A method and system for selecting and recommending power sales companies for retail power users

Abstract

The present invention discloses a method and system for selecting and recommending a power sales company for retail power users. The method includes: clustering the power load curves of retail power users, dividing the user load curves into two categories: non-flat type and flat type; classifying the power demand and green users of retail power users, and constructing a user feature matrix; classifying risks based on the credit rating and registered capital of the power sales company; classifying market influence based on the total number of transactions and the total number of transaction users of the power sales company; constructing a power sales company feature matrix; and enabling users to select or recommend power sales companies based on the feature matrix information of retail power users and power sales companies. The method proposed in the present invention comprehensively considers the user's power consumption information and transaction behavior preferences in mining user needs. It can not only recommend power sales companies to retail power users like the existing methods, but also allows users to quickly screen out power sales companies of interest.

Description

Technical Field

[0001] The present invention relates to the technical field of power retail market data analysis, and in particular to a method and system for selecting and recommending power sales companies for retail power users. Background Art

[0002] In recent years, with the development of the electricity retail market, the electricity retail market has shown the characteristics of covering many entities and large transaction scale. How to continuously stimulate the vitality of the retail market and improve the efficiency of resource allocation has become a development demand of the electricity retail market. If retail electricity users can quickly select or obtain electricity sales companies that suit their personalized needs, it will help reduce the user's selection information cost and increase the enthusiasm and return of retail electricity users to participate in the retail market. However, there are currently few studies on recommending electricity sales companies that meet their personalized needs to retail electricity users, and the data characteristics of retail electricity users and electricity sales companies are multi-dimensionally complex. Therefore, how to effectively screen the data of the two market entities, retail electricity users and electricity sales companies, reasonably classify the two market entities, and recommend electricity sales companies that meet the personalized needs of retail electricity users has become an urgent problem to be solved in the process of coordinated development of the operation side and the supply side and promoting the healthy development of electricity market reform.

[0003] In the power retail market data, the data related to power users and power sales companies are multi-type and multi-dimensional. It is still unclear what type of data to choose to recommend power sales companies that meet the personalized needs of retail power users. Therefore, in the power retail market environment, a power sales company selection and recommendation method for retail power users can recommend power sales companies that meet the personalized needs of retail power users, which provides a solution for reducing the information cost of user selection and optimizing market resource allocation.

[0004] At present, there are Chinese patent documents that study the recommendation method of power sales companies (CN115393008A). This method collects user demand information or preference information, but only extracts relevant information from semantic data, ignoring historical transaction and electricity consumption data related to demand and preference information. Summary of the invention

[0005] In order to solve the above technical problems, the present application provides a method and system for selecting and recommending power sales companies for retail power users, which fully considers the risk reputation, package conditions, market influence, and power sales price of the power sales companies. It can not only allow retail power users to quickly screen power sales companies according to their personal wishes, but also recommend power sales companies that may be of interest to power users; it can recommend power sales companies that meet the personalized needs of retail power users, increase the activity of the power retail market, and improve the efficiency of resource allocation.

[0006] The purpose of the present invention is achieved by at least one of the following technical solutions.

[0007] A method for selecting and recommending a power sales company for retail power users comprises the following steps:

[0008] S1. The electricity consumption curves of retail power users are clustered by the K-Medoids clustering method based on cosine similarity, and two types of electricity consumption curves of retail power users are obtained. Based on these, the users are divided into two types: non-flat type and flat type.

[0009] S2. Classify the electricity demand of retail electricity users and green users, and construct a characteristic matrix of retail electricity users based on the electricity curve patterns, electricity demand categories and coding of green user categories of retail electricity users;

[0010] S3. Encode the credit rating of the power sales company, and classify the risk of the power sales company using the GMM clustering algorithm based on the credit rating and registered capital of the power sales company, and encode the risk classification results;

[0011] S4. Based on the total number of transactions and the total number of transaction users of the power sales companies, the GMM clustering algorithm is used to classify the market influence of the power sales companies, and the market influence classification results are encoded;

[0012] S5. According to the packages provided by the power sales company, the average power sales price is obtained to determine whether green electricity is provided, and a characteristic matrix of the power sales company is constructed based on the risk category, market influence category, green electricity situation and average power sales price of the power sales company;

[0013] S6. Based on the feature matrix of retail electricity users and power sales companies and in accordance with the feature coding matching principle, a manual selection method or an automatic matching method is used to select and recommend power sales companies.

[0014] Furthermore, in step S1, the electricity consumption curves of retail power users are clustered by the K-Medoids clustering method based on cosine similarity to obtain two types of electricity consumption curves of retail power users, and the users are divided into two types: non-flat type and flat type. The specific steps include:

[0015] S1.1 Input N The daily load data of retail electricity users is normalized by the minimum-maximum method. L Convert to a value in the interval [0,1] :

[0016]

[0017] In the formula, L max is the maximum value in the data,L min is the minimum value in the data;

[0018] S1.2. Clustering of daily load data of retail electricity users Clustering is performed, the number of clusters is 2, and finally two types of electricity consumption curves are obtained;

[0019] S1.3, according to Calculate the peak-to-valley difference between two types of electricity consumption curves , For the e The maximum value in the class power consumption curve, For the e The minimum value in the class power consumption curve, e Choose 1 or 2;

[0020] S1.4. According to and Size, will The large power consumption curve is recorded as non-flat type. Small electricity consumption curves are recorded as flat, and these two electricity consumption modes are coded as 1 and 2 respectively; it is generally believed that non-flat retail users have stronger demand response capabilities and are easily affected by prices, and are recorded as price-sensitive users; flat users have low demand response capabilities and are not easily affected by prices, and are recorded as price-insensitive users.

[0021] Furthermore, in step S1.2, the daily load data of retail power users are clustered by a clustering method, and the number of clusters is 2. The specific process is as follows:

[0022] S1.2.1. From N Randomly select two user data from retail electricity users as the centroids of the two clusters;

[0023] S1.2.2. Calculate the cosine similarity between other user data and the two centroids, and group the users into clusters with large cosine similarity to the centroids. The formula for calculating cosine similarity is as follows:

[0024]

[0025] The first i The normalized load data series of retail electricity users, , For the i Retail electricity users in n Normalized load data of each collection point; For the j The load data series after centroid normalization, , For the j The centroid is n The load data after normalization of each collection point is j Choose 1 or 2; M is the number of sampling points of daily load, where M Take 24; The closer it is to 1, the i Users and j The closer the power consumption curves of the centroids are, the i Users are assigned to j The cluster where the centroid is located, the final users are divided into two clusters;

[0026] S1.2.3. Calculate j The sum of the cosine similarities between a centroid and the non-centroid sample points in its cluster , the calculation formula is as follows:

[0027]

[0028] In the formula, F Representative j The centroid of the cluster has F non-centroid sample points;

[0029] S1.2.4. Randomly select a sample point from each of the two clusters as the new centroid of the cluster to which it belongs. The sample point cannot be the original centroid. Calculate the j The sum of the cosine similarities between the new centroid and the non-centroid sample points in its cluster , and compare and The size between the two, the centroid corresponding to the larger value is taken as the new centroid;

[0030] S1.2.5. Repeat steps S1.2.2, S1.2.3 and S1.2.4 until the center of mass remains unchanged.

[0031] Furthermore, in step S2, the retail power users are classified into electricity demand classification and green user classification, and a retail power user characteristic matrix is ​​constructed based on the encoding of the three dimensions of the retail power users' electricity curve pattern, electricity demand category, and green user category. The specific steps include:

[0032] S2.1. Calculate the total power , For the Retail electricity users in p The total power consumption in a power cycle, P is the total power consumption cycle;

[0033] S2.2, perform minimum-maximum normalization processing on the total power in step S2.1, and the total power after processing is recorded as ;

[0034] S2.3. Retail electricity users with a value less than 0.5 are recorded as users with low electricity demand and coded as 1. Retail electricity users with a value greater than 0.5 are recorded as users with high electricity demand and coded as 2;

[0035] S2.4. Enter the retail package purchased by the retail electricity user. If the environmental premium ratio in the package is not zero, it is recorded as a green user and coded as 1; if the environmental premium ratio in the package is zero, it is recorded as a non-green electricity user and coded as 0;

[0036] S2.5. Construct a retail electricity user characteristic matrix based on the coding of three dimensions: electricity consumption curve pattern, electricity demand category, and green user category.

[0037] Furthermore, in step S3, the credit rating of the power sales company is encoded, and based on the credit rating and registered capital of the power sales company, the GMM clustering algorithm is used to classify the risk of the power sales company, and the risk classification result is encoded, as follows:

[0038] S3.1 Input m The credit rating of the power sales company includes AAA, AA, A, B, and C. The credit ratings of AAA, AA, A, B, and C are scored as 5, 4, 3, 2, and 1 respectively. a The credit score of the electricity sales company is ; It is generally believed that the higher the score, the better the credit of the electricity sales company;

[0039] S3.2 Input m The registered capital of each power sales company is processed using the minimum-maximum normalization method. a The normalized registered capital of a power sales company is ; It is generally believed that the larger the registered capital, the stronger the risk resistance of the power sales company;

[0040] S3.3. The scoring results in step S3.1 and the normalized results in step S3.2 constitute the sample data of the power sales company ,The power sales companies are classified by GMM clustering algorithm, the number of categories is 3, and three categories are obtained;

[0041] S3.4 defines three categories as high risk, medium risk, and low risk, and codes the three categories as 1, 2, and 3.

[0042] Furthermore, the specific steps of step S3.3 are as follows:

[0043] S3.3.1. Initialize the model parameters of GMM clustering , , ; For the q The weight of each sub-model, For the q The mean of the sub-models, For the q The covariance of the sub-models;

[0044] S3.3.2. Calculate the a Sample data of electricity sales companies By q The probability generated by the sub-model ,in: , For the l The weight of each sub-model, For sample In the q The probability density function of a Gaussian distribution is For sample In the l The probability density function of a Gaussian distribution;

[0045] S3.3.3. Update the model parameters according to the following formula:

[0046] ,

[0047] ,

[0048]

[0049] in, , , Respectively q The updated mean, covariance, and weight of each sub-model, m is the total number of power sales companies; then repeat steps S3.3.1 and S3.3.2 according to the new model parameters and stop when the maximum number of iterations is reached;

[0050] S3.3.4. Each sample is divided into into the corresponding clusters, Indicates a The maximum probability of a power sales company in the sub-model is obtained, that is, each sample is classified into the cluster of the model with the highest probability of coming from the sub-model, and finally three categories are obtained.

[0051] Furthermore, in step S4, based on the total number of transactions and the total number of transaction users of the power sales company, the GMM algorithm is used to classify the market influence of the power sales company, and the market influence classification result is encoded. The specific method is as follows:

[0052] S4.1. m The total number of transactions and the total number of transaction users of each power sales company constitute the sample data, and the minimum-maximum normalization method is used for processing;

[0053] S4.2, clustering the normalized data in step S4.1 by using the GMM clustering algorithm, the number of categories is 3, and three categories are obtained; the GMM clustering algorithm process is the same as step S3.3;

[0054] S4.3. Define the three categories divided in step S4.2 as low market influence, medium market influence, and high market influence, and code them as 1, 2, and 3.

[0055] Furthermore, in step S5, according to the packages provided by the power sales company, the average power sales price is obtained to determine whether it provides green electricity, and a characteristic matrix of the power sales company is constructed based on the risk category, market influence category, green electricity situation, and average power sales price of the power sales company. The specific method is as follows:

[0056] S5.1. Input the retail package of the electricity sales company. If the environmental premium ratio in the package is not zero, it is recorded as green electricity and coded as 1; if the environmental premium ratio in the package is zero, it is recorded as non-green electricity and coded as 0;

[0057] S5.2. According to Calculate the The average electricity price of each electricity sales company, For the Electricity sales company r The price of a power package. T a For the The total number of packages of each electricity sales company;

[0058] S5.3. Perform minimum-maximum normalization on the average electricity selling price of the electricity selling company; record the normalized average electricity selling price as ,if If it is less than 0.5, the average electricity price is considered low and coded as 1; If it is greater than 0.5, the average electricity price is considered high and coded as 2;

[0059] S5.4. Construct a characteristic matrix of power sales companies based on three aspects: risk category, market influence category, and green electricity situation.

[0060] Furthermore, in step S6, if manual selection is selected, it is necessary to independently select the codes of the three characteristics of the power sales company's risk category, market influence category, and green electricity situation to obtain the power sales company that matches the selected code.

[0061] Furthermore, in step S6, if automatic matching is selected, recommendations are made according to the following rules:

[0062] For users who consume a lot of electricity and are sensitive to prices, priority will be given to power companies with low average power prices and low risks;

[0063] For users who consume a lot of electricity and are not sensitive to prices, low-risk power sales companies are recommended first;

[0064] For users with low electricity consumption and price sensitivity, priority will be given to power companies with low average electricity prices;

[0065] For users with low electricity consumption and insensitive to prices, priority will be given to electricity sales companies with great market influence;

[0066] Green users give priority to power companies that sell green electricity.

[0067] A power sales company selection and recommendation system for retail power users, comprising:

[0068] User classification module: The K-Medoids clustering method based on cosine similarity is used to cluster the power consumption curves of retail power users, and two types of power consumption curves are obtained. Based on these two types of curves, users are divided into two categories: non-flat type and flat type.

[0069] Retail power user characteristic matrix construction module: classifies retail power users by electricity demand and green users, and constructs a characteristic matrix of retail power users based on the electricity curve pattern, electricity demand category and coding of green user category of retail power users;

[0070] Power sales company risk coding module: Encode the credit rating of the power sales company, use the GMM clustering algorithm to classify the risk of the power sales company based on the credit rating and registered capital of the power sales company, and encode the risk classification results;

[0071] Power sales company market influence coding module: Based on the total number of transactions and the total number of transaction users of the power sales company, the GMM clustering algorithm is used to classify the market influence of the power sales company, and the market influence classification results are encoded;

[0072] Power sales company feature matrix construction module: According to the packages provided by the power sales company, the average power sales price is obtained to determine whether green electricity is provided, and the power sales company's feature matrix is ​​constructed based on the risk category, market influence category, green electricity situation and average power sales price of the power sales company;

[0073] Power sales company selection and recommendation module: Based on the feature matrix of retail power users and power sales companies and the feature coding matching principle, manual selection method or automatic matching method is used to select and recommend power sales companies.

[0074] Compared with the prior art, the advantages of the present invention are:

[0075] Compared with the existing power sales company recommendation method, the method proposed in the present invention comprehensively considers the user's electricity usage information and transaction behavior preferences in exploring user needs. It can not only recommend power sales companies to retail power users like the existing methods, but also allow users to quickly screen out the power sales companies they are interested in. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A schematic diagram of the steps of a method for selecting and recommending a power sales company for retail power users in an embodiment of the present invention;

[0077] Figure 2 A cluster diagram of a flat load curve of retail power users in an embodiment of the present invention;

[0078] Figure 3 A cluster diagram of load curves of non-flat retail power users in an embodiment of the present invention;

[0079] Figure 4 This is a classification diagram of the market influence of the power sales company in the embodiment of the present invention. DETAILED DESCRIPTION

[0080] The specific implementation of the present invention is further described below with reference to the accompanying drawings and examples.

[0081] Example:

[0082] In one embodiment, based on the retail market model of Guangdong Power Exchange, a method for selecting and recommending power sales companies for retail power users is provided. Figure 1 As shown, the following steps are included:

[0083] S1. Clustering the electricity consumption curves of retail power users by the K-Medoids clustering method based on cosine similarity, obtaining two types of electricity consumption curves of retail power users, and dividing the users into two types: non-flat type and flat type. The specific steps include:

[0084] S1.1 Input N The daily load data of retail electricity users is normalized by the minimum-maximum method. L Convert to a value in the interval [0,1] :

[0085]

[0086] In the formula, L max is the maximum value in the data, L min is the minimum value in the data;

[0087] S1.2. Clustering of daily load data of retail electricity users Clustering is performed, the number of clusters is 2, and the final two types of electricity consumption curves are as follows Figure 2 and Figure 3 As shown; in the clustering process, cosine similarity is used as the basis for classification, which can better distinguish the changes in load curves than Euclidean distance, Hamming distance and other methods; the specific process of this step is as follows:

[0088] S1.2.1. From N Randomly select two user data from retail electricity users as the centroids of the two clusters;

[0089] S1.2.2. Calculate the cosine similarity between other user data and the two centroids, and group the users into clusters with large cosine similarity to the centroids. The formula for calculating cosine similarity is as follows:

[0090]

[0091] The first i The normalized load data series of retail electricity users, , For the i Retail electricity users in n Normalized load data of each collection point; For the j The load data series after centroid normalization, , For the j The centroid is n The load data after normalization of each collection point is j Choose 1 or 2; M is the number of sampling points of daily load, where M Take 24; The closer it is to 1, the i Users and j The closer the power consumption curves of the centroids are, the i Users are assigned to j The cluster where the centroid is located, the final users are divided into two clusters;

[0092] S1.2.3. Calculate jThe sum of the cosine similarities between a centroid and the non-centroid sample points in its cluster , the calculation formula is as follows:

[0093]

[0094] In the formula, F Representative j The cluster where the centroid is located has F non-centroid sample points;

[0095] S1.2.4. Randomly select a sample point from each of the two clusters as the new centroid of the cluster to which it belongs. The sample point cannot be the original centroid. Calculate the j The sum of the cosine similarities between the new centroid and the non-centroid sample points in its cluster , and compare and The size between the two, the centroid corresponding to the larger value is taken as the new centroid;

[0096] S1.2.5. Repeat steps S1.2.2, S1.2.3 and S1.2.4 until the center of mass remains unchanged.

[0097] S1.3, according to Calculate the peak-to-valley difference between two types of electricity consumption curves , For the e The maximum value in the class power consumption curve, For the e The minimum value in the class power consumption curve, e Choose 1 or 2;

[0098] S1.4. According to and Size, will Large power consumption curve (such as Figure 3 ) is recorded as non-smooth type, and Small power consumption curve (such as Figure 2 ) is recorded as the smooth type, and these two electricity consumption modes are coded as 1 and 2 respectively; it is generally believed that non-smooth retail users have stronger demand responsiveness and are easily affected by prices, and are recorded as price-sensitive users; smooth users have low demand responsiveness and are not easily affected by prices, and are recorded as price-insensitive users.

[0099] S2. Classify the electricity demand and green users of retail electricity users, and construct a characteristic matrix of retail electricity users based on the coding of the three dimensions of electricity curve pattern, electricity demand category and green user category of retail electricity users. The specific steps include:

[0100] S2.1. Calculate the total power , For the Retail electricity users in p The total power consumption in a power cycle, P is the total power consumption cycle;

[0101] S2.2, perform minimum-maximum normalization processing on the total power in step S2.1, and the total power after processing is recorded as ;

[0102] S2.3. Retail electricity users with a value less than 0.5 are recorded as users with low electricity demand and coded as 1. Retail electricity users with a value greater than 0.5 are recorded as users with high electricity demand and coded as 2;

[0103] S2.4. Enter the retail package purchased by the retail electricity user. If the environmental premium ratio in the package is not zero, it is recorded as a green user and coded as 1; if the environmental premium ratio in the package is zero, it is recorded as a non-green electricity user and coded as 0;

[0104] S2.5. Construct a retail power user characteristic matrix based on the coding of the three dimensions of retail power users' power consumption curve pattern, power demand category, and green user category. The retail power user characteristic matrix is ​​shown in Table 1:

[0105] Table 1 Retail electricity user characteristics matrix

[0106]

[0107] S3. Encode the credit rating of the power sales company, and use the GMM clustering algorithm to classify the risk of the power sales company based on the credit rating and registered capital of the power sales company, and encode the risk classification results, as follows:

[0108] S3.1 Input m The credit rating of the power sales company includes AAA, AA, A, B, and C. The credit ratings of AAA, AA, A, B, and C are scored as 5, 4, 3, 2, and 1 respectively. a The credit score of the electricity sales company is ; It is generally believed that the higher the score, the better the credit of the electricity sales company;

[0109] S3.2 Input m The registered capital of each power sales company is processed using the minimum-maximum normalization method. a The normalized registered capital of a power sales company is ; It is generally believed that the larger the registered capital, the stronger the risk resistance of the power sales company;

[0110] S3.3. The scoring results in step S3.1 and the normalized results in step S3.2 constitute the sample data of the power sales company , the GMM clustering algorithm is used to classify the power sales companies. The number of categories is 3, and three categories are obtained. The specific steps are as follows:

[0111] S3.3.1. Initialize the model parameters of GMM clustering , , ; For the q The weight of each sub-model, For the q The mean of the sub-models, For the q The covariance of the sub-models;

[0112] S3.3.2. Calculate the a Sample data of electricity sales companies By q The probability generated by the sub-model ,in: , For the l The weight of each sub-model, For sample In the q The probability density function of a Gaussian distribution is For sample In the l The probability density function of a Gaussian distribution;

[0113] S3.3.3. Update the model parameters according to the following formula:

[0114] ,

[0115] ,

[0116]

[0117] in, , , Respectively q The updated mean, covariance, and weight of each sub-model, m is the total number of power sales companies; then repeat steps S3.3.1 and S3.3.2 according to the new model parameters and stop when the maximum number of iterations is reached;

[0118] S3.3.4. Each sample is divided into into the corresponding clusters, Indicates aThe maximum probability of a power sales company in the sub-model is obtained, that is, each sample is classified into the cluster of the model with the highest probability of coming from the sub-model, and finally three categories are obtained.

[0119] S3.4 defines three categories as high risk (relatively low registered capital and credit score), medium risk (medium registered capital and credit score), and low risk (relatively high registered capital and credit score), and codes the three categories as 1, 2, and 3.

[0120] S4. Based on the total number of transactions and the total number of transaction users of the power sales company, the GMM clustering algorithm is used to classify the market influence of the power sales company, and the market influence classification results are encoded. The specific method is as follows:

[0121] S4.1. m The total number of transactions and the total number of transaction users of each power sales company constitute the sample data, and the minimum-maximum normalization method is used for processing;

[0122] S4.2, clustering the normalized data in step S4.1 by using the GMM clustering algorithm, the number of categories is 3, and three categories are obtained; the GMM clustering algorithm process is the same as step S3.3;

[0123] S4.3, define the three categories divided in step S4.2 as low market influence (relatively low total number of transactions and total number of transaction users), medium market influence (medium total number of transactions and total number of transaction users), and high market influence (relatively high total number of transactions and total number of transaction users), and code them as 1, 2, and 3; the results of these three categories are as follows Figure 4 As shown in the figure, it can be seen that the three categories have obvious distinction and differences.

[0124] S5. According to the packages provided by the power sales company, the average power sales price is obtained to determine whether green electricity is provided. The characteristic matrix of the power sales company is constructed based on the risk category, market influence category, green electricity situation and average power sales price of the power sales company. The specific method is as follows:

[0125] S5.1. Input the retail package of the electricity sales company. If the environmental premium ratio in the package is not zero, it is recorded as green electricity and coded as 1; if the environmental premium ratio in the package is zero, it is recorded as non-green electricity and coded as 0;

[0126] S5.2. According to Calculate the The average electricity price of each electricity sales company, For the Electricity sales company r The price of a power package. T a For the The total number of packages of each electricity sales company;

[0127] S5.3. Perform minimum-maximum normalization on the average electricity selling price of the electricity selling company; record the normalized average electricity selling price as ,if If it is less than 0.5, the average electricity price is considered low and coded as 1; If it is greater than 0.5, the average electricity price is considered high and coded as 2;

[0128] S5.4. Construct a characteristic matrix of power sales companies based on the risk category, market influence category, and green electricity situation of power sales companies. The characteristic matrix of power sales companies is shown in Table 2:

[0129] Table 2. Characteristics matrix of power sales companies

[0130]

[0131] S6. Based on the feature matrix of retail power users and power sales companies and in accordance with the feature coding matching principle, a manual selection method or an automatic matching method is used to select or recommend power sales companies.

[0132] If you choose manual selection, you need to select the codes of the three characteristics of the power sales company: risk category, market influence category, and green power situation, and get the power sales company that matches the selected code;

[0133] In one embodiment, when the risk category code is 1, the market influence category code is 1, and the green electricity situation code is 1, the power sales company with the three characteristic codes consistent therewith will be output.

[0134] If you select automatic matching, recommendations are made based on the following rules:

[0135] For users who consume a lot of electricity and are sensitive to prices, priority will be given to power companies with low average power prices and low risks;

[0136] For users who consume a lot of electricity and are not sensitive to prices, low-risk power sales companies are recommended first;

[0137] For users with low electricity consumption and price sensitivity, priority will be given to power companies with low average electricity prices;

[0138] For users with low electricity consumption and insensitive to prices, priority will be given to electricity sales companies with great market influence;

[0139] Green users give priority to power companies that sell green electricity.

[0140] In one embodiment, the recommendation rules may be shown in the following table:

[0141] Table 3 Recommended rules

[0142]

[0143] The x in the code indicates that the code position can be any code within the code range. For example, the recommended code for power sales companies is [3, x, x, 1]. The second code may be 1, 2, or 3, and x means that it can be any one of 1, 2, or 3.

[0144] In one embodiment, the characteristics of some retail power users are coded as [2,1,0], which means that this is a type of user with a flat power consumption curve, low power consumption, and no record of purchasing green power. The power consumption behavior of this type of user does not change much in a day, and is generally not easily affected by the power price. The average power price, whether it is green power, and risk are not likely to become their focus. Therefore, it is possible to consider pushing some power sales companies with great market influence, that is, power sales companies coded as [x, 3, x, x].

[0145] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other modifications, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and should be included in the protection scope of the present invention.

Claims

1. A method for selecting and recommending a power sales company for retail power users, characterized in that: The following steps are involved: S1. The electricity consumption curves of retail power users are clustered by the K-Medoids clustering method based on cosine similarity to obtain two types of electricity consumption curves, and the users are divided into two types: non-flat type and flat type. Specifically, the following are included: S1.1 Input N The daily load data of retail electricity users is normalized by the minimum-maximum method. L Convert to a value in the interval [0,1] : In the formula, L max is the maximum value in the data, L min is the minimum value in the data; S1.

2. Clustering of daily load data of retail electricity users Clustering is performed, the number of clusters is 2, and finally two types of electricity consumption curves are obtained; the daily load data of retail power users are clustered by clustering method, the number of clusters is 2, and the specific process is as follows: S1.2.

1. From N Randomly select two user data from retail electricity users as the centroids of the two clusters; S1.2.

2. Calculate the cosine similarity between other user data and the two centroids, and group the users into clusters with large cosine similarity to the centroids. The formula for calculating cosine similarity is as follows: The first i The normalized load data series of retail electricity users, , For the i Retail electricity users in n Normalized load data of each collection point; For the j The load data series after centroid normalization, , For the j The centroid is n The load data after normalization of each collection point is j Choose 1 or 2. n The value is 1 -M ; M is the number of sampling points for daily load; The closer it is to 1, the i Users and j The closer the power consumption curves of the centroids are, the i Users are assigned to j The cluster where the centroid is located, the final users are divided into two clusters; S1.2.

3. Calculate j The sum of the cosine similarities between a centroid and the non-centroid sample points in its cluster , the calculation formula is as follows: In the formula, F Representative j The centroid of the cluster has F non-centroid sample points; S1.2.

4. Randomly select a sample point from each of the two clusters as the new centroid of the cluster to which it belongs. The sample point cannot be the original centroid. Calculate the j The sum of the cosine similarities between the new centroid and the non-centroid sample points in its cluster , and compare and The size between the two, the centroid corresponding to the larger value is taken as the new centroid; S1.2.

5. Repeat steps S1.2.2, S1.2.3 and S1.2.4 until the center of mass remains unchanged; S1.3, according to Calculate the peak-to-valley difference between two types of electricity consumption curves , For the e The maximum value in the class power consumption curve, For the e The minimum value in the class power consumption curve; e Take 1 or 2, corresponding to and ; S1.

4. According to and Size comparison, The large power consumption curve is recorded as non-flat type. Small electricity consumption curves are recorded as flat, and these two electricity consumption patterns are coded as 1 and 2 respectively; non-flat retail users are recorded as price-sensitive users; flat users are recorded as price-insensitive users; S2. Classify the electricity demand of retail electricity users and green users, and construct a characteristic matrix of retail electricity users based on the electricity curve patterns, electricity demand categories and coding of green user categories of retail electricity users; S3. Encode the credit rating of the power sales company, and classify the risk of the power sales company using the GMM clustering algorithm based on the credit rating and registered capital of the power sales company, and encode the risk classification results; S4. Based on the total number of transactions and the total number of transaction users of the power sales companies, the GMM clustering algorithm is used to classify the market influence of the power sales companies, and the market influence classification results are encoded; S5. According to the packages provided by the power sales company, the average power sales price is obtained to determine whether green electricity is provided, and a characteristic matrix of the power sales company is constructed based on the risk category, market influence category, green electricity situation and average power sales price of the power sales company; S6. Based on the feature matrix of retail electricity users and power sales companies and in accordance with the feature coding matching principle, a manual selection method or an automatic matching method is used to select and recommend power sales companies.

2. A method for selecting and recommending a power sales company for retail power users according to claim 1, characterized in that: Step S2 specifically includes: S2.

1. Calculate the total power , For the Retail electricity users in p The total power consumption in a power cycle, P is the total power consumption cycle; S2.2, perform minimum-maximum normalization processing on the total power in step S2.1, and the total power after processing is recorded as ; S2.

3. Retail electricity users with a value less than 0.5 are recorded as users with low electricity demand and coded as 1. Retail electricity users with a value greater than 0.5 are recorded as users with high electricity demand and coded as 2; S2.

4. Enter the retail package purchased by the retail electricity user. If the environmental premium ratio in the package is not zero, it is recorded as a green user and coded as 1; if the environmental premium ratio in the package is zero, it is recorded as a non-green electricity user and coded as 0; S2.

5. Construct a retail electricity user characteristic matrix based on the coding of three dimensions: electricity consumption curve pattern, electricity demand category, and green user category.

3. A method for selecting and recommending a power sales company for retail power users according to claim 1, characterized in that: In step S3, the credit rating of the power sales company is encoded, and based on the credit rating and registered capital of the power sales company, the GMM clustering algorithm is used to classify the risk of the power sales company, and the risk classification result is encoded, as follows: S3.1 Input m The credit rating of the power sales company includes AAA, AA, A, B, and C. The credit ratings of AAA, AA, A, B, and C are scored as 5, 4, 3, 2, and 1 respectively. a The credit score of the electricity sales company is ; The higher the score, the better the credit of the electricity sales company; S3.2 Input m The registered capital of each power sales company is processed using the minimum-maximum normalization method. a The normalized registered capital of each power sales company is ; The larger the registered capital, the stronger the risk resistance of the power sales company; S3.

3. The scoring results in step S3.1 and the normalized results in step S3.2 constitute the sample data of the power sales company ,The power sales companies are classified by GMM clustering algorithm, the number of categories is 3, and three categories are obtained; S3.4 defines three categories as high risk, medium risk, and low risk, and codes the three categories as 1, 2, and 3.

4. A method for selecting and recommending a power sales company for retail power users according to claim 1, characterized in that: The specific steps of step S3.3 are as follows: S3.3.

1. Initialize the model parameters of GMM clustering , , ; For the q The weight of each sub-model, For the q The mean of the sub-models, For the q The covariance of the sub-models; S3.3.

2. Calculate the a Sample data of electricity sales companies By q The probability generated by the sub-model ,in: , For the l The weight of each sub-model, For sample In the q The probability density function of a Gaussian distribution is For sample In the l The probability density function of a Gaussian distribution; S3.3.

3. Update the model parameters according to the following formula: ， ， in, , , Respectively q The updated mean, covariance, and weight of each sub-model, m is the total number of power sales companies; then repeat steps S3.3.1 and S3.3.2 according to the new model parameters and stop when the maximum number of iterations is reached; S3.3.

4. Each sample is divided into into the corresponding clusters, Indicates a The maximum probability of a power sales company in the sub-model is obtained, that is, each sample is classified into the cluster of the model with the highest probability of coming from the sub-model, and finally three categories are obtained.

5. According to claim 1, a method for selecting and recommending a power sales company for retail power users is characterized in that: Step S4 specifically includes: S4.

1. m The total number of transactions and the total number of transaction users of each power sales company constitute the sample data, and the minimum-maximum normalization method is used for processing; S4.2, clustering the normalized data in step S4.1 by using the GMM clustering algorithm, the number of categories is 3, and three categories are obtained; the GMM clustering algorithm process is the same as step S3.3; S4.

3. Define the three categories divided in step S4.2 as low market influence, medium market influence, and high market influence, and code them as 1, 2, and 3.

6. A method for selecting and recommending a power sales company for retail power users according to claim 1, characterized in that: In step S5, according to the packages provided by the power sales company, the average power sales price is obtained to determine whether it provides green electricity, and a characteristic matrix of the power sales company is constructed based on the risk category, market influence category, green electricity situation, and average power sales price of the power sales company. The specific method is as follows: S5.

1. Input the retail package of the electricity sales company. If the environmental premium ratio in the package is not zero, it is recorded as green electricity and coded as 1; if the environmental premium ratio in the package is zero, it is recorded as non-green electricity and coded as 0; S5.

2. According to Calculate the The average electricity price of each electricity sales company, For the Electricity sales company r The price of a power package. T a For the The total number of packages of each electricity sales company; S5.

3. Perform minimum-maximum normalization on the average electricity selling price of the electricity selling company; record the normalized average electricity selling price as ,if If it is less than 0.5, the average electricity price is considered low and coded as 1; If it is greater than 0.5, the average electricity price is considered high and coded as 2; S5.

4. Construct a characteristic matrix of power sales companies based on three aspects: risk category, market influence category, and green electricity situation.

7. A method for selecting and recommending a power sales company for retail power users according to claim 1, characterized in that: In step S6, if manual selection is selected, it is necessary to independently select the codes of the three characteristics of the power sales company's risk category, market influence category, and green power situation, and obtain the power sales company that matches the selected code; If you select automatic matching, recommendations are made based on the following rules: For users who consume a lot of electricity and are sensitive to prices, priority will be given to power companies with low average power prices and low risks; For users who consume a lot of electricity and are not sensitive to prices, low-risk power sales companies are recommended first; For users who have low electricity consumption and are price-sensitive, priority will be given to power companies with low average electricity prices; For users with low electricity consumption and insensitive to prices, priority will be given to electricity sales companies with great market influence; Green users give priority to power companies that sell green electricity.

8. A system for implementing the method for selecting and recommending a power sales company for retail power users as described in claim 1, characterized in that: include: User classification module: The K-Medoids clustering method based on cosine similarity is used to cluster the power consumption curves of retail power users, and two types of power consumption curves are obtained. Based on these two types of curves, users are divided into two categories: non-flat type and flat type. Retail power user characteristic matrix construction module: classifies retail power users by electricity demand and green users, and constructs a characteristic matrix of retail power users based on the electricity curve pattern, electricity demand category and coding of green user category of retail power users; Power sales company risk coding module: Encode the credit rating of the power sales company, use the GMM clustering algorithm to classify the risk of the power sales company based on the credit rating and registered capital of the power sales company, and encode the risk classification results; Power sales company market influence coding module: Based on the total number of transactions and the total number of transaction users of the power sales company, the GMM clustering algorithm is used to classify the market influence of the power sales company, and the market influence classification results are encoded; Power sales company feature matrix construction module: According to the packages provided by the power sales company, the average power sales price is obtained to determine whether green electricity is provided, and the power sales company's feature matrix is ​​constructed based on the risk category, market influence category, green electricity situation and average power sales price of the power sales company; Power sales company selection and recommendation module: Based on the feature matrix of retail power users and power sales companies and the feature coding matching principle, manual selection method or automatic matching method is used to select and recommend power sales companies.

Images