Virtual power plant aggregation user optimization system and method considering multiple types of users

Abstract

The invention discloses a virtual power plant aggregation user optimization system and method considering multiple types of users. Carrying out clustering analysis on the load curves of the residential users through a clustering algorithm, determining power utilization habits of different residential users, and constructing a multi-type user optimization index system; a cloud model is adopted to generate the membership degree of each index in the reference cloud; introducing an uncertain focal element and converting the uncertain focal element into basic probability distribution in an evidence theory; aiming at the problem that strong conflict evidence fusion may generate paradox in the traditional evidence theory, conflict evidences are identified according to a combined weight, conflict correction and fusion are carried out, and the closeness between basic probability distribution of each user and positive and negative ideal clouds is compared to determine the aggregation priority of the user; according to the method, a user optimization index system adapting to the characteristics of the virtual power plant is constructed, comprehensive evaluation is carried out through the cloud model and the improved evidence theory, and the randomness, fuzziness and uncertainty existing in virtual power plant user optimization can be effectively solved at the same time.

Description

Technical Field

[0001] The present invention relates to the field of virtual power plants, and in particular to a virtual power plant user optimization system and method that considers multiple types of users. Background Art

[0002] With the deepening of energy transformation strategies such as the "dual carbon" goal and the construction of a new power system with new energy as the main body, new things such as virtual power plants, load aggregators, and demand-side resources continue to emerge. However, independent demand-side resources are difficult to meet the access conditions of the power market and power system dispatch due to their many points, small quantity, wide range, and heterogeneity. With its strong aggregation ability, the virtual power plant can aggregate massive demand-side resources to form a huge load integrator, acting as an agent for users to participate in power trading and power dispatch. Moreover, as a demand-side load-type virtual power plant, one of its main uses is to participate in system peak-shaving auxiliary services, while some users themselves are not suitable for peak-shaving or have relatively weak peak-shaving capabilities. Therefore, in the face of many uneven load users to be aggregated, how to evaluate their peak-shaving performance and aggregate optimization is one of the important practical problems that need to be solved urgently.

[0003] In the context of large-scale demand-side resource access to the power grid, in order to improve the accuracy of user peak-shaving performance identification of virtual power plants under massive heterogeneous load environments, it is urgent to propose a virtual power plant user optimization system and method that considers multiple types of users. Summary of the invention

[0004] In order to improve the accuracy of identifying user peak-shaving performance of virtual power plants under massive heterogeneous load environments, the present invention proposes a virtual power plant user optimization system and method that considers multiple types of users.

[0005] The solution adopted by the present invention to solve the technical problem is:

[0006] A virtual power plant user optimization system and method considering multiple types of users, comprising the following steps:

[0007] Step 1: Taking the evaluation set {level 1, level 2, level 3, level 4, level 5} of the optimal evaluation as the identification framework, the cloud models corresponding to each indicator are established respectively, and the membership of each indicator value at different levels is obtained, that is, the membership matrix;

[0008] Step 2: Integrate the dynamic weight and static weight of the evidence to obtain the combined weight, then determine the conflicting evidence based on the combined weight, and modify its basic probability distribution function through the discount factor;

[0009] Step 3, identifying conflicting evidence by combining weights and correcting only the conflicting evidence;

[0010] Step 4, drawing on the basic idea of ​​TOPSIS, determines the user's preference order by calculating the closeness between the basic probability distribution function formed by each user after evidence fusion and the basic probability distribution function corresponding to the positive and negative ideal solutions.

[0011] Furthermore, in step 4, if there are multiple decision objects belonging to the same level and these decision objects cannot be further compared, the optimal and worst values ​​of each indicator among all users are selected in combination with the indicator attributes, and substituted into the cloud model to solve the membership matrix of the positive ideal cloud and the negative ideal cloud, and then the aforementioned evidence theory fusion method is used to obtain the probability distribution of the positive and negative ideal clouds after evidence fusion, which represent the most ideal solution and the least ideal solution, respectively.

[0012] Preferably, in step 3, assuming that there are o groups of evidence, the average weight 1 / o assigned to each piece of evidence is used as the threshold, and when the combined weight Wi ≥ 1 / o, the evidence is considered to be non-conflicting evidence and should be retained. When the combined weight Wi < 1 / o, the evidence is considered to be conflicting evidence and should be revised.

[0013] The step 2 is based on the combined weighting method of game theory, which integrates the dynamic weight and static weight of the evidence to obtain the combined weight, and then determines the conflicting evidence based on the combined weight, and corrects its basic probability distribution function through the discount factor. This method not only fully considers the importance difference between the evidences, but also only corrects the conflicting evidence, which can retain the original information as much as possible, identify the correct focal element more efficiently, and thus improve the accuracy of evidence fusion.

[0014] Furthermore, combined with the indicator attributes, the optimal and worst values ​​of each indicator among all users are selected and substituted into the cloud model to obtain the membership matrix of the positive ideal cloud and the negative ideal cloud. Then, the aforementioned evidence theory fusion method is used to obtain the probability distribution of the positive and negative ideal clouds after evidence fusion, which represent the most ideal solution and the least ideal solution, respectively.

[0015] Furthermore, the closeness of each user's basic probability distribution function formed after evidence fusion is calculated with the basic probability distribution function of the positive and negative ideal clouds, and the average of the user's fit is calculated. Based on the average fit, the difference between the average fit of each user and the maximum and minimum average fit is calculated. The larger the difference, the closer the user is to the most ideal solution, and the users are ranked accordingly.

[0016] The technical solution of the present invention has the following advantages:

[0017] A. In terms of evaluation optimization methods, the cloud model, as a qualitative and quantitative uncertainty transformation model, not only replaces the traditional membership generation function, but also can effectively take into account the randomness and ambiguity existing in the evaluation problem, greatly improving the credibility of the evaluation results.

[0018] B. The evidence theory adopted in the present invention can not only fuse the membership of multi-dimensional user indicators for decision making due to its information fusion capability, but also effectively resolve uncertainty through uncertainty focal elements.

[0019] C. Based on the differences in peak-shaving performance among users of virtual power plants, the present invention designs a virtual power plant aggregated user optimization evaluation index system from four dimensions: overall load level, peak load period, late load reduction period, and daily load fluctuation rate. The membership value between the sample to be tested and the benchmark cloud is calculated through a cloud model, which realizes the mutual conversion between quantitative values ​​and qualitative concepts, and overcomes the ambiguity and randomness of traditional membership functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the specific embodiments of the present invention, the drawings required for use in the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 It is a schematic flow chart of the method of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0023] In the present invention, all embodiments, implementations and features of the present invention can be combined with each other without contradiction or conflict. In the present invention, conventional equipment, devices, components, etc. can be purchased commercially or made according to the contents disclosed in the present invention. In the present invention, in order to highlight the key points of the present invention, some conventional operations and equipment, devices, components are omitted or only briefly described.

[0024] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0025] Refer to the following Figure 1 The technical solutions of some embodiments of the present invention are described.

[0026] Classify and analyze the residential load data, that is, cluster the load curves of residential users through clustering algorithms to find the patterns of electricity consumption, and then conduct a refined analysis of the corresponding electricity consumption characteristics to determine the electricity consumption habits of different residential users;

[0027] Determine user categories from four dimensions: overall load level, peak load period, and evening load reduction period, and build a multi-type user optimization index system;

[0028] Aiming at the randomness and fuzziness of users at the boundary of the preferred level, a cloud model is used to generate the membership of each indicator in the benchmark cloud;

[0029] The uncertainty focal element is introduced to transform it into the basic probability distribution in the theory of evidence;

[0030] Referring to the idea of ​​game theory, the static and dynamic weights of evidence are combined in a game, and conflicting evidence is identified according to the combined weights, and conflict correction and integration are performed;

[0031] Drawing on the idea of ​​approximating the ideal solution sorting technique, the closeness of each user's basic probability distribution to the positive and negative ideal clouds is compared to determine the user's aggregation priority;

[0032] Furthermore, before classifying and analyzing the residents' load data, load data preprocessing must be performed, including data cleaning and data normalization.

[0033] The data cleaning is to clean the collected daily load data, mainly using Newton interpolation to fill in the missing data. It is known that there are n points (x1, f(x1)), (x2, f(x2)), ... (x n ,f(x n )) with the missing point being ((x i ,f(x i ))), where x∈R, i∈[1,n], then the interpolation polynomial is:

[0034] f(x i )=f(x1)+(x i -x1)f[x2,x1]+(x i -x1)(x i -x2)f[x3,x2,x1]

[0035] +…+(x i -x1)(x i -x2)…(x i -x n-1 )f[x n ,x n-1 ,…,x2,x1]

[0036] +(x i -x1)(x i -x2)…(x i -x n )f[x n ,x n-1 ,…,x1,x i ]

[0037] =P(x i )+R(x i )

[0038] Among them, f(x i ) is the missing value obtained by Newton interpolation, P(x i ) is the Newton interpolation approximation function, R(x i ) is the error function.

[0039] P(x i )=f(x1)+(x i -x1)f[x2,x1]+(x i -x1)(x i -x2)f[x3,x2,x1]

[0040] +…+(X i -X1)(X i -X2)…(X i -X n-1 )f[X n ,X n-1 ,…,X2,X1]

[0041] R(x i )=(x i -x1)(x i -x2)…(x i -x n )f[x n ,x n-1 ,…,x1,x i ]

[0042] The data normalization is to normalize the cleaned data. The linear normalization method is used to convert the value of the original data to between [0, 1]. The formula used is as follows:

[0043] p′i =(p i -min(p)) / (max(p)-min(p))

[0044] Where p is the original data and p' is the normalized data.

[0045] Furthermore, the preprocessed load data are clustered using a two-step method, which includes a pre-clustering stage and a clustering stage. When the two-step clustering method is used, the clustering criterion is based on the Schwarz Bayesian criterion, also known as the Bayesian Information Criterion (BIC), and the distance measurement uses the log-likelihood distance.

[0046] In the pre-clustering stage, the idea of ​​CF (Clustering Feature) tree growth in the BIRCH (Balanced Iterative Reducing and Clustering Using Hierarchies) algorithm is adopted to read data points in the data set one by one. While generating the CF tree, the data points in the dense area are pre-clustered to form several sub-clusters.

[0047] The clustering stage takes the result of the pre-clustering stage, the sub-clusters, as the object, and uses the agglomerative method (i.e., treating each sub-cluster as a cluster and then repeatedly merging the two closest clusters therein) to merge the sub-clusters one by one until the desired number of clusters is reached.

[0048] When clustering, the Bayesian Information Criterion is mainly used to find the approximate range of the optimal number of clusters. The BIC criterion is defined as:

[0049] BIC=-2ln(L)+ln(h)·Y

[0050] Among them, L represents the maximum likelihood function value in the model; h is the sample size of the model; and Y represents the number of unknown parameters in the model.

[0051] Log-likelihood distance is a method based on statistical theory to calculate the difference between clusters. Assume that the current data object is divided into J clusters: C1, ..., C J If C s and C t (C t (s,t∈{1,…,J})) two clusters are merged into one cluster, denoted as C ＜s,t＞ , define cluster C s and C t The log-likelihood distance between is as follows:

[0052] d(C s ,Ct )=ζ s +ζ t -ζ s,t>

[0053] Among them, ξ is the specific form of the log-likelihood function, which is defined as follows:

[0054]

[0055] Among them, K A and K B are numerical variables and categorical variables, respectively. and are the total variance of the k-th numerical variable and the variance in the v-th class, respectively. ν and N νkl are the sample size of the vth class, and the sample size of the lth class for the kth categorical variable in the vth class. k Introduced in the formula The purpose is to solve the problem that the variance of the vth category may be 0 and the logarithm cannot be calculated.

[0056] Furthermore, the user categories are determined based on four characteristic indicators: overall load level, peak power load period, evening load reduction period, and daily load fluctuation rate.

[0057] (1) Overall load level:

[0058]

[0059] Among them, P level Represents the overall load level; P averoge is the average daily load; P max is the maximum load in all sampled data. According to the calculation results in the formula, it is divided into five categories: relatively high level (greater than or equal to 50%), high level (greater than or equal to 40%, less than 50%), medium level (greater than or equal to 20%, less than 40%), low level (greater than or equal to 10%, less than 20%) and very low level (less than 10%).

[0060] (2) Peak electricity consumption period:

[0061] It is mainly divided into four time periods: 5:00-6:00 in the morning, 11:00-12:00 in the morning, 19:00-20:00 in the evening, and 20:00-21:00 in the evening.

[0062] (3) Late load reduction period:

[0063] The time is divided into two categories: 21:00 and 22:00.

[0064] (4) Daily load fluctuation rate:

[0065]

[0066] Among them, P wave represents the daily load fluctuation rate; P error is the standard deviation of daily load; P average Daily average load: According to the calculation results of the above formula, it is divided into two categories: with fluctuation (30% and above) and without fluctuation (below 30%).

[0067] Based on the above characteristic indicators, the user category can be determined according to the following rules.

[0068] Rule 1: Office workers + children combined household

[0069] Characteristic index (1) is at a high level; characteristic index (2) exists in two time periods (11:00-12:00 and 19:00-20:00); characteristic index (3) is around 21:00; characteristic index (4) is fluctuating.

[0070] Rule 2: Office workers + elderly residents

[0071] Characteristic index (1) is at a medium level; characteristic indexes (2) and (3) are the same as those in rule (b); characteristic index (4) is non-volatile.

[0072] Rule 3: Elderly Households

[0073] The characteristic index (1) is at a low level; the characteristic index (2) exists in two periods (5:00-6:00 and 20:00-21:00); the characteristic index (3) is around 21:00; the characteristic index (4) is without fluctuation.

[0074] Rule 4: Single working tenants

[0075] Characteristic index (1) is at an extremely low level; characteristic index (2) only exists in one period (from 20:00 to 21:00); characteristic index (3) is around 22:00; characteristic index (4) is non-volatile.

[0076] Rule 5: Comprehensive multi-person households

[0077] The characteristic index (1) is at a relatively high level; the characteristic index (2) exists in two periods (11:00-12:00 and 20:00-21:00); the characteristic index (3) is around 22:00; the characteristic index (4) is without fluctuation.

[0078] Furthermore, in view of the randomness and fuzziness of the user at the boundary of the preferred level, the cloud model is used to generate the membership of each indicator in the benchmark cloud; the cloud model is a mathematical model that combines probability theory with traditional fuzzy mathematics theory, which can fuzzily transform the quantitative evaluation value into a qualitative comment set to obtain the membership of all indicator evaluation values ​​in each evaluation level; wherein: the basic definition of the cloud: let U Q is the domain of the quantity, c is the domain of U Q Qualitative concept; if the value x∈U Q , and x is a random realization of a qualitative concept c and satisfies μ(x)∈[0,1], where μ(x) is the membership of x to c, which is a random number, namely μ:U Q →[0,1], x~μ(x), x in U Q The distribution on is called a cloud, and x is called a cloud droplet. A cloud is composed of many cloud droplets, and each cloud droplet is a mapping of a qualitative concept in the quantity domain. Since the mapping of cloud droplets to [0,1] is one-to-many, the cloud graph is not a fixed curve, but follows a certain probability distribution. The more cloud droplets there are, the more it can reflect the randomness and fuzziness of the qualitative concept, and the more the overall characteristics of the qualitative concept can be reflected;

[0079] Cloud by Expectation E x 、Entropy E n and super entropy H e These three digital features reflect the quantitative characteristics of qualitative concepts. They reflect the center of gravity of the cloud and indicate the central value of the corresponding fuzzy information converted into quantitative evaluation; n It reflects the dispersion and fluctuation range of cloud droplets and is an x An estimate of uncertainty represents the degree of ambiguity that can be accepted by a qualitative concept; H e It reflects the stability of cloud droplets, is the entropy of the entropy value, and indicates the uncertainty of entropy.

[0080] Furthermore, the uncertainty focus is introduced to transform it into the basic probability distribution in the evidence theory; the cloud model corresponding to each indicator is established respectively with the evaluation set {level I, level II, level III, level IV, level V} as the recognition framework. x 、E n , H e ), and obtain the membership of each indicator value at different levels, that is, the membership matrix. According to the definition of evidence theory, the sum of the membership of each indicator at different levels should be 1, but the sum of the membership obtained by the cloud model is not 1, that is, it does not fully meet the definition of the basic probability distribution function in evidence theory. Therefore, it is necessary to introduce the uncertainty probability of the i-th indicator level distribution, and transform the membership obtained by the cloud model into a basic probability distribution that meets the definition of evidence theory;

[0081] Furthermore, the reference game theory combines the static and dynamic weights of evidence, identifies conflicting evidence based on the combined weights, and corrects and merges the conflict; integrates the dynamic weights and static weights of the evidence to obtain the combined weights, and then determines the conflicting evidence based on the combined weights, and corrects its basic probability distribution function through the discount factor. By identifying conflicting evidence through combined weights and correcting only the conflicting evidence, it can not only reduce the amount of calculation, but also retain the original data as much as possible, thereby improving the credibility of the decision; assuming that there are o groups of evidence, with the average weight 1 / o assigned to each piece of evidence as the threshold, when the combined weight W i ≥1 / o, the evidence is considered non-conflicting and should be retained. i When ≤1 / o, the evidence is considered conflicting evidence and should be revised;

[0082] Furthermore, the idea of ​​approximating the ideal solution sorting technique is used to compare the closeness of each user's basic probability distribution to the positive and negative ideal clouds to determine the user's aggregation priority; the user's preference order is determined by calculating the closeness between the basic probability distribution function formed by each user after evidence fusion and the basic probability distribution function corresponding to the positive and negative ideal solutions; the steps are as follows:

[0083] 1) Calculate the probability distribution function of positive and negative ideal clouds

[0084] Combined with the indicator attributes, the optimal and worst values ​​of each indicator among all users are selected and substituted into the cloud model to obtain the membership matrix of the positive ideal cloud and the negative ideal cloud. Then, the aforementioned evidence theory fusion method is used to obtain the probability distribution of the positive and negative ideal clouds after evidence fusion, which represent the most ideal solution and the least ideal solution respectively.

[0085] 2) Calculate the fit between each candidate user and the ideal cloud

[0086] Calculate the closeness between the basic probability distribution function of each user formed after evidence fusion and the basic probability distribution function of the positive and negative ideal clouds, and calculate the average of the user's fit;

[0087] 3) Calculate the average fit difference

[0088] Based on the average fit, calculate the difference between the average fit of each user and the maximum and minimum average fit. The larger the difference, the closer the user is to the most ideal solution, and use this to rank the users;

[0089] According to the above method, a virtual power plant user optimization system considering multiple types of users is established, and the system is used to implement the steps of a virtual power plant user optimization system and method considering multiple types of users described in the above method. The system includes:

[0090] A user data processing module is used to perform cluster analysis on the load curve of residential users through a clustering algorithm to obtain the clustering results of the residential electricity load;

[0091] The user optimization index selection module is used to determine the user category from four dimensions: overall load level, peak load period, evening load reduction period, and daily load fluctuation rate, and to build a multi-type user optimization index system;

[0092] Construct a model module to generate the membership of each indicator in the benchmark cloud using a cloud model based on the randomness and fuzziness of users at the boundary of the preferred level;

[0093] Probability assignment module, used to introduce uncertainty focal element and transform it into basic probability assignment in evidence theory;

[0094] The conflict correction and fusion module is used to refer to the game theory to combine the static and dynamic weights of evidence, identify conflicting evidence according to the combined weights, and perform conflict correction and fusion;

[0095] The comparison and determination user priority module is used to compare the closeness of each user's basic probability distribution with the positive and negative ideal clouds to determine the user's aggregation priority;

[0096] Furthermore, the user data processing module in the system uses the silhouette coefficient as an evaluation index to evaluate the clustering results. The value range of the silhouette coefficient is [-1, 1]. The closer it is to 1, the more reasonable the clustering result of the sample is. The silhouette coefficient of sample i is defined as

[0097]

[0098] Among them, a(i) is the average distance between i and other objects in the cluster to which i belongs, which is called the intra-cluster dissimilarity of sample i; b(i) is the minimum average distance from i to all clusters that do not contain i, which is called the inter-cluster dissimilarity of sample i. dist represents the Euclidean distance between two samples; |C s |、|C t |represent the number of samples contained in clusters s and t respectively.

[0099] The evaluation index calculation formula is as follows:

[0100]

[0101] Among them, m is the number of clustering, M is the maximum number of clustering, and T is the average value of the silhouette coefficient.

[0102] Example 1

[0103] The present invention provides an embodiment, which is a virtual power plant user optimization system and method considering multiple types of users, specifically a virtual power plant aggregate user optimization method based on a cloud model and improved evidence theory, comprising the following steps:

[0104] [S001] Calculate the degree of membership of each indicator in each evaluation standard level, and fuse the degree of membership of each indicator with multi-source information to form a comprehensive degree of membership to determine the comprehensive evaluation result.

[0105] The present invention adopts a cloud model. Compared with the traditional model, which is usually based on expert scoring or membership function, it has strong subjectivity on the one hand and cannot take into account the randomness and fuzziness in multi-attribute evaluation decision-making on the other hand. The cloud model, as a mathematical model that combines probability theory with traditional fuzzy mathematics theory, can fuzzily transform quantitative evaluation values ​​into qualitative comment sets, obtain the membership of all indicator evaluation values ​​at each evaluation level, and effectively overcome the randomness and fuzziness in the user optimization of virtual power plants. The fuzziness (either this or that) is reflected in the super entropy value He in the cloud model, and the randomness (either this or that) is reflected in the normal random distribution function in the cloud model.

[0106] As a strict "and" operation rule, the evidence theory, with its strong information fusion ability, introduces uncertainty focal elements to fuse the membership of various indicators obtained by the cloud model at different levels, calculates their degree of membership at each evaluation level, and obtains the comprehensive evaluation decision result, thereby effectively solving the uncertainty problem in the optimization of virtual power plant users. In summary, the use of cloud models and evidence theory can effectively solve the randomness, fuzziness and uncertainty in the optimization of virtual power plant users, making the evaluation and optimization results more reliable.

[0107] Basic definition of cloud: Q is the domain of the quantity, c is the domain of U Q The qualitative concept of Q , and x is a random realization of the qualitative concept c and satisfies μ(x)∈[0,1], where μ(x) is the membership of x to c, which is a random number. QThe distribution on is called a cloud, and x is called a cloud droplet. A cloud is composed of many cloud droplets, and each cloud droplet is a mapping of a qualitative concept in the quantity domain. Since the mapping of cloud droplets to [0, 1] is one-to-many, the cloud graph is not a fixed curve, but follows a certain probability distribution. The more cloud droplets there are, the more it can reflect the randomness and fuzziness of the qualitative concept, and the more the overall characteristics of the qualitative concept can be reflected. For this reason, this embodiment sets the number of cloud droplets to 3000. The cloud is characterized by three digital features: expectation Ex, entropy En, and super entropy He to embody the quantitative characteristics of qualitative concepts. Ex reflects the center of gravity of the cloud, indicating the central value of the corresponding fuzzy information converted into a quantitative evaluation; En reflects the dispersion and fluctuation range of cloud droplets, which is an estimate of the uncertainty of Ex and represents the degree of fuzziness that can be accepted by qualitative concepts; He reflects the stability of cloud droplets, is the entropy of the entropy value, and indicates the uncertainty of entropy.

[0108] For an identification frame Θ consisting of J elements, the elements in the frame Θ are mutually exclusive. J is the number of all subsets of Θ, and is in the interval [0, 1]. If the function mass satisfies ∑mass(Q)=1, then it is called the basic probability distribution function of the identification framework Θ; Q is called the focal element of the basic probability distribution function; mass(Q) is called the support degree of the relevant evidence for the focal element Q, also known as the membership degree. Assume that m p With m q They are the pth and qth information sources belonging to the same identification frame θ, and the corresponding focal elements are Q p1 ,Q p2 ,…,Q pa With Q q1 ,Q q2 ,…,Q qb .

[0109]

S002

[0110]

S003

[0111] [S004] The user preference order is determined by calculating the closeness between the basic probability distribution function formed by each user after evidence fusion and the basic probability distribution function corresponding to the positive and negative ideal solutions.

[0112] Because evidence synthesis is to fuse multiple pieces of evidence to form a comprehensive membership vector, the level of the decision object can often be obtained through the maximum membership principle, but when there are multiple decision objects and they belong to the same level, these decision objects cannot be further compared. The present invention uses the basic idea of ​​TOPSIS to determine the user's preferred order by calculating the closeness between the basic probability distribution function formed by each user after evidence fusion and the basic probability distribution function corresponding to the positive and negative ideal solutions.

[0113]

S0041

[0114] [S0042] Calculate the closeness of the basic probability distribution function of each user formed after evidence fusion with the basic probability distribution function of the positive and negative ideal clouds, and calculate the average of the user's fit.

[0115] [S0043] Based on the average fit, calculate the difference between the average fit of each user and the maximum and minimum average fit. The larger the difference, the closer the user is to the most ideal solution, and use this to rank the users.

[0116] Example

[0117] In order to verify the effectiveness of the proposed virtual power plant user optimization system and method considering multiple types of users, the following steps are specifically implemented:

[0118] Step 1: Improve the validity of evidence theory

[0119] Step 1.1 Evidence Fusion

[0120] Let {α, β, γ} represent a complete identification framework, m1, m2, …, m5 represent evidence, and the probability distribution of each piece of evidence on different focal elements α, β, and γ is shown in Table 1. Taking the basic probability distribution and the corresponding static weights in Table 1 as an example, the improved method proposed in this paper and the common methods in other literatures are used to fuse evidence, and the results are compared to verify the effectiveness and superiority of the improved method in this paper. It means that the evidence m1, m2, ..., m5 are fused. The evidence fusion results of each method are shown in Table 2.

[0121] Table 1 Basic probability distribution and static weights

[0122]

[0123] Table 2 Evidence fusion results of various methods

[0124]

[0125]

[0126] Step 1.2 Analysis of fusion results

[0127] It can be seen from Table 1 that evidence m3 has the largest probability distribution to the focus element γ, and the remaining evidence has the largest probability distribution to the focus element α. The evidence fusion result should maximize the probability distribution to the focus element α and give a certain probability distribution to γ.

[0128] As shown in Table 2, the method of the present invention is effective for strong conflicting evidence. The fusion result is mass(γ)=0.0003, indicating that the focal element γ is also assigned a corresponding probability, while the first three methods have defects in dealing with strong conflicts, making the probability of focal element γ assigned to 0, which is inconsistent with intuitive judgment. The single weight coefficient correction method also assigns a certain probability to the focal element γ, indicating that it has certain advantages in dealing with strong conflict evidence, but the fusion result of the method described in the present invention shows mass(α)=0.9997. Compared with the single weight coefficient correction method, the focal element α gets a higher probability assignment, making the correction result more ideal, thereby verifying the effectiveness and superiority of the method described in the present invention.

[0129] Step 2: Actual example analysis

[0130] Six different types of field survey data from a city in Zhejiang Province were selected for example verification. Among them, users B and F are school district housing communities; user E is a community with convenient indoor transportation; users A and D are single apartments for young people; and user C is a retirement community for retired employees of a certain university. The statistical cycle of the equipment failure rate and the number of credit assessments in the quantitative indicators is 1 year, and the statistical cycle of the other quantitative indicators is 20 days.

[0131] Step 2.1 Parameter setting

[0132] Among the selected survey data, the actual situation and scoring basis of qualitative indicators are shown in Table 3, and the quantitative indicator data are shown in Table 4.

[0133] Table 3 Actual situation of user qualitative indicators and scoring basis

[0134]

[0135] Table 4 User quantitative indicator data

[0136]

[0137] Step 2.2 Establish the grading criteria for indicators

[0138] The grading standards for setting up indicators are shown in Table 5. It is worth pointing out that the indicator grading standards can be changed according to actual needs without affecting the applicability of the model. Among them, A1, A2, and C3 are scoring values. Taking indicator A1 as an example, the cloud digital features (Ex, En, He) at 5 levels are calculated as (1, 0.85, 0.08), (3, 0.85, 0.08), (5, 0.85, 0.08), (7, 0.85, 0.08), and (9, 0.85, 0.08).

[0139] Table 5 Index classification standards

[0140]

[0141]

[0142] Step 2.3 Calculate the average user proximity and difference

[0143] Taking user A as an example, combined with the digital features and indicator data of the benchmark cloud, the average closeness and average closeness difference of each user to the positive and negative ideal clouds are calculated as shown in Table 6. According to the principle that the larger the average closeness difference, the closer the user is to the most ideal result, the user aggregation preference order is finally determined to be user B>user E>user C>user A>user F>user D, where > represents the operator symbol of better than.

[0144] Table 6 Average closeness and difference between each user and the positive and negative ideal clouds

[0145]

[0146] The preferred order using the cloud model and fuzzy comprehensive evaluation method is: user E>user B>user F>user C>user D>user A, which is inconsistent with the actual aggregation situation of a virtual power plant operator in Zhejiang Province, while the method proposed in the present invention is consistent with the actual situation, further proving the effectiveness and superiority of the method proposed in the present invention.

[0147] Anything not described in the present invention is applicable to the prior art.

[0148] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. However, the obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A virtual power plant user optimization system and method considering multiple types of users, characterized by: The steps include: Step 1: Taking the evaluation set {level 1, level 2, level 3, level 4, level 5} of the optimal evaluation as the identification framework, the cloud models corresponding to each indicator are established respectively, and the membership of each indicator value at different levels is obtained, that is, the membership matrix; Step 2: Integrate the dynamic weight and static weight of the evidence to obtain the combined weight, then determine the conflicting evidence based on the combined weight, and modify its basic probability distribution function through the discount factor; Step 3, identifying conflicting evidence by combining weights and correcting only the conflicting evidence; Step 4: Based on the basic idea of ​​TOPSIS, the priority order of users is determined by calculating the closeness between the basic probability distribution function formed by each user after evidence fusion and the basic probability distribution function corresponding to the positive and negative ideal solutions; Furthermore, in step 4, if there are multiple decision objects belonging to the same level, and it is impossible to further compare these decision objects, the optimal and the worst values ​​of each indicator among all users are selected in combination with the indicator attributes, and the membership matrices of the positive ideal cloud and the negative ideal cloud are obtained by substituting them into the cloud model, and then the aforementioned evidence theory fusion method is used to obtain the probability distribution of the positive and negative ideal clouds after evidence fusion, which represent the most ideal solution and the least ideal solution respectively; Preferably, in step 3, assuming that there are o groups of evidence, the average weight 1 / o assigned to each piece of evidence is used as a threshold, when the combined weight Wi ≥ 1 / o, the evidence is considered to be non-conflicting evidence and should be retained; when the combined weight Wi < 1 / o, the evidence is considered to be conflicting evidence and should be revised; The step 2 is a combined weighting method based on game theory, which integrates the dynamic weight and static weight of the evidence to obtain the combined weight, and then determines the conflicting evidence based on the combined weight, and corrects its basic probability distribution function through the discount factor; this method not only fully considers the importance difference between the evidences, but also only corrects the conflicting evidence, which can retain the original information as much as possible, identify the correct focal element more efficiently, and thus improve the accuracy of evidence fusion; Furthermore, combined with the indicator attributes, the optimal and worst values ​​of each indicator among all users are selected and substituted into the cloud model to obtain the membership matrix of the positive ideal cloud and the negative ideal cloud. Then, the aforementioned evidence theory fusion method is used to obtain the probability distribution of the positive and negative ideal clouds after evidence fusion, which represent the most ideal solution and the least ideal solution respectively. Furthermore, the closeness of the basic probability distribution function of each user formed after evidence fusion and the basic probability distribution function of the positive and negative ideal clouds is calculated, and the average of the user's fit is calculated; based on the average fit, the difference between the average fit of each user and the maximum and minimum average fit is calculated; The larger the difference, the closer the user is to the most ideal solution, and the users are ranked accordingly.

2. According to claim 1, a virtual power plant user optimization system and method considering multiple types of users is characterized by: Classify and analyze the residential load data, that is, cluster the load curves of residential users through clustering algorithms to find the patterns of electricity consumption, and then conduct a refined analysis of the corresponding electricity consumption characteristics to determine the electricity consumption habits of different residential users; Determine user categories from four dimensions: overall load level, peak load period, and evening load reduction period, and build a multi-type user optimization index system; Aiming at the randomness and fuzziness of users at the boundary of the preferred level, a cloud model is used to generate the membership of each indicator in the benchmark cloud; The uncertainty focal element is introduced to transform it into the basic probability distribution in the theory of evidence; Referring to the idea of ​​game theory, the static and dynamic weights of evidence are combined in a game, and conflicting evidence is identified according to the combined weights, and conflict correction and integration are performed; Drawing on the idea of ​​approximating the ideal solution sorting technique, the closeness of each user's basic probability distribution to the positive and negative ideal clouds is compared to determine the user's aggregation priority.

3. A virtual power plant user selection system and method considering multiple types of users according to claim 1, characterized in that: Before classifying and analyzing the residential load data, load data preprocessing must be performed, including data cleaning and data normalization; The data cleaning is to clean the collected daily load data, mainly using Newton interpolation to fill in the missing data; it is known that there are n points (x1, f(x1)), (x2, f(x2)), ... (x n ,f(x n )) with the missing point being ((x i ,f(x i ))), where x∈R, i∈[1,n], then the interpolation polynomial is: f(x i )=f(x1)+(x i -x1)f[x2,x1]+(x i -x1)(x i -x2)f[x3,x2,x1] +…+(x i −x1)(x i -x2)…(x i -x n-1 )f[x n ,x n-1 ,…,x2,x1] +(x i −x1)(x i -x2)…(x i -x n )f[x n ,x n-1 ,…,x1,x i ] =P(x i )+R(x i ) Among them, f(x i ) is the missing value obtained by Newton interpolation, P(x i ) is the Newton interpolation approximation function, R(x i ) is the error function; P(x i )=f(x1)+(x i -x1)f[x2,x1]+(x i -x1)(x i -x2)f[x3,x2,x1] +…+(X i −X1)(X i -X2)…(X i -X n-1 )f[X n ,X n-1 ,…,X2,X1] R(x i )=(x i −x1)(x i -x2)…(x i -x n )f[x n ,x n-1 ,…,x1,x i ] The data normalization is to normalize the cleaned data; the linear normalization method is used to convert the value of the original data to between [0, 1]. The formula used is as follows: on i ′=(p i -min(p)) / (max(p)-min(p)) Where p is the original data and p' is the normalized data.

4. A virtual power plant user selection system and method considering multiple types of users according to claim 1, characterized in that: The preprocessed load data is clustered in two steps, namely the pre-clustering stage and the clustering stage. When the two-step clustering method is used, the clustering criterion is based on the Schwarz Bayesian criterion, also known as the Bayesian Information Criterion (BIC), and the distance measurement uses the log-likelihood distance. In the pre-clustering stage, the idea of ​​CF (Clustering Feature) tree growth in the BIRCH (Balanced Iterative Reducing and Clustering Using Hierarchies) algorithm is adopted to read the data points in the data set one by one, and while generating the CF tree, the data points in the dense area are pre-clustered to form several sub-clusters; In the clustering stage, the result of the pre-clustering stage, that is, the sub-clusters, is used as the object, and the agglomerative method is used (i.e., each sub-cluster is regarded as a cluster, and then the two closest clusters are repeatedly merged), and the sub-clusters are merged one by one until the desired number of clusters is reached; Clustering mainly relies on the Bayesian Information Criterion to find the approximate range of the optimal number of clusters; the BIC criterion is defined as: BIC=-2ln(L)+ln(h)·Y Among them, L represents the maximum likelihood function value in the model; h is the sample size of the model; Y represents the number of unknown parameters in the model; Log-likelihood distance is a method based on statistical theory to calculate the difference between clusters. Assume that the current data object is divided into J clusters: C1, ..., C J If C s and C t (C t (s,t∈{1,…,J})) two clusters are merged into one cluster, denoted as C ＜s,t＞ , define cluster C s and C t The log-likelihood distance between is as follows: d(C s ,C t )=ζ s +g t -g s,t> Among them, ξ is the specific form of the log-likelihood function, which is defined as follows: Among them, K A and K B are numerical variables and categorical variables, respectively. and are the total variance of the k-th numerical variable and the variance in the v-th class; N ν and N νkl are the sample size of the vth class, and the sample size of the lth class for the kth categorical variable in the vth class; the kth categorical variable has L k categories; the formula introduces The purpose is to solve the problem that the variance of the vth category may be 0 and the logarithm cannot be calculated.

5. A virtual power plant user selection system and method considering multiple types of users according to claim 1, characterized in that: The user category is determined based on four characteristic indicators: overall load level, peak load period, evening load reduction period, and daily load fluctuation rate: (1) Overall load level: Among them, P level Represents the overall load level; P averoge is the average daily load; P max is the maximum load in all sampled data; according to the calculation results in the formula, it is divided into five categories: high level (greater than or equal to 50%), high level (greater than or equal to 40%, less than 50%), medium level (greater than or equal to 20%, less than 40%), low level (greater than or equal to 10%, less than 20%) and very low level (less than 10%); (2) Peak electricity consumption period: It is mainly divided into four time periods: 5:00-6:00 in the morning, 11:00-12:00 in the morning, 19:00-20:00 in the evening, and 20:00-21:00 in the evening; (3) Late load reduction period: The time is divided into two categories: 21:00 and 22:00; (4) Daily load fluctuation rate: Among them, P wave represents the daily load fluctuation rate; P error is the standard deviation of daily load; P average Daily average load: divided into two categories according to the calculation results of the above formula: with fluctuation (30% and above) and without fluctuation (below 30%); Based on the above characteristic indicators, the user category is determined according to the following rules: Rule 1: Office workers + children combined household Characteristic index (1) is at a high level; characteristic index (2) exists in two periods (11:00-12:00 and 19:00-20:00); characteristic index (3) is around 21:00; characteristic index (4) is fluctuating; Rule 2: Office workers + elderly residents Characteristic index (1) is at a medium level; characteristic indexes (2) and (3) are the same as those in rule (b); characteristic index (4) is non-volatile; Rule 3: Elderly Households Characteristic index (1) is at a low level; characteristic index (2) exists in two periods (5:00-6:00 and 20:00-21:00); characteristic index (3) is around 21:00; characteristic index (4) is non-volatile; Rule 4: Single working tenants Characteristic indicator (1) is at an extremely low level; characteristic indicator (2) only exists in one period (20:00-21:00); characteristic indicator (3) is around 22:00; characteristic indicator (4) is non-volatile; Rule 5: Comprehensive multi-person households The characteristic index (1) is at a relatively high level; the characteristic index (2) exists in two periods (11:00-12:00 and 20:00-21:00); the characteristic index (3) is around 22:00; the characteristic index (4) is without fluctuation.

6. A virtual power plant user optimization system and method considering multiple types of users according to claim 1, characterized in that: Aiming at the randomness and fuzziness of the user at the boundary of the preferred level, the cloud model is used to generate the membership of each indicator in the benchmark cloud; the cloud model is a mathematical model that combines probability theory with traditional fuzzy mathematics theory, which can fuzzily transform the quantitative evaluation value into a qualitative comment set to obtain the membership of all indicator evaluation values ​​in each evaluation level; wherein: the basic definition of the cloud: let U Q is the domain of the quantity, c is the domain of U Q Qualitative concept; if the value x∈U Q , and x is a random realization of a qualitative concept c and satisfies μ(x)∈[0,1], where μ(x) is the membership of x to c, which is a random number, namely μ:U Q →[0,1], x~μ(x), x in U Q The distribution on is called a cloud, and x is called a cloud droplet. A cloud is composed of many cloud droplets, and each cloud droplet is a mapping of a qualitative concept on the quantity domain. Since the mapping of cloud droplets to [0,1] is one-to-many, the cloud graph is not a fixed curve, but follows a certain probability distribution. The more cloud droplets there are, the more it can reflect the randomness and fuzziness of the qualitative concept, and the more the overall characteristics of the qualitative concept can be reflected. Cloud by Expectation E x 、Entropy E n and super entropy H e These three digital features reflect the quantitative characteristics of qualitative concepts; they reflect the center of gravity of the cloud and represent the central value of the corresponding fuzzy information converted into quantitative evaluation; n It reflects the dispersion and fluctuation range of cloud droplets and is an x An estimate of uncertainty, representing the degree of ambiguity that can be accepted by a qualitative concept; H e It reflects the stability of cloud droplets, is the entropy of the entropy value, and indicates the uncertainty of entropy.

7. A virtual power plant user optimization system and method considering multiple types of users according to claim 1, characterized in that: The uncertainty focus is introduced to transform it into the basic probability distribution in the evidence theory; the cloud model corresponding to each indicator is established respectively with the evaluation comment set {level I, level II, level III, level IV, level V} as the identification framework. x 、E n , H e ), and obtain the membership of each indicator value at different levels, that is, the membership matrix; according to the definition of evidence theory, the sum of the membership of each indicator at different levels should be 1, but the sum of the membership obtained by the cloud model is not 1, that is, it does not fully meet the definition of the basic probability distribution function in evidence theory; therefore, it is necessary to introduce the uncertainty probability of the i-th indicator level distribution, and transform the membership obtained by the cloud model into a basic probability distribution that conforms to the definition of evidence theory.

8. A virtual power plant user optimization system and method considering multiple types of users according to claim 1, characterized in that: The reference game idea combines the static and dynamic weights of evidence through game, identifies conflicting evidence according to the combined weight, and corrects and merges the conflict; integrates the dynamic weight and static weight of the evidence to obtain the combined weight, and then determines the conflicting evidence based on the combined weight, and corrects its basic probability distribution function through the discount factor; identifies the conflicting evidence through the combined weight and only corrects the conflicting evidence, which can not only reduce the amount of calculation, but also retain the original data as much as possible, thereby improving the credibility of the decision; assuming that there are o groups of evidence, the average weight 1 / o assigned to each evidence is used as the threshold, when the combined weight W i ≥1 / o, the evidence is considered non-conflicting and should be retained; when the combined weight W i When ≤1 / o, the evidence is considered conflicting evidence and should be revised.

9. A virtual power plant user optimization system and method considering multiple types of users according to claim 1, characterized in that: The idea of ​​approximating the ideal solution sorting technique is used to compare the closeness of each user's basic probability distribution to the positive and negative ideal clouds to determine the user's aggregation priority; the user's preference order is determined by calculating the closeness between the basic probability distribution function formed by each user after evidence fusion and the basic probability distribution function corresponding to the positive and negative ideal solutions; the steps are as follows: 1) Calculate the probability distribution function of positive and negative ideal clouds Combined with the indicator attributes, the optimal and worst values ​​of each indicator among all users are selected and substituted into the cloud model to obtain the membership matrix of the positive ideal cloud and the negative ideal cloud. Then, the aforementioned evidence theory fusion method is used to obtain the probability distribution of the positive and negative ideal clouds after evidence fusion, which represent the most ideal solution and the least ideal solution respectively. 2) Calculate the fit between each candidate user and the ideal cloud Calculate the closeness between the basic probability distribution function of each user formed after evidence fusion and the basic probability distribution function of the positive and negative ideal clouds, and calculate the average of the user's fit; 3) Calculate the average fit difference According to the average fit, the difference between the average fit of each user and the maximum and minimum average fit is calculated; The larger the difference, the closer the user is to the most ideal solution, and the users are ranked accordingly.

10. A virtual power plant user optimization system considering multiple types of users according to claim 1, the system is used to implement the steps of a virtual power plant user optimization system and method considering multiple types of users according to any one of claims 1 to 9, the system comprising: A user data processing module is used to perform cluster analysis on the load curve of residential users through a clustering algorithm to obtain the clustering results of the residential electricity load; The user optimization index selection module is used to determine the user category from four dimensions: overall load level, peak load period, evening load reduction period, and daily load fluctuation rate, and to build a multi-type user optimization index system; Construct a model module to generate the membership of each indicator in the benchmark cloud using a cloud model based on the randomness and fuzziness of users at the boundary of the preferred level; Probability assignment module, used to introduce uncertainty focal element and transform it into basic probability assignment in evidence theory; The conflict correction and fusion module is used to refer to the game theory to combine the static and dynamic weights of evidence, identify conflicting evidence according to the combined weights, and perform conflict correction and fusion; The comparison and determination user priority module is used to compare the closeness of each user's basic probability distribution with the positive and negative ideal clouds to determine the user's aggregation priority; The user data processing module uses the silhouette coefficient as an evaluation index to evaluate the clustering results; the value range of the silhouette coefficient is [-1, 1], and the closer it is to 1, the more reasonable the clustering result of the sample is; the silhouette coefficient of sample i is defined as Where a(i) is the average distance between i and other objects in the cluster to which i belongs, which is called the intra-cluster dissimilarity of sample i; b(i) is the minimum average distance from i to all clusters that do not contain i, which is called the inter-cluster dissimilarity of sample i; dist represents the Euclidean distance between two samples; |C s |、|C t |represents the number of samples contained in clusters s and t respectively; The evaluation index calculation formula is as follows: Among them, m is the number of clustering, M is the maximum number of clustering, and T is the average value of the silhouette coefficient.

Images