet the minimum duration specified by the scheduling authority as follows:

$$u_{in}^i\le u_{RD}^i$$

(11)

$$u_{RD}^i=\left\{\begin{array}{l}1,F_{RD}^i\ge F_{RD,\min }\\ 0,F_{RD}^i<F_{RD,\min }\end{array}\right.$$

(12)

where $u_{RD}^i$ is a 0–1 variable indicating whether the demand-side resource $i$ meets the minimum duration specified by the scheduling authority; $F_{RD,\min }$ is the minimum duration specified by the scheduling authority.

(2)

Response rate constraints

In order to efficiently organize demand-side resources to provide responsive power to the power system, aggregators usually require each resource entity of the agent to meet its minimum response rate as follows:

$$u_{in}^i\le u_{RS}^i$$

(13)

$$u_{RS}^i=\left\{\begin{array}{l}1,F_{RS}^i\ge F_{RS,\min }\\ 0,F_{RS}^i<F_{RS,\min }\end{array}\right.$$

(14)

where $u_{RS}^i$ is a 0–1 variable, indicating whether the demand-side resource $i$ meets the minimum response rate specified by the aggregator; $F_{RS,\min }$ is the minimum response rate.

Drawing from the aforementioned objective function and constraints, the demand-side resource multi-objective aggregation optimization model is established as follows:

$$\begin{array}{l}\max B_{DR}={\displaystyle \sum _{t\in T_{ad}}\left(u_{C,t}{p}_{com,t}{\displaystyle \sum _{i\in \Omega }{u}_{in}^i{F}_{RC,t}^i}\right)}\min V_{DR}={\lambda }_{RR}{V}_{RR}+{\lambda }_{RL}{V}_{RL}\\ s.t.\left\{\begin{array}{l}\mathrm{duration} \mathrm{constraints}: \mathrm{Equations} (11) \mathrm{and} (12)\\ \mathrm{response} \mathrm{rate} \mathrm{constraints}: \mathrm{Equations} (13) \mathrm{and} (14)\end{array}\right.\end{array}$$

(15)

3.3. Integration of Subjective and Objective Empowerment

The demand-side resource aggregation optimization model has two objective functions, because the AI swarm algorithm has a great reduction in the solution efficiency when optimizing the aggregation of massive resources, so the model is converted from a multi-objective to a single-objective solution by weight division. It is shown as follows:

$$\begin{array}{c}\min F_{DR}={\lambda }_V{V}_{DR}-{\lambda }_B{B}_{DR}\\ s.t.\left\{\begin{array}{l}\mathrm{duration} \mathrm{constraints}: \mathrm{Equations} (11) \mathrm{and} (12)\\ \mathrm{response} \mathrm{rate} \mathrm{constraints}: \mathrm{Equations} (13) \mathrm{and} (14)\end{array}\right.\end{array}$$

(16)

where $F_{DR}$ is the combined optimization objective of the single-objective aggregation optimization model for demand-side resources; ${\lambda }_V$ and ${\lambda }_B$ are the weights of the objective of minimizing the risk of bias penalties and the objective of maximizing potential market gains, respectively. Since both objective functions aim at market profit and loss, the multi-objective aggregation model does not require de-quantification. In order to take into account the influence of subjective and objective factors on the model solution, this section uses the subjective–objective integration method to assign weights to the two objective functions [23] for the two objective functions.

The integration of Subjective and Objective Empowerment method includes three parts: (1) calculation of subjective factor weights, (2) calculation of objective factor weights, and (3) subjective and objective integration weighting. First, the expert scoring method is used to evaluate the importance of the two optimization objectives to determine the subjective factor weight, then the entropy weight method is used to determine the objective factor weight according to the data dispersion degree, and finally, the subjective factor weight is integrated according to the subjective factor bias coefficient, so as to make the weight more reasonable and avoid the subjective shortcomings.

3.3.1. Calculation of Subjective Factor Weights

In this section, the expert scoring method is used to classify the subjective factor weights for the above two optimization objectives in the following steps [24].

(1)

Determine the evaluation level of importance of the objective function, which is divided into five levels: not important at all, not important, general, important, and very important;

(2)

A number of experts in the field were invited to evaluate the importance of the two optimization objectives;

(3)

Summarize and analyze the results of the experts’ evaluations and calculate the weights of the subjective factors for each objective function $u_j$.

$$S_j={\displaystyle \frac{{\displaystyle \sum _i{S}_{ij}}}{n}}$$

(17)

$$u_j={\displaystyle \frac{S_j}{{\displaystyle \sum _{j=1}^m{S}_j}}}$$

(18)

where $S_{ij}$ is the rating of the $j$ objective function by the $i$ expert, and the above five evaluation levels can be taken as 1, 2, 3, 4, and 5 points, respectively; $S_j$ is the final evaluation result of the $j$ objective function; $n$ is the number of experts. The calculated subjective factor weights of the objective function of maximizing potential market returns and minimizing the risk of bias penalties are recorded as $w_B^s$ and $w_V^s$, respectively.

3.3.2. Calculation of Objective Factor Weights

In this section, the entropy weighting method is used to classify the objective factor weights for the above two optimization objectives in the following steps:

(1)

Construct the matrix of objective functions under different decision scenarios $F={(f_{ij})}_{n\times m}$, $n$ is the number of scenarios and $m$ is the number of objective functions;

(2)

Then, the degree of information bias for each objective function is calculated as follows:

$$E_j=-{\displaystyle \frac{1}{\ln n}}{\displaystyle \sum _{i=1}^n{r}_{ij}\ln (r_{ij})}$$

(19)

$$D_j=1-E_j$$

(20)

where $r_{ij}$ is the normalized value of $f_{ij}$; $E_j$ is the information entropy of each objective function; $D_j$ is the information deviation of each objective function.

(3)

Calculate the objective factor weights of each objective function $v_j$ as follows:

$$v_j={\displaystyle \frac{D_j}{{\displaystyle \sum _{j=1}^m{D}_j}}}$$

(21)

Based on the above operations, the calculated objective factor weights for the potential market return maximization objective function and the deviation penalty risk minimization objective function are denoted as $w_B^o$ and $w_V^o$, respectively.

3.3.3. Subjective and Objective Integration Weighting

Due to the limitations of the single-weight division method, the final weights of the two objective functions are determined based on the weights of subjective and objective factors [25] that are shown below:

$$\left\{\begin{array}{l}{w}_B={\left(w_B^s\right)}^r{\left(w_B^o\right)}^{1-r}/W\\ w_V={\left(w_V^s\right)}^r{\left(w_V^o\right)}^{1-r}/W\end{array}\right.$$

(22)

$$W={\left(w_B^s\right)}^r{\left(w_B^o\right)}^{1-r}+{\left(w_V^s\right)}^r{\left(w_V^o\right)}^{1-r}$$

(23)

where $r$ represents the decision maker’s bias coefficient for subjective factors, with a value range of 0–1, and $1-r$ is the bias coefficient for objective factors. In Equation (23), the closer the value of $r$ is to 1, the closer the result of subjective–objective integrated weighting is to the weight of subjective factors; conversely, the closer it is to the weight of objective factors. After that, $w_B$ and $w_V$ are normalized respectively to obtain the final weights ${\lambda }_V$ and ${\lambda }_B$, respectively.

4. A Robust Trading Decision Model for Demand-Side Resource Aggregators

4.1. Robust Optimization Theory

Assuming that the decision problems in both the day-ahead and real-time phases are linear programming problems and that the uncertainty variables take values in the space of discrete finite point sets or polyhedra, the two-phase robust optimization model can be expressed by the following equation [26].

$$\begin{array}{l}\underset{\mathit{x}}{\min }\left({\mathit{c}}^T\mathit{x}+\underset{\mathit{u}\in \mathit{U}}{\max }\underset{\mathit{y}\in \Omega (\mathit{u},\mathit{x})}{\min }{\mathit{d}}^T\mathit{y}\right)\\ s.t.\left\{\begin{array}{l}\mathit{A}\mathit{x}\ge \mathit{b}\\ \mathit{G}\mathit{y}\ge \mathit{h}-\mathit{E}\mathit{x}-\mathit{M}\mathit{u}\\ \mathit{x}\in S_x,\mathit{y}\in S_y\end{array}\right.\end{array}$$

(24)

where $\mathit{x}$ and $\mathit{y}$ denote the decision variables in the first and second stages; $\mathit{u}$ and $\mathit{U}$ denote the value space of the uncertainty variables and uncertainty variables, respectively; $\Omega (u,x)$ is the value space of the uncertainty variables given $\mathit{x}$ and $\mathit{u}$; $\mathit{A}$, $\mathit{G}$, $\mathit{E}$ and $\mathit{M}$ denote the coefficient matrices of the corresponding constraints, respectively; $\mathit{b}$ and $\mathit{h}$ are vectors of constant columns; and $S_x$ and $S_x$ are the ranges of the values of x and y, respectively. The first-stage model formulates the decision scenario $\mathit{x}$ with the uncertainty variable $\mathit{u}$ unknown, the second-stage outer model solves the worst-case scenario of $\mathit{u}$ with $\mathit{x}$ known, and the second-stage inner model formulates the decision scenario $\mathit{y}$ with $\mathit{x}$ and $\mathit{u}$ known. The space of values $\mathit{U}$ of the uncertainty variable $\mathit{u}$ can usually be expressed by the following equation [27]:

$$\mathit{u}\in \mathit{U}=[\tilde{\mathit{u}}-\Delta \underset{\_}{\mathit{u}},\tilde{\mathit{u}}+\Delta \overline{\mathit{u}}]$$

(25)

$$\Delta \overline{\mathit{u}}=\overline{\mathit{u}}-\tilde{\mathit{u}}$$

(26)

$$\Delta \underset{\_}{\mathit{u}}=\tilde{\mathit{u}}-\underset{\_}{\mathit{u}}$$

(27)

where $\tilde{\mathit{u}}$ represents the predicted value of $\mathit{u}$; $\Delta \overline{\mathit{u}}$ and $\Delta \underset{\_}{\mathit{u}}$ represent the upward and downward fluctuation ranges of $\mathit{u}$; $\overline{\mathit{u}}$ and $\underset{\_}{\mathit{u}}$ represent the upper and lower bounds of $\mathit{u}$, respectively. Since the definition of the ‘worst case’ scenario is often more in line with the definition of the uncertainty variables when they are taken to the boundaries, Equation (25) can be rewritten in the following form, using the lower bound as an example of the worst case scenario:

$$\mathit{u}=\tilde{\mathit{u}}-\Gamma \Delta \underset{\_}{\mathit{u}}$$

(28)

where $\Gamma$ is the robustness coefficient; the larger the value, the higher the degree of badness of the fluctuation of the uncertainty variable, and the more robust the decision-making program is. However, the above equation tends to make the decision-making scheme too conservative, so the adjustment parameter $L$ is introduced to control the ‘worst’ degree of uncertainty variables. Therefore, Equation (28) can be rewritten as follows:

$$\mathit{u}=\tilde{\mathit{u}}-\Gamma \mathit{B}\Delta \underset{\_}{\mathit{u}}$$

(29)

$${\displaystyle \sum \mathit{B}}\le L$$

(30)

where $\mathit{B}$ is the variable controlling the ‘worst degree’ of uncertainty, and its value ranges from 0 to 1.

4.2. Robust Trading Decision Models

4.2.1. Day-Ahead Bidding Decision Model

The aggregator decides on the optimal regulation capacity and price based on the expected market clearing price and resource responsiveness with the objective of maximizing the expected net income from the market. The objective function is specified as follows:

$$\max U_{DR}^{DA,1}={\displaystyle \sum _{t\in T_{ad}}\left[\begin{array}{l}{p}_{com,t}\min \left({\alpha }_{re}{P}_{DR,t}^b,{\displaystyle \sum _{i\in I}{P}_{DR,t}^i}\right)-{\displaystyle \sum _{i\in I}{p}_{buy}^i{P}_{DR,t}^i}\\ -p_{pun,t}\max \left({\alpha }_{pun}{P}_{DR,t}^b-{\displaystyle \sum _{i\in I}{P}_{DR,t}^i},0\right)\end{array}\right]}$$

(31)

where $U_{DR}^{DA,1}$ represents the expected net income of the demand-side resource aggregator participating in the market transaction; ${\alpha }_{re}$ is the upper limit ratio of effective response power; $P_{DR,t}^b$ is the regulating capacity declared by the aggregator; $P_{DR,t}^i$ is the expected response power of resource i in time period t; $p_{buy}^i$ is the aggregator’s purchase price for the response power of resource i; $p_{pun,t}$ is the deviation penalty price of time period t; ${\alpha }_{pun}$ is the lower limit ratio of the deviation penalty; and $I$ is the set of the aggregator’s proxy resources. In the day-ahead bidding decision model, $P_{DR,t}^b$ and $P_{DR,t}^i$ have the following relationship:

$$P_{DR,t}^b={\displaystyle \sum _{i\in I}{P}_{DR,t}^i}$$

(32)

In addition, the day-ahead bidding decision model includes the following constraints:

(1)

Regulation Capacity Constraints

The aggregator’s declared regulation capacity needs to be greater than the minimum regulation capacity specified by the scheduler as follows:

$$P_{DR,t}^b\ge P_{DR,\min }^b$$

(33)

where $P_{DR,\min }^b$ is the minimum regulation capacity specified by the dispatching authority.

(2)

Response capacity constraint

The expected response power of the demand-side resource needs to be less than its maximum response capacity, as follows:

$$0\le P_{DR,t}^i\le P_{DR,t}^{i,\max }$$

(34)

The initial values of $p_{com,t}$ and $P_{DR,t}^{i,\max }$ are the predicted values of the market clearing price and the resource like a capacity, i.e., $p_{com,t}^f$ and $P_{DR,t}^{i,f}$, respectively. The optimal solution to the above model yields the aggregator’s declared regulation capacity $P_{DR,t}^b$. On this basis, the aggregator’s unit regulation capacity cost is accounted for as follows:

$$C_{DR,t}^b={\displaystyle \frac{{\displaystyle \sum _{i\in I}{p}_{buy}^i{P}_{DR,t}^i}+p_{pun,t}\max \left({\alpha }_{pun}{P}_{DR,t}^b-{\displaystyle \sum _{i\in I}{P}_{DR,t}^i},0\right)}{P_{DR,t}^b}}$$

(35)

Substituting Formula (32) into Formula (35) yields

$$C_{DR,t}^b={\displaystyle \frac{{\displaystyle \sum _{i\in I}{p}_{buy}^i{P}_{DR,t}^i}}{P_{DR,t}^b}}$$

(36)

where $C_{DR,t}^b$ is the aggregator’s acquisition cost per unit of responsive power in the time period $t$. Then, the aggregator’s regulation price is calculated based on the aggregator’s minimum income expectation function as follows:

$$p_{DR,t}^b=a{C}_{DR,t}^b+b$$

(37)

where $a$ and $b$ are the primary and constant term coefficients of the aggregator’s minimum income expectation function; $p_{DR,t}^b$ is the regulated price declared by the aggregator during the period t.

Ultimately, the aggregator’s declared capacity optimized by the previous day’s bidding decision model is passed on to the expected trading decision model, and the declared price does not need to be passed on because it has little impact on the market clearing price.

4.2.2. Expected Trading Decision Model

In the expected transaction decision-making stage, the aggregator decides the real-time response plan based on the market clearing price and resource response capacity, with the goal of maximizing net income, and arrives at the ‘worst case’ scenario, with the following objective function.

$$\max U_{DR}^{DA,2}={\displaystyle \sum _{t\in T_{ad}}\left[\begin{array}{l}{p}_{com,t}\min \left({\alpha }_{re}{P}_{DR,t}^b,{\displaystyle \sum _{i\in I}{P}_{DR,t}^{i,r}}\right)-{\displaystyle \sum _{i\in I}{p}_{buy}^i{P}_{DR,t}^{i,r}}\\ -p_{pun,t}\max \left({\alpha }_{pun}{P}_{DR,t}^b-{\displaystyle \sum _{i\in I}{P}_{DR,t}^{i,r}},0\right)\end{array}\right]}$$

(38)

where $U_{DR}^{DA,2}$ denotes the net income of the aggregator; $P_{DR,t}^{i,r}$ is the real-time responsive power of resource $i$ in time period $t$.

The demand-side resource’s responsive power needs to be less than its real-time responsive capacity, as follows:

$$0\le P_{DR,t}^{i,r}\le P_{DR,t}^{i,\max }$$

(39)

In addition, the following constraints on clearing prices and responsiveness are required to be consistent with the definition of the ‘worst’ scenario:

$$p_{com,t}=p_{com,t}^f-{\Gamma }_{com}{B}_{com,t}\Delta p_{com,t}^d$$

(40)

$$\Delta p_{com,t}^d=p_{com,t}^f-p_{com,t}^d$$

(41)

$${\displaystyle \sum _{t\in T_{ad}}{B}_{com,t}}\le L_{com}$$

(42)

$$P_{DR,t}^{i,\max }=P_{DR,t}^{i,f}-{\Gamma }_{DR}^i\Delta P_{DR,t}^{i,d}$$

(43)

$$\Delta P_{DR,t}^{i,d}=P_{DR,t}^{i,f}-P_{DR,t}^{i,d}$$

(44)

where $\Delta p_{com,t}^d$ and $\Delta P_{DR,t}^{i,d}$ are the range of fluctuations in market clearing prices and responsiveness towards the ‘worst case’ scenario for the time period $t$, respectively; ${\Gamma }_{com}$ and ${\Gamma }_{DR}^i$ are robust coefficients controlling the degree of conservatism in market clearing prices and responsiveness towards the ‘worst case’ scenario, respectively; $B_{com,t}$ is a variable controlling the ‘worst case’ level of market clearing prices; with a value range of 0 to 1; $L_{com}$ is the adjustment parameter controlling the overall ‘worst degree’ of market clearing price on the execution day.

On the basis of the above model, the aggregator’s two-stage robust bidding model is written in the following form.

$$\begin{array}{l}\underset{\mathit{x}}{\max }\left(\underset{\mathit{u}\in \mathit{U}}{\min }\underset{\mathit{y}\in \Omega (\mathit{u},\mathit{x})}{\max }{U}_{DR}^{DA}\right)\\ s.t.\left\{\begin{array}{l}\mathit{A}\mathit{x}=\mathit{a}\hspace{1em}\hspace{1em} \to \mathrm{Equation} (32)\\ \mathit{B}\mathit{x}\ge \mathit{b}\hspace{1em}\hspace{1em} \to \mathrm{Equation} (33)\\ \mathit{M}\mathit{x}+\mathit{Y}\mathit{u}\ge \mathbf{0}\to \mathrm{Equation} (34)\\ \mathit{K}\mathit{y}+\mathit{G}\mathit{u}\ge \mathbf{0}\to \mathrm{Equation} (39)\\ \mathrm{Equations} (40)-(44)\end{array}\right.\end{array}$$

(45)

The outer max problem is a day-ahead bidding decision model with optimization variables $\mathit{x}$, mainly the demand-side resource aggregator’s declared regulation capacity $P_{DR,t}^b$ and the demand-side resource’s expected response power $P_{DR,t}^i$. The inner min-max problem is a prospective trading decision model with optimization variables $\mathit{y}$ and $\mathit{u}$, namely the demand-side resource’s real-time response power $P_{DR,t}^{i,r}$, and the market clearing price $p_{com,t}$ and the demand-side resource’s responsiveness $P_{DR,t}^{i,\max }$, respectively. The iterative resolution of the aforementioned two-stage robust bidding model for demand-side resource aggregators leads to the determination of their optimal trading strategy, i.e., regulated capacity $P_{DR,t}^b$ and regulated price $p_{DR,t}^b$; they can be obtained after the model converges.

5. Algorithmic Analysis

5.1. Basic Data

Henan Province has a large scale of development in the field of energy and a good data foundation. It is easy to obtain demand-side resource data such as electric vehicles, commercial air conditioners, and data centers in the community. The power load in Henan Province is similar to the consumption structure of many regions in China, so the research results have strong universality. In addition, Henan Province actively explores energy transformation, and the research of community demand-side resource aggregation operation model and transaction decision model under this background can well reflect the practical application situation. This section takes a community in Henan Province as a simulation object, which contains 16 electric vehicle users (EV 1–EV 16), 4 commercial air conditioning users (AC 1–AC 4), and 1 data center user (ES), all of which have the ability to interact with demand-side resource aggregators in a bidirectional manner. The data center has a backup energy storage unit with a rated capacity of 100 kW-h and a rated charging and discharging power of 20 kW, with a user load state range of 0.2 to 0.9, which is maintained at 0.6 under normal operation. Each of the parameters for EVs and commercial air conditioners in this community are specifically shown in

Table A1. Basic parameters for electric vehicles and commercial air conditioners.

Electric Vehicle				Commercial Air Conditioning
Serial Number (ev)	Charge/Discharge Power (kw)	Battery Capacity (kw·h)	Charge/Discharge Efficiency	Serial Number (ac)	Efficiency Ratio	Average Thermal Resistance (°C/kw)	Indoor Heating (kw)
1, 2, 3, 4	4	36	0.85/0.90	1	0.90	0.260	84
5, 6, 7, 8, 9	6	45	0.90/0.95	2	0.95	0.283	76
10, 11, 12, 13, 14	5	40	0.95/0.95	3	0.90	0.205	104
15, 16	4	42	0.90/0.90	4	0.92	0.224	92

in the attached Exhibit. It includes the charge/discharge power, battery capacity, and charge/discharge efficiency of each electric vehicle; as well as the energy efficiency ratio, average thermal resistance, and indoor heat source of each commercial air conditioner.

The regulating hours mainly include the valley filling hours: 11:00–12:00 and 12:00–13:00, and the peak shaving hours: 19:00–20:00 and 20:00–21:00. Figure 5 and Figure 6 show historical response data for electric vehicle users and commercial air conditioning users, respectively. In addition, the data center backup storage response capacity is relatively fixed, with response capacities of 20 kW, 13.3 kW, 20 kW, and 8.5 kW for each regulation period, and there is no response power deviation.

The market clearing price levels and deviation penalty prices considered in the demand-side resource cluster aggregation optimization are shown in

Table 2. Market clearing price levels and deviation penalty prices.

Time Period	Clearing Price (¥/kW·h)	Penalty Price (¥/kW·h)
11:00–12:00	0.45	0.90
12:00–13:00	0.36	0.70
19:00–20:00	1.28	2.50
20:00–21:00	0.94	2.00

. In addition, the scheduling agency requires that the duration of the responsive power provided by the demand-side entities participating in the transaction should be no less than 15 min, and the aggregators require that the response rate of the demand-side resource users acting on their behalf should be no more than 30 s in response to their scheduling instructions.

5.2. Analysis of the Results

5.2.1. Cluster Aggregation Result

The index values of response capacity, response fluctuation range, and response fluctuation degree are first calculated for each demand-side resource, as shown in Figure A1, Figure A2 and Figure A3 in Appendix B, respectively. The data center has the highest response capacity; each commercial air conditioner has the next highest response capacity; and each electric vehicle has a relatively low response capacity. The data center was the most reliable in terms of responsiveness, with a response bias of 0. Both commercial air conditioning and EVs had response biases to varying degrees, but EVs had a response bias of 0 from 20:00 to 21:00 during the peak shaving hours. All demand-side resources satisfy the minimum duration, and only EV 8, EV 12, and AC 4 do not satisfy the minimum response rate constraint.

Then, the optimal demand-side resource clusters are determined. For the deviation penalty risk minimization objective function, the response fluctuation range and the response fluctuation degree weight coefficients are 0.3 and 0.7, respectively. For subjective factor weights, demand-side resource aggregators are assumed to focus on minimizing the risk of bias penalties. For objective factor weights, two objective weights are set to increase from 0 to 1 in steps of 0.1 and solved separately to obtain 11 decision schemes, and then the objective factor weights of the above two optimization objectives are calculated according to the entropy weight method. In addition, considering that the demand-side resource aggregators have a bias coefficient of 0.4 for subjective factors, the results of subjective and objective aggregation division are shown in

Table 3. Subjective and objective integration weight division results.

Weighting Methodology	Maximizing Potential Market Revenue	Minimizing the Risk of Bias Penalties
subjective factors	0.3	0.7
objective factor	0.45	0.55
integration of subjective and objective factors	0.39	0.61

.

Finally, the demand-side resources multi-objective aggregation optimization model is solved and the resulting optimal demand-side resource clusters are {EV 1, EV 2, EV 3, EV 5, EV 6, EV 7, EV 10, EV 11, AC 1, AC 2, AC 3, ES}. To analyze the value of different demand-side resources to the aggregator, the cluster aggregation optimization results are shown in Figure 7.

The value order of demand-side resource participation in electricity market trading can be projected when the deviation penalty risk minimization objective weights are varied and different resources are included in clusters. The decision maker’s bias coefficient is increased from 0.4 to 0.8, and the bias penalty risk weight is increased by 0.0582, which reduces the risk by about 29.57% and the benefit by about 18.62%, indicating that the proposed methodology can flexibly adjust the aggregation scheme.

5.2.2. Results of Market Transactions

In this section, based on the results of the cluster aggregation optimization in Section 5.2.1, the aggregators’ regulation capacity and regulation price, i.e., the worst-case market clearing price, are determined for each regulation period on the execution day by solving the robust trading decision model of the demand-side resource aggregators, as shown in

Table 4. Demand-side resource aggregators’ declared regulation capacity and regulation prices for each regulation period.

Time Interval	Adjustment Capacity/(kW·h)		Regulating Prices/(¥/kW·h)	Worst Clearing Price/(¥/kW·h)
valley filling time	11:00–12:00	49.59	0.3549	0.3960
	12:00–13:00	34.09	0.3485	0.3384
peak reduction time	19:00–20:00	101.55	0.9009	1.1264
	20:00–21:00	79.69	0.7511	0.8272

.

As can be seen from Table 4, market clearing prices are high in the 19:00–20:00 and 20:00–21:00 periods, and uncertainty fluctuations have a large impact on the returns of demand-side resource aggregators, whose trading risk considerations focus on these two periods.

It can also be seen that the regulated prices for Time Periods 11:00–12:00, 19:00–20:00, and 20:00–21:00 are lower than the worst-clearing price, and only the regulated price for Time Period 12:00–13:00 is greater than the worst-clearing price. The period between 12:00 and 13:00 is the peak period for electricity consumption, with high pressure on system supply and difficulty in balancing, increasing resource regulation and input costs, while commercial customers’ load shedding affects their operations during this critical period, with high opportunity costs and high compensation required by aggregators. All these factors lead to an increase in the average compensation cost, which in turn leads to a relative increase in the regulation price declared by the aggregator in order to meet the minimum revenue expectation. The expected response power of demand-side resources during each regulation period is shown in Figure 8 and Figure A4.

Commercial air conditioning users have a high percentage of expected response power during peak shaving hours, and zero expected trading power during valley filling hours in AC2 and AC3 due to low compensation prices. The expected response power of electric vehicle users varies by time period. The expected response power of data center users accounts for a high percentage of the total traded power of each responding entity, but the percentage of expected response power is relatively low during valley filling hours.

5.2.3. Comparative Analysis Results

In this section, the following two scenarios are set up to analyze the benefits of demand-side resource aggregators acting as agents for each demand-side resource to participate in the trading of peaking ancillary services.

Scenario 1. 

The aggregators use a two-stage robust trading decision model to consider the worst-case scenario of market clearing prices to develop the regulated capacity and regulated prices of the aggregated response entities;

Scenario 2. 

The aggregators use a conventional bidding model to set the regulated capacity and regulated prices of the aggregated response entities based on market forecasts of clearing prices.

The demand-side resource aggregators’ declared regulation capacity and regulation prices in Scenario 2 are shown in

Table 5. Demand-side resource aggregators’ declared regulation capacity and regulation prices in Scenario 2.

Time Interval		Adjustment Capacity/(kW·h)	Price Adjustment/(¥/kW·h)
Valley Filling Period	11:00–12:00	56.79	0.3802
	12:00–13:00	39.79	0.3656
Peak Reduction Period	19:00–20:00	118.65	0.9783
	20:00–21:00	89.65	0.7908

, with generally higher regulation capacity and prices across regulation periods in Scenario 2 compared to Scenario 1. This is due to the fact that aggregators in Scenario 2 do not take into account market clearing price uncertainty and fully utilize their demand-side resource response potential, resulting in higher marginal compensation costs for the aggregator response entities and a consequent higher adjustment price.

Given that the actual clearing price in the peaking ancillary services market varies uniformly between the worst to the forecast clearing price in Scenario 1, the net market benefits to demand-side resource aggregators are calculated separately for the two scenarios, and, as shown in

Table 6. Market net returns for demand-side resource aggregators in Scenarios 1 and 2 at different clearing prices.

Serial Number	Scenario 1 Net Gain/$				Scenario 2 Net Gain/$
	Slot 12	Slot 13	Slot 20	Slot 21	Slot 12	Slot 13	Slot 20	Slot 21
1	7.35	0.00	41.01	18.72	7.27	0.00	40.58	18.35
2	7.62	0.00	42.58	19.62	7.56	0.00	42.32	19.31
3	7.89	0.00	44.24	20.52	7.85	0.00	44.04	20.28
4	8.16	0.00	45.90	21.41	8.14	0.00	45.78	21.23
5	8.45	0.00	47.56	22.32	8.43	0.00	47.50	22.19
6	8.73	3.67	49.24	23.22	8.72	0.00	49.24	23.15
7	9.02	3.75	50.96	24.13	9.01	0.00	51.00	24.12
8	9.31	3.83	52.67	25.08	9.32	0.00	52.75	25.08
9	9.60	3.91	54.39	26.05	9.62	0.00	54.53	26.06
10	9.90	3.99	56.09	27.04	9.93	0.00	56.33	27.07
11	10.19	4.07	57.81	28.03	10.24	0.00	58.15	28.08

, the net benefits are greater for Scenario 1 than for Scenario 2 when the market clearing price is low. The net return to aggregators in Scenario 2 approaches or even exceeds that of Scenario 1 as the actual market clearing price tapers from the worst to the predicted clearing. This is because the marginal compensation cost of Scenario 2 is high when prices are low, and the excess of winning bids over Scenario 1 is no longer economic but still needs to be completed in order to avoid high penalties, resulting in a lower net return, which is referred to in this section as “over-response”.

Conversely, as the actual clearing price gradually increases, the economics of the responding subject’s winning power bid in Scenario 2 also rise, bringing the aggregators’ net returns gradually closer to Scenario 1, or even exceeding Scenario 1. For example, when at the Serial No. 11 clearing price, the aggregators in Scenario 1 do not fully utilize the response potential of the aggregating body, resulting in the loss of some of the potential economic benefits. This section defines this phenomenon as “under-response”. In addition, the probability of a successful bid is higher in Scenario 1 than in Scenario 2 due to a more conservative filing strategy.

In summary, the two-stage robust trading decision model can assist aggregators in coping with the risk of market clearing price uncertainty, preventing “over-response” losses and increasing the probability of winning the bid, but also bringing potential losses of “under-response”. With the gradual maturity of demand-side subjects participating in electricity market trading, the scheduling agency assessment is more accurate, and the risk aversion effect of the two-stage robust trading decision model is more significant with potential losses. Therefore, aggregators need to be flexible in developing day-ahead bidding strategies based on market rules and price fluctuations.

6. Conclusions

In response to the new power system’s need to tap into the flexible regulatory potential of demand-side resources, this paper introduces a multi-objective clustering aggregation optimization approach for demand-side resources, taking into account both economic efficiency and risk., based on which a robust trading decision model for demand-side resource aggregators is constructed to improve the aggregator’s ability to resist the market trading risk. The validity and sophistication of the proposed model are verified through the arithmetic simulation of a community in Henan Province. The main conclusions are as follows:

(1)

The proposed multi-objective cluster aggregation optimization method comprehensively evaluates the response performance of demand-side resources by considering the three dimensions of response capacity, response reliability, and response flexibility. The use of the subjective–objective integrated empowerment method reduces the demand-side resource aggregator’s deviation penalty risk by about 33.12% and improves the comprehensive optimization objective by about 18.10%, which reasonably balances the market returns and risk preferences;

(2)

The aggregator’s robust trading decision model is able to increase net revenue by about 3.1% under the “worst” scenario of fluctuating uncertainties, which can help demand-side resource aggregators cope with the market trading risks brought by uncertainties, and comparative analyses show that the proposed model can help demand-side resource aggregators reduce The comparative analysis shows that the proposed model can help demand-side resource aggregators reduce over-response losses and increase the chances of winning bids, but it may also bring the risk of under-response, which requires flexibility in formulating response strategies. Due to the robustness coefficient and adjustment parameters considered in this model, the uncertainty of price fluctuations can be quantified. So, the same applies to larger dynamic communities, considering the price fluctuations due to weather conditions and grid demand patterns;

(3)

The multi-objective cluster aggregation optimization method and robust trading decision-making model proposed in this paper enhance the ability of demand-side resources to take part in electricity market trading, which is conducive to demand-side resource aggregators to make more accurate and economical trading decisions in the complex and volatile electricity market, and enhance the aggregators’ ability to cope with the potential risks in the market.

In terms of policy prospects, in the process of power market reform, policies will focus on demand-side main body trading norms. In view of the advantages and disadvantages of the two-stage robust transaction decision model, future policies will refine market rules, clarify the rights and responsibilities of aggregators, introduce reward and punishment or compensation mechanisms to deal with “over-response” and “under-response”, and strengthen supervision over the decision-making process of aggregators, so as to ensure market stability and efficiency and achieve optimal allocation of resources.

In the future work direction, the strategy optimization should be based on the market rules to refine the day before the bidding strategy, the construction of price monitoring and analysis system to cope with price fluctuations. Technology research and development need to improve the innovation model, integrate more factors, improve operational efficiency, and enhance adaptability with intelligent technology. Cooperation and communication should strengthen coordination with dispatching agencies, share information, participate in testing, and promote the exchange of experience and joint research as well as the development of aggregators in the industry, so as to promote the overall level of transaction decision-making and sustainable development of the industry.