An information recommendation method based on federated learning and energy internet

Abstract

The application provides an information recommendation method based on federated learning and an energy internet, comprising the following steps: a local work platform acquires a local feature data set; the local work platform adopts a horizontal federated learning algorithm to perform federated training on the local feature data set, and obtains a horizontal federated recommendation model and local training model parameters; the local work platform, a server, a local energy supplier client and a local energy demander client update the local training model parameters; and the horizontal federated recommendation model pushes demand information of the local energy supplier client or demand information of the local energy demander client to the local energy demander client or the local energy supplier client according to individual requirements. The information recommendation method based on federated learning and the energy internet has the characteristics of individualization, strong privacy, simple process, accurate prediction and the like, and can be widely applied to the field of internet information recommendation.

Description

Technical Field

[0001] This invention relates to information recommendation technology, and in particular to an information recommendation method based on federated learning and the energy internet. Background Technology

[0002] The Energy Internet, comprised of the Internet and smart energy, represents the overall vision for the future energy supply and consumption system. It is a comprehensive system providing integrated energy services, distributed energy stations, and energy trading based on a big data platform. With the development of new energy technologies, a wide range of energy suppliers with varying scales of energy supply capabilities will emerge. A new challenge for the Energy Internet is how to coordinate energy supply and demand information, optimize the allocation of energy supplier resources, and meet the energy needs of residential, commercial, and industrial users based on energy interaction demands. Simultaneously, the Energy Internet should also meet the personalized needs of users.

[0003] Currently, information recommendation systems work by uploading users' historical information to a central database, where recommendation models are then trained on this information; information is then recommended to users based on the training results. However, the process of users uploading their historical information to the central database carries the risk of leaking user privacy.

[0004] Therefore, it is evident that, in the current technology, there is no energy information recommendation method that can both meet users' personalized needs and protect user privacy. Summary of the Invention

[0005] In view of this, the main objective of the present invention is to provide an information recommendation method based on federated learning and the energy internet that can both meet users' personalized needs and protect user privacy.

[0006] To achieve the above objectives, the technical solution proposed by this invention is as follows:

[0007] An information recommendation method based on federated learning and the energy internet includes the following steps:

[0008] Step 1: The local working platform will collect local energy supplier characteristic datasets and local energy demander characteristic datasets to form a local characteristic dataset; wherein, the number of local energy suppliers is more than one and the number of local energy demanders is also more than one.

[0009] Step 2: The local working platform uses a horizontal federated learning algorithm to perform federated training on the local feature dataset, obtaining the parameters of the horizontal federated recommendation model and the local training model. The local working platform updates its own local training parameters. The local working platform also sends the local training model parameters to the local energy supplier client, the local energy demander client, and the server. The server, the local energy supplier client, and the local energy demander client update the local training model parameters.

[0010] Step 3: The horizontal federated recommendation model pushes the demand information of the local energy supplier client or the local energy demander client to the corresponding local energy demander client or local energy supplier client according to the personalized requirements of the local energy supplier client or the local energy demander client.

[0011] In summary, in the information recommendation method based on federated learning and the energy internet described in this invention, the local working platform collects all local energy supplier feature datasets and local energy demander feature datasets to form a local training dataset. The local working platform uses a horizontal federated learning algorithm to train the local training dataset, obtaining the horizontal federated recommendation model and the parameters of the local training model. The local working platform updates its own local training model parameters and sends them to the server, all local energy supplier clients, and local energy demander clients. All local energy supplier clients and local energy demander clients update their own local training model parameters. After the update, the local working platform selects the appropriate local energy supplier client or local energy demander client based on their personalized requirements. As can be seen, the information recommendation method based on federated learning and the energy internet described in this invention has a relatively simple process, without needing to consider the cumbersome details of each local energy supplier client or local energy demander client. Moreover, since the local working platform, server, and local energy supplier clients or local energy demander clients all update the parameters of the locally trained model in a timely manner, the trained federated recommendation model makes relatively accurate predictions. In addition, the information recommendation method based on federated learning and the energy internet described in this invention also considers the personalized needs of each local energy supplier client or local energy demander client. Therefore, the information recommendation method based on federated learning and the energy internet described in this invention has broad application value. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall process of an information recommendation method based on federated learning and the energy internet as described in this invention.

[0013] Figure 2 This is a schematic diagram of the overall framework of the Internet-based energy supply and demand system described in this invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] Figure 1 This is a schematic diagram illustrating the overall process of an information recommendation method based on federated learning and the energy internet as described in this invention. Figure 1 As shown, the information recommendation method based on federated learning and the energy internet of the present invention includes the following steps:

[0016] Step 1: The local working platform will collect local energy supplier characteristic datasets and local energy demander characteristic datasets to form a local characteristic dataset; wherein, the number of local energy suppliers is more than one and the number of local energy demanders is also more than one.

[0017] In step 1 of this invention, the local energy supplier feature dataset is a collection of energy supplier feature information and its corresponding feature data extracted by the local working platform from the local energy supplier client; the local energy demander feature dataset is a collection of energy demand feature information and its corresponding feature data extracted by the local working platform from the local demander client. The local energy supplier feature information includes: supply type, supply inventory, and maximum supply quantity; the energy demander feature information includes: demand type, demand quantity, and demand time.

[0018] Step 2: The local working platform uses a horizontal federated learning algorithm to perform federated training on the local feature dataset, obtaining the parameters of the horizontal federated recommendation model and the local training model. The local working platform updates its own local training parameters. The local working platform also sends the local training model parameters to the local energy supplier client, the local energy demander client, and the server. The server, the local energy supplier client, and the local energy demander client update the local training model parameters.

[0019] Step 3: The horizontal federated recommendation model pushes the demand information of the local energy supplier client or the local energy demander client to the corresponding local energy demander client or local energy supplier client according to the personalized requirements of the local energy supplier client or the local energy demander client.

[0020] In summary, the information recommendation method based on federated learning and the energy internet described in this invention involves a local working platform collecting all local energy supplier feature datasets and local energy demander feature datasets to form a local training dataset. The local working platform then trains the local training dataset using a horizontal federated learning algorithm to obtain the parameters of the horizontal federated recommendation model and the local training model. The local working platform updates its own local training model parameters and sends these parameters to the server, all local energy supplier clients, and all local energy demander clients. Each local energy supplier client and local energy demander client updates its own local training model parameters. After the update, the local working platform selects the appropriate parameters based on the personalized requirements of the local energy supplier clients or local energy demander clients. As can be seen, the information recommendation method based on federated learning and the energy internet described in this invention has a relatively simple process, without needing to consider the cumbersome details of each local energy supplier client or local energy demander client. Moreover, since the local working platform, server, and local energy supplier clients or local energy demander clients all update the parameters of the locally trained model in a timely manner, the trained federated recommendation model makes relatively accurate predictions. In addition, the information recommendation method based on federated learning and the energy internet described in this invention also considers the personalized needs of each local energy supplier client or local energy demander client. Therefore, the information recommendation method based on federated learning and the energy internet described in this invention has broad application value.

[0021] In practical applications, the energy supply and demand system involved in the information recommendation method based on federated learning and the energy internet described in this invention includes: a server, a local working platform, and more than one supply and demand group, each of which includes more than one local energy supplier client and more than one local energy demander client. Figure 2 This is a schematic diagram of the overall framework of the Internet-based energy supply and demand system described in this invention. Figure 2 In the embodiment described, the Internet-based energy supply and demand system includes a supply and demand group, which includes M local energy suppliers B1, B2, ..., BM and N local energy demanders A1, A2, ..., AN. Local energy supplier B1 includes n local energy supplier clients B11, B12, ..., B1n; local energy supplier BM includes p local energy supplier clients BM1, BM2, ..., BMp; local energy demander A1 includes m local energy demand clients A11, A12, ..., A1m; and local energy demander AN includes q local energy demand clients A11, A12, ..., A1q. Wherein, M, N, m, n, p, and q are all natural numbers. Figure 2The system described in this embodiment operates in accordance with the information recommendation method based on federated learning and the energy internet as described in this invention.

[0022] In this invention, step 1 specifically includes the following steps:

[0023] Step 11: The local working platform performs correlation analysis on each energy supplier and each energy demander to determine each supply-demand group, and identifies the local energy supplier and local energy demander in each supply-demand group; the local energy supplier and local energy demander are corresponding; wherein, the number of the supply-demand groups is more than one.

[0024] Step 12: The local work platform further conducts correlation analysis on all local energy suppliers and all local energy demanders in each supply and demand group to obtain characteristic information of each local energy supplier and each local energy demander.

[0025] Step 13: For each supply and demand group, the local working platform collects the characteristic information of each local energy supplier and the corresponding data of each local energy demander to form the characteristic datasets of each local energy supplier and each local energy demander.

[0026] Step 14: Align and normalize the feature datasets of each local energy supplier and each local energy demander, and use the corresponding energy supplier feature vectors and energy demander feature vectors as local sample training sets.

[0027] In this invention, step 2 specifically includes the following steps:

[0028] Step 21: The local working platform uses a horizontal federated learning algorithm to perform federated training on the local training sample set.

[0029] Step 22: The horizontal federated learning algorithm calculates the predicted energy demand, energy loss function, and gradient of the characteristic information of each local energy supplier client for each local energy demander client; at the same time, it generates local training model parameters; the local working platform updates the local training model parameters and sends the local training model parameters to the server, local energy supplier clients, and local energy demander clients.

[0030] Step 23: The server determines whether the horizontal federated learning algorithm has converged. If it has converged, the horizontal federated learning algorithm is used as the horizontal federated recommendation model, and step 24 is executed. If it has not converged, the server returns to step 21 until it converges.

[0031] Step 24: The server, each local energy supplier client, and each local energy demander client update the parameters of the local training model.

[0032] In this invention, step 3 specifically includes the following steps:

[0033] Step 31: The horizontal federated recommendation model obtains the demand list of each local energy supplier and the demand list of each local energy demander in the corresponding supply and demand group, and obtains the current supply demand characteristic data in the demand list of each local energy supplier and the current demand demand characteristic data in the demand list of each local energy demander.

[0034] Step 32: The horizontal federated recommendation model predicts and processes the current supply demand of each local energy supplier or the current demand of each local energy demander: sorts all predicted values ​​of the current supply demand of each local energy demander in descending order; sorts all predicted values ​​of the current demand of each local energy supplier in descending order.

[0035] Step 33: Obtain the personalized requirements of each local energy supplier client or each local energy demander client using the horizontal federated recommendation model.

[0036] Step 34: The horizontal federated recommendation model recommends the information corresponding to the maximum predicted value of the current supply demand of each local energy supplier to each local energy demander client in the corresponding supply-demand group according to the personalized requirements of each local energy supplier client; the horizontal federated recommendation model sends the information corresponding to the maximum predicted value of the current demand of each local energy demander to each local energy supplier client in the corresponding supply-demand group according to the personalized requirements of each local energy demander client.

[0037] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An information recommendation method based on federated learning and the energy internet, characterized in that, The information recommendation method includes the following steps: Step 1: The local work platform will construct a local feature dataset from the collected local energy supplier feature dataset and local energy demander feature dataset; wherein there is more than one local energy supplier and more than one local energy demander; specifically including the following steps: Step 11: The local working platform performs correlation analysis on each energy supplier and each energy demander to determine each supply-demand group, and identifies the local energy supplier and local energy demander in each supply-demand group; the local energy supplier and local energy demander are corresponding; wherein, the number of the supply-demand groups is more than one. Step 12: The local work platform further conducts correlation analysis on all local energy suppliers and all local energy demanders in each supply and demand group to obtain characteristic information of each local energy supplier and characteristic information of each local energy demander. Step 13: For each supply and demand group, the local working platform collects the characteristic information of each local energy supplier and the corresponding data of each local energy demander to form the characteristic dataset of each local energy supplier and the characteristic dataset of each local energy demander. Step 14: Align and normalize the feature datasets of each local energy supplier and each local energy demander, and use the corresponding energy supplier feature vectors and energy demander feature vectors as local sample training sets. Step 2: The local working platform uses a horizontal federated learning algorithm to perform federated training on the local feature dataset, obtaining the parameters of the horizontal federated recommendation model and the locally trained model. The local working platform then updates its own local training parameters. The local working platform also sends the local training model parameters to the local energy supplier client, the local energy demander client, and the server. The server, the local energy supplier client, and the local energy demander client update their local training model parameters. Specifically, this includes the following steps: Step 21: The local working platform uses a horizontal federated learning algorithm to perform federated training on the local training sample set; Step 22: The horizontal federated learning algorithm calculates the predicted energy demand, energy loss function, and gradient of the characteristic information of each local energy supplier client for each local energy demander client; at the same time, it generates local training model parameters; the local working platform updates the local training model parameters and sends the local training model parameters to the server, local energy supplier clients, and local energy demander clients. Step 23: The server determines whether the horizontal federated learning algorithm has converged. If it has converged, the horizontal federated learning algorithm is used as the horizontal federated recommendation model, and step 24 is executed. If it has not converged, the server returns to step 21 until it converges. Step 24: The server, each local energy supplier client, and each local energy demander client update the parameters of the local training model. Step 3: The horizontal federated recommendation model pushes the demand information of the local energy supplier client or the local energy demander client to the corresponding local energy demander client or local energy supplier client according to the personalized requirements of the local energy supplier client; specifically, it includes the following steps: Step 31: The horizontal federated recommendation model obtains the demand list of each local energy supplier and the demand list of each local energy demander in the corresponding supply and demand group, and obtains the current supply demand characteristic data and the current demand demand characteristic data in the demand list of each local energy supplier. Step 32: The horizontal federated recommendation model predicts and processes the current supply demand of each local energy supplier or the current demand of each local energy demander: sorts all predicted values ​​of the current supply demand of each local energy demander in descending order; sorts all predicted values ​​of the current demand of each local energy supplier in descending order. Step 33: Obtain the personalized requirements of each local energy supplier client or each local energy demander client using the horizontal federated recommendation model. Step 34: The horizontal federated recommendation model recommends the information corresponding to the maximum predicted value of the current supply demand of each local energy supplier to each local energy demander client in the corresponding supply-demand group according to the personalized requirements of each local energy supplier client; the horizontal federated recommendation model sends the information corresponding to the maximum predicted value of the current demand of each local energy demander to each local energy supplier client in the corresponding supply-demand group according to the personalized requirements of each local energy demander client.

2. The information recommendation method based on federated learning and the energy internet according to claim 1, characterized in that, In step 1, the local energy supplier feature dataset is a collection of energy supplier feature information and its corresponding feature data extracted by the local working platform from the local energy supplier client; the local energy demander feature dataset is a collection of energy demand feature information and its corresponding feature data extracted by the local working platform from the local demander client.

3. The information recommendation method based on federated learning and the energy internet according to claim 2, characterized in that, The local energy supplier characteristics include: supply type, supply inventory, and maximum supply volume; the energy demander characteristics include: demand type, demand quantity, and demand time.

Images