Virtual power plant scheduling and optimization method based on multiple game relationships among users

Abstract

The invention relates to the technical field of virtual power plant resource scheduling, and discloses a virtual power plant scheduling and optimization method based on multiple game relationships among users, comprising the following steps: step 1, user characteristic analysis and similarity description; step 2, user similarity calculation; 3, building a game model between users; 4, responding to an income expectation model; 5, optimizing the target and the constraint; and 6, solving and optimizing the model. According to the method, the enhancement effect of cooperation on user response is quantified by introducing the cooperation correction factor. Through combination of the similarity and the response level, the response capability of user cooperation is dynamically adjusted by the correction factor, so that the model can reflect the gain effect under the real cooperation condition, and the beneficial effects of optimizing the overall scheduling effect and realizing the more efficient response capability of the virtual power plant are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual power plant resource scheduling, and particularly to a virtual power plant scheduling and optimization method based on various game relationships among users. Background Technique

[0002] With the development of the flexible and economic operation mode of microgrids, virtual power plants have gradually received wide attention because they can effectively address the security and reliability challenges brought about by the large-scale access of distributed energy to the power grid. By integrating a large number of dispersed energy resources and participating in demand response, virtual power plants not only improve the power system's ability to balance power supply and demand but also significantly relieve the pressure of energy investment. To achieve effective response strategies in a complex and changing environment, it is particularly important to formulate targeted virtual power plant scheduling strategies.

[0003] In traditional virtual power plant scheduling, participating users usually respond independently and are not affected by other users. However, the relationships among users are often intricate, and they may adopt various participation methods such as negotiation and cooperation or non-cooperation, which have important impacts on response strategies. Fully considering the game relationships among users and their impacts can greatly improve the accuracy of response evaluation and thus enhance the effectiveness of virtual power plant demand response. Summary of the Invention

[0004] Technical problems to be solved:

[0005] Aiming at the deficiencies of the prior art, the present invention provides a virtual power plant scheduling and optimization method based on various game relationships among users, which has the advantages of optimizing the overall scheduling effect, achieving a more efficient virtual power plant response ability, and realizing innovative optimization of the virtual power plant response strategy through user characteristic analysis, game modeling, and optimization solution, thus solving the above technical problems.

[0006] Technical solution:

[0007] To achieve the above object, the present invention provides the following technical solution: A virtual power plant scheduling and optimization method based on various game relationships among users, comprising the following steps:

[0008] Step 1. User characteristic analysis and similarity characterization: The characteristics include numerical characteristics and categorical characteristics. The numerical characteristics are: normalizing historical load data and geographical location information, and extracting load data, coefficient of variation, and average load during the demand period. The categorical characteristics are: industry attribute, historical effective response situation, and enterprise scale. All characteristics are concatenated into a comprehensive feature vector;

[0009] The industry attribute is one-hot encoded;

[0010] The historical effective response situation is to calculate the probability distribution of effective response;

[0011] The enterprise scale is label - encoded and divided into three categories: large, medium, and small.

[0012] Step 2: User similarity calculation: Use KL - divergence to calculate the difference in the effective response probability distribution between users, use Euclidean distance to calculate the difference in other features, and combine KL - divergence and Euclidean distance to obtain the comprehensive similarity between users.

[0013] Step 3: Game model between users: According to the similarity between users, calculate the probability of user cooperation. When the cooperation probability is higher than a specific value, users cooperate; otherwise, users participate independently.

[0014] Step 4: Response revenue expectation model: Calculate the revenue of independent users participating in demand response; for cooperative user pairs, introduce a cooperation correction factor, consider the additional revenue when users cooperate, and calculate the total expected response capacity and revenue of cooperative user pairs.

[0015] Step 5: Optimization objectives and constraints: The objective is to maximize the total expected revenue of the virtual power plant. The constraint conditions are the total demand capacity constraint: the sum of the response capacities of all users is equal to the total demand response volume; the non - negative constraint: the response capacity of users is non - negative.

[0016] Step 6: Model solution and optimization: Use the Cplex optimization solver to find the optimal capacity allocation strategy and improve the effectiveness and revenue level of the demand response of the virtual power plant.

[0017] Preferably, for feature extraction in Step 1, for numerical features, including historical load data and geographical location information, first normalize the data. The geographical location information uses the normalized data as features lo i , la i , which are the longitude and latitude of user i respectively. For historical load, extract the load data corresponding to the demand period [T 1 , T 2 . According to the demand date, filter to obtain the load data sequence of the corresponding demand period of similar days. Similar days are based on the same day of the week. Calculate the coefficient of variation and average load of each period, denoted as Va i , Pa i ;

[0018] For the categorical feature of industry attribute, perform one - hot encoding. Suppose there are a total of M industry types among n users, and the industry feature of user i is an M - dimensional vector z i = [0, …, 0, 1 i , 0, …, 0], where 1 i = 1, indicating that the i - th element is 1.

[0019] Preferably, for the historical effective response situation of categorical features in the first step, calculate the effective response probability distribution, and denote the user's effective response ratio as the random variable X i , calculate the effective response ratio based on the user's historical declared capacity and effective response capacity in the following manner:

[0020] Declared capacity a i Greater than the effective response capacity b i ,

[0021] Declared capacity a i Less than or equal to the effective response capacity b i , the effective response probability is c i = 1;

[0022] Divide the interval [0, 1] into L intervals Based on the historical response data of user i, count the effective response times in each interval for each response. When L is large, it is reasonable to consider each interval as a constant, that is, in the interval The random constant variable is equal to this interval c i The average value of the effective response probability. Let the effective response times in each interval be N i,k , k = 0, …, L - 1, representing the number of times of user i in interval k, and denote the total number of times as

[0023] Calculate the effective response distribution Pr of user i based on the empirical distribution i As:

[0024]

[0025] For the enterprise scale, divide it into three categories of large, medium, and small according to the scale size, and then numerically encode the scale using label encoding. The scale feature of user i is a 3D vector s i = [0, …, 0, 1 i , 0, …, 0], where 1 i = 1, indicating that the i-th element is 1.

[0026] Preferably, for the numerical features in the first step, including historical load data and geographical location information, first normalize the data. The geographical location information is denoted as features lo i , la i , which are the longitude and latitude of user i respectively. For the historical load, extract the load data corresponding to the demand period [T 1 , T 2 , filter to obtain the load data sequence corresponding to the similar days in the corresponding demand period according to the demand date. The similar days are based on the same day of the week, and calculate the coefficient of variation and average load of each period, denoted as Vai , Pa i ;

[0027] In the first step, the categorical feature industry attribute is one-hot encoded. Suppose there are a total of M industry types among n users, and the industry feature of user i is an M-dimensional vector z i = [0, …, 0, 1 i , 0, …, 0], where 1 i = 1, indicating that the i-th element is 1.

[0028] Preferably, the calculation of user similarity characterization is as follows: For the effective response probability function, the KL divergence between each user is calculated, and the divergence is used to quantify the difference index of the effectiveness of user participation in the response. For user i and user j, the divergence is calculated as:

[0029]

[0030] For other features, after being transformed into numerical features as above, they are fused into a multi-dimensional feature vector, denoted as [lo i , la i , Va i , Pa i , s i , z i . The Euclidean distance is used to calculate the difference between users. For user i and user j, it is denoted as R i,j ;

[0031] Calculate the overall similarity, KL max = max i,j KL i,j , R max = max i,j R i,j The similarity between user i and user j is calculated by the following formula

[0032] Preferably, the construction of the user game model in the third step is as follows:

[0033] First, a cooperation probability mapping is established based on the similarity. For the similarity S i,j between user i and user j, it is mapped to the cooperation probability P i,j . The mapping function is constructed as follows

[0034]

[0035] where γ is the smoothing parameter and θ is the cooperation similarity threshold. When P i,j is higher than a specific value, users cooperate with each other; when it is lower than the threshold, users participate independently. To simplify the model, each user only cooperates with the user with the highest cooperation probability.

[0036] Preferably, the response revenue expectation model is constructed as follows:

[0037] Before the aggregator considers cooperation, the effective response distributions of user i and user j are Pr i and Pr j respectively. After the scheduling metrics Q i and Q j are sent to the users, the aggregator evaluates the expected effective responses of each user based on the distributions Pr i and Pr j . The total effective response capacity evaluation is Q i Pr i +Q j Pr j . When the aggregator considers two cooperative users with better response guarantees, a new strategy is needed to evaluate the expected effective response capacities of user i and user j. First, a cooperation correction factor ρ i,j is introduced to quantify the enhancement effect of cooperation on the effective responses of the two users, which is characterized by the similarity and cooperation tightness between users. Specifically:

[0038] ρ i,j =∈P i,j (1 - exp(-β(Q (1 - exp(-β(Q i +Q j )))

[0039] where ∈ and β are adjustment coefficients, reflecting the enhancement degree of cooperation on the response. Q i +Q j is the response level between users. The higher the response level, the larger the response of the correction factor, indicating that the cooperation has a greater effect on the capacity response. Based on the correction factor, the expected response capacities of the two users are corrected as follows (Q i Pr i +Q j Pr j )(ρ i,j +1), obtaining the set C of cooperative user pairs and the set S of independent responding users.

[0040] Preferably, the calculation expression of the total expected revenue in step four is as follows

[0041] Set C of cooperative user pairs:

[0042] ∑ (i,j)∈C (Q i Pr i +Q j Pr j )(ρ i,j +1)p;

[0043] Set S of independent responding users:

[0044] ∑ k∈S Q k Pr k p。

[0045] Preferably, after the construction and analysis of the total expected revenue calculation model, the scheduling objective function is obtained as follows

[0046] maxH(Q 1 ,…,Q n )=∑ (i,j)∈C (Q i Pr i +Q j Pr j )(ρ i,j +1)p+∑ k∈S Q k Pr k p。

[0047] Preferably, the constraint conditions are as follows:

[0048] Total demand capacity constraint: Q 1 +…+Q n =Q;

[0049] Non - negative constraint: Q i ≥0,i=1,…,n。

[0050] Compared with the prior art, the present invention provides a virtual power plant scheduling and optimization method based on various game relationships among users, having the following beneficial effects:

[0051] 1. By introducing a cooperative mapping function of the game among users, the present invention constructs a cooperative probability mapping function based on similarity. Through the setting of the smoothing parameter and the cooperative similarity threshold, it can better judge the cooperative relationship among users, select that each user only cooperates with the user with the highest cooperative probability, simplifies the construction of the game model, reduces the computational complexity, and effectively reflects the gain effect of user cooperation in actual response. The cooperation correction factor is introduced to quantify the enhancement effect of cooperation on user response. Through the combination of similarity and response level, the correction factor dynamically adjusts the response ability of user cooperation, enabling the model to reflect the gain effect under real cooperation conditions, enhancing the reality and accuracy of the evaluation. A revenue expectation model including cooperative and independent users is constructed. By calculating the corrected expected response capacity, the revenue gain brought by cooperation is reasonably evaluated. The calculation of the total revenue not only considers the individual response ability of users but also includes the revenue enhanced by cooperation, making the revenue evaluation more comprehensive and three - dimensional, achieving the beneficial effect of optimizing the overall scheduling effect and realizing a more efficient virtual power plant response ability.

[0052] 2. By deeply mining and integrating user characteristics, such as geographical location information, historical load data, industry attributes, historical effective response situations, enterprise scale, etc., the present invention constructs a multi-dimensional comprehensive feature vector, and combines methods such as KL divergence and Euclidean distance to finely characterize user similarity, which helps to better evaluate the potential cooperation relationship between users, achieving the beneficial effect of innovatively optimizing the response strategy of the virtual power plant through user characteristic analysis, game modeling, and optimization solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the step flow of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0055] A virtual power plant scheduling and optimization method based on various game relationships between users includes the following steps:

[0056] Step 1. User characteristic analysis and similarity characterization: The characteristics include numerical features and categorical features. The numerical features are: normalizing the historical load data and geographical location information, and extracting the load data, load variation coefficient, and average load during the demand period. The categorical features are: industry attributes, historical effective response situations, and enterprise scale. All the features are concatenated into a comprehensive feature vector;

[0057] The industry attribute is one-hot encoded;

[0058] The historical effective response situation is to calculate the effective response probability distribution;

[0059] The enterprise scale is label encoded and divided into three categories: large, medium, and small;

[0060] Step 2. User similarity calculation: Use KL divergence to calculate the difference in the effective response probability distribution between users, use Euclidean distance to calculate the difference in other features, and combine KL divergence and Euclidean distance to obtain the comprehensive similarity between users;

[0061] Step 3. Game model between users: According to the similarity between users, calculate the probability of user cooperation. When the cooperation probability is higher than a specific value, the users cooperate; otherwise, the users participate independently;

[0062] Step 4, Respond to the revenue expectation model: Calculate the revenue of independent users participating in demand response; for cooperative user pairs, introduce a cooperation correction factor, consider the additional revenue when users cooperate, and calculate the total expected response capacity and revenue of the cooperative user pairs.

[0063] Step 5, Optimization objectives and constraints: The objective is to maximize the total expected revenue of the virtual power plant, and the constraint conditions are the total demand capacity constraint: the sum of the response capacities of all users is equal to the total demand response volume; the non-negativity constraint: the response capacity of users is non-negative.

[0064] Step 6, Model solution and optimization: Use the Cplex optimization solver to find the optimal capacity allocation strategy to improve the effectiveness and revenue level of the demand response of the virtual power plant.

[0065] Example 1

[0066] Please refer to Figure 1

[0067] The relevant data of three users are as follows:

[0068]

[0069] 1.1 Calculate the effective response ratio ci:

[0070] For user 1:

[0071]

[0072] For user 2:

[0073]

[0074] For user 3:

[0075] c 3 = 1;

[0076] 1.2 Calculate the effective response distribution Pr i :

[0077] For the partition interval with L = 4, calculate the effective response distribution:

[0078]

[0079] 1.3 Calculate the KL divergence between users

[0080] Considering the effective response distributions of user 1 and user 2, calculate the KL divergence:

[0081]

[0082] Example 2

[0083] The data of another group of users are as follows:

[0084]

[0085] 2.1 Calculate the effective response ratio c i :

[0086] For user 4:

[0087]

[0088] For user 5:

[0089]

[0090] For user 6:

[0091] c 6 = 1

[0092] 2.2 Calculate the effective response distribution Continue to use the divided interval L = 4 to calculate the effective response distribution:

[0093]

[0094] 2.3 Calculate the KL divergence between users Calculate the KL divergence between user 4 and user 5:

[0095]

[0096] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A virtual power plant scheduling and optimization method based on multiple game relationships between users, characterized in that: The following steps are involved: Step 1: User characteristic analysis and similarity characterization: The features include numerical features and categorical features. The numerical features are: normalizing historical load data and geographic location information, extracting load data, coefficient of variation and average load during the demand period, and categorical features are: industry attributes, historical effective response and enterprise scale. All features are spliced ​​into a comprehensive feature vector. The industry attributes are one-hot encoded; The historical effective response situation is to calculate the effective response probability distribution; The enterprise size is coded by the label and is divided into three categories: large, medium and small; Step 2: User similarity calculation: Use KL divergence to calculate the difference in the probability distribution of effective responses between users, use Euclidean distance to calculate the difference in other features, and combine KL divergence and Euclidean distance to obtain the comprehensive similarity between users; Step 3: Game model between users: Calculate the probability of user cooperation based on the similarity between users. When the probability of cooperation is higher than a certain value, users cooperate. Otherwise, users participate independently; Step 4: Response benefit expectation model: Calculate the benefits of independent users participating in demand response; introduce cooperation correction factors for cooperative user pairs, consider the additional benefits of user cooperation, and calculate the total expected response capacity and benefits of cooperative user pairs; Step 5. Optimization objectives and constraints: The objective is to maximize the total expected revenue of the virtual power plant. The constraints are the total demand capacity constraint: the sum of the response capacities of all users is equal to the total demand response; non-negative constraint: the user's response capacity is non-negative; Step 6: Model solving and optimization: Use the Cplex optimization solver to find the optimal capacity allocation strategy to improve the demand response effectiveness and profitability of the virtual power plant.

2. According to claim 1, a virtual power plant scheduling and optimization method based on multiple game relationships between users is characterized by: In the feature extraction of step 1, for numerical features, including historical load data and geographic location information, the data is first normalized, and the geographic location information uses the normalized data as the feature lo i ,la i , are the longitude and latitude of user i, respectively. For historical load, extract the load data of the corresponding demand period [T1, T2]. According to the demand date, filter to obtain the load data sequence of the corresponding demand period on similar days. Similar days are based on the same week date. Calculate the coefficient of variation and average load of each period, respectively, denoted as Va i ,Pa i ; For the categorical feature industry attributes, one-hot encoding is performed. Suppose there are M industry types among n users, and the industry feature of user i is the M-dimensional vector z i =[0,…,0,1 i ,0,…,0], where 1 i =1, indicating that the i-th element is 1.

3. The virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 2 is characterized by: In the step 1, for the historical effective response of the categorical feature, the effective response probability distribution is calculated, and the effective response ratio of the user is recorded as the random variable X i , based on the user's historical declared capacity and effective response capacity, the effective response ratio is calculated as follows: Declared capacitya i Greater than the effective response capacity b i , Declared capacitya i Less than or equal to the effective response capacity b i , the probability of effective response is c i =1; Divide the interval [0,1] into L intervals Based on the historical response data of user i, the number of valid responses in each interval is counted. When L is large, it is reasonable to regard each interval as a constant, that is, in the interval The random constant variable is equal to this interval c i The average value of the probability of effective response, assuming the number of effective responses in each interval is N i,k , k=0,…,L-1, represents the number of times user i is in interval k, and the total number is Calculate the effective response distribution Pr of user i based on the empirical distribution i for: For enterprise scale, we divide it into three categories: large, medium and small. Then we use label coding to digitize the scale. The scale feature of user i is a 3D vector s i =[0,…,0,1 i ,0,…,0], where 1 i =1, indicating that the i-th element is 1.

4. A virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 3, characterized in that: The numerical features in step 1 include historical load data and geographic location information. The data is first normalized, and the geographic location information uses the normalized data as the feature lo i ,la i , are the longitude and latitude of user i, respectively. For historical load, extract the load data of the corresponding demand period [T1, T2]. According to the demand date, filter to obtain the load data sequence of the corresponding demand period on similar days. Similar days are based on the same week date. Calculate the coefficient of variation and average load of each period, respectively, denoted as Va i ,Pa i ; The category-type characteristic industry attribute in step 1 is processed by one-hot encoding. Suppose there are M industry types in total among n users, and the industry characteristic of user i is M latitude vector z i =[0,…,0,1 i ,0,…,0], where 1 i =1, indicating that the i-th element is 1.

5. A virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 4, characterized in that: The user similarity characterization calculation is as follows: for the effective response probability function, the KL divergence between each user is calculated, and the difference index of the effectiveness of user participation response is quantified by the divergence. For user i and user j, the divergence is calculated as: For other features, after being converted into numerical features above, they are fused into a multidimensional feature vector, denoted as [lo i ,la i ,Va i ,Pa i ,s i ,z i ]. Using the Euclidean example to calculate the difference between users, for user i and user j, denoted by R i,j ; Calculate the overall similarity, KL max =max i,j KL i,j ,R max =max i,j R i,j , the similarity between user i and user j is calculated by the following formula 6. A virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 1, characterized in that: The game model between users in step 3 is constructed as follows: First, a cooperation probability mapping is established based on similarity. For the similarity S between user i and user j, i,j , mapped to cooperation probability P i,j , construct the mapping function as follows Where γ is the smoothing parameter, θ is the cooperation similarity threshold, when P i,j When the value is above a certain level, users cooperate with each other, and when the value is below the threshold, users participate independently. To simplify the model, each user only cooperates with the user with the highest probability of cooperation.

7. A virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 6, characterized in that: The response return expectation model is constructed as: Before the aggregator does not consider cooperation, the effective response distributions of user i and user j are Pr i With Pr j , send the scheduling indicator Q to the user i With Q j After that, the aggregator distributes Pr i With Pr j The expected effective response of each user is evaluated, and the total effective response capacity evaluation is Q i Pr i +Q j Pr j , the aggregator considers the existence of two cooperative users, which has better response guarantee. A new strategy is needed to evaluate the expected effective response capacity of user i and user j. First, the cooperation correction factor ρ is introduced i,j , which is used to quantify the enhancement effect of cooperation on the effective response of two users, and is characterized based on the similarity between users and the closeness of cooperation, specifically: ρ i,j =∈P i,j (1-exp(-β(Q i +Q j ))) Where ∈, β is the adjustment coefficient, reflecting the degree of enhancement of cooperation on response. i +Q j is the response level between users. The higher the response level, the larger the correction factor response is, indicating that the greater the effect of cooperation on capacity response. Based on the correction factor, the expected response capacity of the two users is corrected as follows (Q i Pr i +Q j Pr j )(ρ i,j +1), and obtain the cooperative user pair set C and the independent response user set S.

8. The virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 7 is characterized by: The total expected return calculation expression in step 4 is as follows: Collaborative user pair set C: (i,j)∈C (Q i Pr i +Q j Pr j )(ρ i,j +1)p;​ The set of independent responding users S: Σ k∈S Q k Pr k p。 9. A virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 1, characterized in that: The Cplex model calculation in step 6 is: maxH(Q1,…,Q n ) (i,j)∈C (Q i Pr i +Q j Pr j )(ρ i,j +1)p+∑ k∈S Q k Pr k p。 10. The virtual power plant scheduling and optimization method based on multiple game relationships between users according to claim 1 is characterized by: The Cplex model calculation constraints are: Total demand capacity constraint: Q1+…+Q n =Q; Non-negative constraint: Q i ≥0,i=1,…,n.

Images