Research On Power Selling Package Recommendation Method Of Power Market Based On Spark

The issuance of the 9th power reform document clarifies the key points and ways for the country to deepen the reform of the power industry.It requires that the power-supply side liberalizes,encourages the development of market-oriented power selling companies,introduces competition in power sales,restores the attributes of power products,and gives users the freedom to choose right.As an important power commodity,the power selling package is a carrier for the power selling company to provide power services to users.In order to gain an advantage in the market competition,the power selling company enhances the refined service level to the customer from all aspects and improves the customer satisfaction.It is one of the means to recommend a power selling package.Pushing the appropriate power selling package service for users is an effective way to directly increase the viscosity of the existing power selling companies and attract a large number of new users for the power selling companies,while the power selling company analyzes and excavates the mass data on the user side and understands the user power consumption.The behavior is also the necessary way to implement the package push.This issue begins with three aspects of the proposed method of power selling packages.Aiming at the large number of users of power selling packages and the inability to guarantee the efficiency of recommendation,a user optimal feature subset discovery algorithm based on weighted incremental item coverage is proposed.The optimal feature subset is used to represent all users and the user group is reduced.First,defining the coverage of weighted incremental items to solve the problem that unpopular packages are easily covered by popular packages,and increase the exposure rate of unpopular packages.Then the concept of the optimal feature subset of package trading users is presented to represent the interest preferences of all users.Finally,the user optimal feature subset is screened in combination with the weighted-increasing item coverage rate,and the user's search scope is narrowed down for subsequent packages to save computing resources.The experiment verifies the representativeness of the user optimal feature subset and the validity of the discovery method.Aiming at the complex and diverse contents of power selling packages in power market transactions and the difficulty of user selection,an power selling package recommendation algorithm based on user optimal feature subsets and bidirectional prediction is proposed.Firstly,a similarity calculation method of power selling packages based on attribute correlation is presented.The attribute weight value of package is obtained by using the analytic hierarchy process,and the similarity is obtained by calculating the similarity of the electricity sales package in combination with the weight.Then use the similarity matrix to find the target package neighborhood set,and get the target user's initial prediction score for the unrated package.Then the target user is compared with each user in the optimal feature subset to obtain the target user's neighbor set,and the final prediction score for the ungraded package is calculated in combination with the time function.Finally,selecting the set of recommended results according to the result of the rating.The experiment verifies the accuracy of the proposed algorithm.In the recommendation process,it is often necessary to mine user behavior in a large amount of data.In order to quickly respond to user needs,a large-data processing capability is required for implementation of the power selling package recommendation system.Therefore,a Spark-based selling package recommendation system based on the proposed algorithm is established.The overall architecture of the system is designed.The main functions of the system are implemented of user interaction,data collection and storage,data preprocessing,and package recommendation engine.In the experiment,the performance of the recommendation engine and the functions of the system were tested,and it was verified that the Spark package recommendation system has good big data processing capability and high availability.