Inter-cell load balancing method and device, readable storage medium and program product
By constructing graph structure and graph neural network model, predicting inter-cell user migration and network speed, the problem of inter-cell load imbalance in communication networks is solved, and the inter-cell load balancing and energy saving efficiency are improved.
Patent Information
- Application Number
- CN202411270629.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-12-10
AI Technical Summary
The energy-saving strategies of the existing communication network cannot accurately locate the energy-saving cells that need to be closed, resulting in unbalanced network load and affecting communication quality and user experience.
By obtaining historical data of the target cell and adjacent cells, a graph structure and pre-trained graph neural network model are constructed, the number of user migrations and network speed are predicted, the load balancing between cells is judged, and the antenna transmission power is adjusted according to the prediction results to achieve load balancing .
It realizes inter-cell load balancing, improves network performance and energy saving efficiency, and ensures user experience and communication quality.
Smart Images
Figure CN119110352A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of wireless communication technology, and in particular to a method and device for load balancing between cells, a readable storage medium, and a computer program product. Background Art
[0002] In current communication networks, energy-saving strategies are formulated to shut down certain cells and migrate users in energy-saving cells to other cells in order to save energy. Existing energy-saving strategies often cannot accurately locate energy-saving cells that need to be shut down. After users migrate from shut-down cells to other cells, some cells in the communication network are overloaded while other cells are lightly loaded, resulting in uneven network performance. This will affect the overall network communication quality and user experience.
[0003] How to strike a good balance between cell energy saving and load balancing to improve energy saving efficiency and network performance is a technical problem that needs to be solved at present. Summary of the invention
[0004] The purpose of the embodiments of the present application is to provide a method and device for load balancing between cells, a readable storage medium, and a computer program product to solve the problem of low communication quality caused by energy saving in cells of existing communication networks.
[0005] In order to solve the above technical problems, this specification is implemented as follows:
[0006] In a first aspect, a method for load balancing between cells is provided, comprising:
[0007] Acquire historical data of a target cell and adjacent cells of the target cell, wherein the historical data includes the number of users in the cell, received signal strength, and network speed;
[0008] Based on the graph structure constructed by the historical data and the pre-trained graph neural network model, the number of users migrated from the target cell to each adjacent cell after the target cell is closed is predicted, the graph structure uses the target cell and each adjacent cell as nodes, and the user migration ratio between the target cell and each adjacent cell as edges, the graph neural network model uses the graph structure constructed corresponding to the energy-saving cell and the corresponding non-energy-saving adjacent cell as samples, and the number of users in the energy-saving cell to migrate to each non-energy-saving adjacent cell after the energy-saving cell is closed as labels for training, and the user migration ratio is related to the received signal strength of each adjacent cell corresponding to the user;
[0009] Based on the historical data and the predicted number of users, predict the network speed of each adjacent cell;
[0010] Based on the predicted number of users and the predicted network speed, determining whether adjacent cells are load balanced after the target cell is shut down;
[0011] If the loads of the adjacent cells are balanced, the target cell is shut down.
[0012] Optionally, the graph structure constructed based on the historical data and the pre-trained graph neural network model predicts the number of users that migrate from the target cell to each adjacent cell after the target cell is closed, including:
[0013] Determine, based on the maximum received signal strength of each user in the target cell corresponding to each adjacent cell in the historical data, a migration ratio of each user in the target cell to each adjacent cell after the target cell is closed;
[0014] A graph structure is constructed with the target cell and each adjacent cell as nodes and the migration ratio between the target cell and each adjacent cell as edges, wherein the characteristics of each node include the number of users, received signal strength and network speed of the cell in the corresponding historical data, and the weight of each edge is the corresponding migration ratio;
[0015] The graph structure is input into a pre-trained graph neural network model to obtain the number of users who migrate from the target cell to each adjacent cell after the target cell is closed.
[0016] Optionally, the determining, based on the maximum received signal strength of each user in the target cell corresponding to each neighboring cell in the historical data, a migration ratio of each user in the target cell to each neighboring cell after the target cell is shut down, includes:
[0017] Determine the number of users with the maximum received signal strength in the target cell corresponding to the target adjacent cell based on the received signal strength in the historical data;
[0018] The ratio of the number of users corresponding to the maximum received signal strength to the total number of users in the target cell is determined as the migration ratio of users migrating from the target cell to the target adjacent cell after the target cell is closed.
[0019] Optionally, predicting the network speed of each adjacent cell based on the historical data and the predicted number of users includes:
[0020] Respectively obtain characteristics of the number of users, received signal strength, network speed, and corresponding predicted number of users of the target adjacent cell in the historical data;
[0021] The acquired features are combined and input into a convolutional neural network model to obtain the average network speed of the target adjacent cell. The convolutional neural network model predicts the average network speed of the cell based on the features of the cell including the number of users, received signal strength, network speed and the corresponding predicted number of users.
[0022] Optionally, judging whether adjacent cells are load balanced after the target cell is shut down based on the predicted number of users and the predicted network speed includes:
[0023] Determining the number of users after migration in each adjacent cell based on the sum of the number of users in each adjacent cell in the historical data and the corresponding predicted number of users;
[0024] Obtaining the network speed of each adjacent cell in the historical data;
[0025] Calculating a change value between a network speed of each adjacent cell in the historical data and a corresponding predicted network speed of each adjacent cell;
[0026] Respectively comparing the number of users after migration corresponding to each adjacent cell with a preset user number threshold, the predicted network speed corresponding to each adjacent cell with a preset network speed threshold, and the network speed change value corresponding to each adjacent cell with the preset network speed change threshold;
[0027] If each adjacent cell satisfies the following conditions: the number of users after migration is less than a preset user number threshold, the predicted network speed is greater than a preset network speed threshold, and the network speed change value is less than a preset network speed change threshold, then it is determined that the load of each adjacent cell is balanced after the target cell is closed;
[0028] Otherwise, it is determined that the loads of the adjacent cells are unbalanced after the target cell is shut down.
[0029] Optionally, after judging whether the adjacent cells are load balanced after the target cell is shut down based on the predicted number of users and the predicted network speed, the method further includes:
[0030] If the loads of adjacent cells are unbalanced, adjust the antenna transmission power of each adjacent cell in a simulated manner;
[0031] Predicting the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmit power;
[0032] Based on the predicted received signal strength, updating the received signal strength in the historical data;
[0033] Based on the updated historical data, repeating the steps of predicting the number of users migrating from the target cell to each adjacent cell after the target cell is shut down, to determining whether each adjacent cell is load balanced after the target cell is shut down;
[0034] If the loads of the adjacent cells are balanced, the target cell is shut down and the antenna transmission power of the adjacent cells is adjusted to the simulated adjusted antenna transmission power.
[0035] Optionally, the predicting the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmit power includes:
[0036] The simulated adjusted antenna transmit power of the target adjacent cell is input into a pre-trained multilayer perceptron model to obtain the received signal strength of the target adjacent cell. The multilayer perceptron model learns the relationship between the antenna transmit power and the received signal strength of the cell based on the antenna transmit power of different cells and the corresponding received signal strength, so as to predict the received signal strength of the cell.
[0037] In a second aspect, a load balancing device between cells is provided, comprising a processor and a memory, wherein the memory stores a program or instruction that can be executed on the processor, and when the program or instruction is executed by the processor, the steps of the method described in the first aspect are implemented.
[0038] In a third aspect, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.
[0039] According to a fourth aspect, a computer program product is provided, the computer program product comprising a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to execute the steps of the method according to the first aspect.
[0040] In an embodiment of the present application, historical data of a target cell and each adjacent cell of the target cell are obtained, wherein the historical data include the number of users in the cell, the received signal strength, and the network speed; a graph structure constructed based on the historical data and a pre-trained graph neural network model are used to predict the number of users migrating from the target cell to each adjacent cell after the target cell is closed. The graph structure uses the target cell and each adjacent cell as nodes, and the user migration ratio between the target cell and each adjacent cell as edges. The graph neural network model uses the graph structure constructed corresponding to the energy-saving cell and the corresponding non-energy-saving adjacent cell as a sample, and the number of users in the energy-saving cell to which each non-energy-saving adjacent cell is migrated after the energy-saving cell is closed as the sample. The quantity is obtained by label training, and the user migration ratio is related to the received signal strength of each adjacent cell corresponding to the user; based on the historical data and the predicted number of users, the network speed of each adjacent cell is predicted; based on the predicted number of users and the predicted network speed, it is judged whether the adjacent cells are load balanced after the target cell is closed; if the adjacent cells are load balanced, the target cell is closed, thereby the graph neural network can be used to accurately predict the number of user migrations and adjacent cells, and by analyzing the cell load balancing after user migration, the balanced distribution of network load is achieved, the load balancing of each cell is ensured, and the performance of the overall network is improved, thereby balancing the relationship between cell energy saving and load balancing well, and improving energy saving efficiency and network performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0042] Figure 1 It is a flow chart of the load balancing method between cells according to an embodiment of the present application.
[0043] Figure 2 It is an overall flow chart of the inter-cell load balancing method according to an embodiment of the present application.
[0044] Figure 3 It is a structural block diagram of the inter-cell load balancing device of an embodiment of the present application. DETAILED DESCRIPTION
[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application. The numbering of the drawings in this application is only used to distinguish the various steps in the scheme, and is not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.
[0046] In order to solve the problems existing in the prior art, the embodiments of the present application provide a method and device for load balancing between cells, a readable storage medium, and a computer program product.
[0047] Figure 1 is a flow chart of the load balancing method between cells according to an embodiment of the present application. Figure 1 As shown, the following steps 102 to 110 are included.
[0048] Step 102, obtaining historical data of the target cell and each adjacent cell of the target cell, wherein the historical data includes the number of users in the cell, received signal strength, and network speed.
[0049] The target cell is a cell that is temporarily closed due to energy saving needs when formulating an energy saving strategy. Each adjacent cell is a neighboring cell of the target cell configured in the communication network. One target cell may be configured with multiple adjacent cells.
[0050] In the preparation stage, historical data of relevant cells are first collected, including: 1. The number of users in the cell, which can be determined by collecting the number of historical radio resource control (RRC) connections in the cell. A user equipment (UE) is connected to an RRC in the communication network, and all service signaling can be carried out through this RRC connection. There is at most one RRC connection between each user equipment and the base station; 2. The received signal strength of the cell, which can be determined by collecting the reference signal receiving power (RSRP) of the cell; 3. The network speed of the cell, which can be determined by the total amount of cell service data collected per unit time, including uplink and downlink traffic.
[0051] Furthermore, the historical data also includes the energy-saving status of each cell, for example, if there is a closed cell among the adjacent cells, the adjacent cell will be removed in advance. It also includes information such as the network topology of each cell. By analyzing these historical data, we can more accurately understand the operation rules of the network and user behavior patterns.
[0052] Step 104, based on the graph structure constructed by the historical data and the pre-trained graph neural network model, predict the number of users who migrate from the target cell to each adjacent cell after the target cell is closed, the graph structure has the target cell and each adjacent cell as nodes, and the user migration ratio between the target cell and each adjacent cell as edges, the graph neural network model is trained with the graph structure constructed corresponding to the energy-saving cell and the corresponding non-energy-saving adjacent cell as samples, and the number of users in the energy-saving cell to which each non-energy-saving adjacent cell is migrated after the energy-saving cell is closed as labels, and the user migration ratio is related to the received signal strength of each adjacent cell corresponding to the user.
[0053] In an embodiment of the present application, a graph structure with an association relationship is constructed based on the features corresponding to the historical data of each cell, and the graph structure is input into a pre-trained graph neural network model to predict the number of users who migrate from the target cell to each adjacent cell after the target cell is closed.
[0054] The historical data obtained in step 102 is used to construct a graph structure, G = {V, E} represents the graph, where G is a graph, V is a node set, and E is an edge set. The cells are nodes of the graph, the source nodes of the target cell are connected to the destination nodes of each adjacent cell, and the user migration ratio between the connected nodes is used as the edge of the graph.
[0055] The migration ratio of a user is related to the received signal strength of each adjacent cell corresponding to the user, and can be obtained through measurement report (MR) data.
[0056] Based on the scheme provided in the above embodiment, optionally, in the above step 104, the graph structure constructed based on the historical data and the pre-trained graph neural network model predict the number of users who migrate from the target cell to each adjacent cell after the target cell is closed, including: determining the migration ratio of each user in the target cell to each adjacent cell after the target cell is closed based on the maximum received signal strength of each user in the target cell corresponding to each adjacent cell in the historical data; constructing a graph structure with the target cell and each adjacent cell as nodes and the migration ratio between the target cell and each adjacent cell as edges, the characteristics of each node including the number of users in the corresponding cell in the historical data, the received signal strength and the network speed, and the weight of each edge is the corresponding migration ratio; inputting the graph structure into the pre-trained graph neural network model to obtain the number of users who migrate from the target cell to each adjacent cell after the target cell is closed.
[0057] Specifically, the method of determining the migration ratio of each user in the target cell to each adjacent cell after the target cell is closed based on the maximum received signal strength of each user in the target cell corresponding to each adjacent cell in the historical data includes: determining the number of users in the target cell with the maximum received signal strength corresponding to the target adjacent cell based on the received signal strength in the historical data; and determining the ratio of the number of users with the maximum received signal strength to the total number of users in the target cell as the migration ratio of users from the target cell to the target adjacent cell after the target cell is closed.
[0058] For example, cell i is the target cell, and N different users all use cell i as the main cell. The MR data of the N users are obtained respectively. The RSRP size of each user's corresponding adjacent cells in the MR data is sorted to determine the adjacent cell with the largest RSRP. The adjacent cell with the largest RSRP is used as the adjacent cell to which the corresponding user is to be migrated after cell i is closed.
[0059] Through statistics, the M users in cell i have the largest RSRP in cell j, so cell j can be determined as the neighboring cell to be migrated by the M users in cell i. After cell i is closed, the migration ratio of the corresponding users that may migrate to cell j is:
[0060]
[0061] Wherein, M is the number of users in cell i with the maximum received signal strength corresponding to the adjacent cell j, and N is the total number of users in cell i.
[0062] After the target cell is closed for energy saving, the migration ratio of each user in the target cell from the target cell to each adjacent cell can be represented by the migration adjacency matrix A. If a user migrates from cell j to cell j, then the migration adjacency matrix A in row i and column j is ij In the migration adjacency matrix, it corresponds to a non-zero value, otherwise it is zero.
[0063] Migration ratio A ij The value of can be considered as, if cell i is shut down by energy saving operation, the users in cell i will have A ij Therefore, if the migration ratio is large, it means that the relationship between the two cells is close, the user can receive a stronger signal strength from cell j, and the communication quality is better.
[0064] The above calculation process is performed on all adjacent cells, and the obtained migration ratio is filled into the corresponding element position of the migration adjacency matrix A, and the migration adjacency matrix of users between the target cell and each adjacent cell is obtained. The migration ratio in the migration adjacency matrix A is used as the weight of the edge from the source node to the destination node.
[0065] Therefore, the user migration adjacency matrix is constructed through the received signal strength of MR data, which accurately reflects the user flow relationship between cells and provides data support for subsequent cell energy-saving strategies and load balancing.
[0066] The node characteristics of each cell (including the target cell and each adjacent cell) in the graph structure may be determined based on the number of users, received signal strength, and network speed of the cell in the historical data corresponding to each cell.
[0067] For example, for cell j, collect the number of users in historical data user j 、Average network speed avg_speed j , and the distribution parameters based on MR grid statistics, extract the maximum value max_rsrp of the received signal strength of cell j j 、mean mean_rsrp j , variance var_rsrp j , forming the node feature X for cell j j The description vector is expressed as follows:
[0068] X j ={max_rsrp j ,mean_rsrp j ,var_rsrp j ,user A ,avg_speed j}.
[0069] Thus, a graph structure can be constructed with the target cell and each adjacent cell as nodes and the migration ratio between the target cell and each adjacent cell as edges.
[0070] By integrating cell characteristic information in multiple dimensions, including the number of users, average network speed, and received signal strength-related parameters, a description vector of each cell node's characteristics is constructed, which can provide a comprehensive description of cell performance and support cell energy saving and performance optimization decisions.
[0071] By inputting the constructed graph structure into the pre-trained graph neural network model, the number of users who migrated from the target cell to each adjacent cell after the target cell was closed can be obtained.
[0072] The characteristics of each cell are not only affected by its own historical characteristics, but also by the characteristics of its neighboring cells. This influence can be well measured using a graph, thereby achieving more accurate prediction and energy saving.
[0073] The pre-trained graph neural network model can adopt a multi-layer graph convolutional network (GCN) model. The graph convolutional network is a deep learning model that specializes in processing graph structured data. It can effectively use the structural information and node features in the graph to capture neighborhood information.
[0074] The input of the graph convolutional network model is a graph structure constructed based on the target cell and adjacent cells, including node features X and the migration adjacency matrix A representing the relationship between cells.
[0075] In one embodiment, a vector indicating whether each cell is in a closed state can also be introduced into the input, which can be a binary variable y n Indicates that if cell n is the target cell to be closed, then y n =0, otherwise y n = 1. The closed state of all cells is vector y, y = (y1, y2, ..., y n ). By inputting the closing state vector, it is possible to distinguish which is the target cell and which is its neighboring cell.
[0076] Each graph convolutional layer of the graph convolutional network model aggregates the node features of each cell with the feature information of its neighboring nodes to capture the mutual influence between cells. The representation of nodes is learned by stacking multiple layers of graph convolutional layers. At each layer, the node features are updated by its neighboring nodes.
[0077] The output layer of the graph convolutional network model outputs a scalar O, where O j It indicates the number of users that migrate from the target cell to be shut down to the neighboring cell j with index j.
[0078] The training objective of the graph convolutional network model is to predict which neighboring cells the users of the target cell will migrate to and the number of migrations when the target cell is shut down. The graph convolutional network model uses the difference between the actual value and the predicted value of the number of migrated users as the loss function for model training.
[0079] The training process of the graph convolutional network model is as follows:
[0080] Forward propagation: Input the graph structure corresponding to the target cell and the adjacent cells, and forward propagate through the GCN layer to obtain the new feature representation of each cell. Backward propagation: Calculate the error according to the loss function, update the model parameters through the back propagation algorithm, and optimize the model performance. Evaluation and verification: During the training process, use the verification set to evaluate the model, adjust the model parameters, and prevent overfitting. Verify the prediction accuracy of the model through experiments to ensure that user migration can be accurately predicted in actual applications.
[0081] Therefore, the GCN model captures the relationship and features between cells and aggregates the information of adjacent nodes through convolution operations. It can effectively predict which cells users will migrate to and how many users will migrate when the target cell is closed.
[0082] In one embodiment, the graph neural network model can also be applied to the learning of dynamic graph structures, taking into account the changes in the network topology after cell energy saving, and realizing intelligent decision-making on user migration. Through the dynamic graph neural network model, the network topology changes are perceived in real time, and the user diversion strategy is adjusted in time to adapt to the dynamic changes in the network environment.
[0083] Step 106: predicting the network speed of each adjacent cell based on the historical data and the predicted number of users.
[0084] After predicting the number of users who will migrate to each cell if the target cell is closed through step 104, the historical data of each adjacent cell (except for the cells where the predicted number of migrated users is 0) can be further combined to predict the network speed of each cell after the new users migrate to the cell, so as to subsequently determine whether the cell will still maintain good network performance and user experience after the new users migrate.
[0085] Specifically, predicting the network speed of each adjacent cell based on the historical data and the predicted number of users includes: respectively obtaining the features of the number of users, received signal strength, network speed and the corresponding predicted number of users of the target adjacent cell in the historical data; merging the obtained features and inputting them into a convolutional neural network model to obtain the average network speed of the target adjacent cell, wherein the convolutional neural network model predicts the average network speed of the cell based on the features of the cell including the number of users, received signal strength, network speed and the corresponding predicted number of users.
[0086] In this step, the new average network speed of the corresponding adjacent cell can be predicted using historical data and the number of users that are about to migrate in. The input vector includes the node features of the adjacent cell and the number of users that are about to migrate in. A convolutional neural network (CNN) model can be used to process the input vector corresponding to each adjacent cell and predict the average network speed S′ of the corresponding adjacent cell.
[0087] Taking the adjacent cell j as an example, the input data of cell j includes node features X j and the predicted number of users to migrate in j Node feature X j The characteristics including parameters related to the number of users, received signal strength, and network speed can be obtained through step 104, including the number of users to be migrated. j It can be predicted in step 104 and is a state variable.
[0088] Merge node feature X j and the number of users O j Merge to form a new node feature I j =[X j ;O j ] and use the multi-layer convolutional neural network of the CNN model to extract the input vector I j The convolutional layer can capture the complex relationship between the characteristics of cell j and the number of users. The feature map after the convolutional layer is flattened and then further processed by several fully connected layers. The final output layer of the CNN model is the regression layer, which is used to predict the new average network speed S of cell j. ′ .
[0089] The convolutional neural network model can be trained based on the features of different cells including the number of users, received signal strength, network speed and the corresponding predicted number of users as samples, and with the average network speed of the corresponding cell as a label.
[0090] Step 108: Based on the predicted number of users and the predicted network speed, it is determined whether the adjacent cells are load balanced after the target cell is shut down.
[0091] After the number of users in each adjacent cell is predicted in step 104 and the network speed of each adjacent cell is predicted in step 106, it can be determined whether the corresponding adjacent cells to which users migrate after the target cell is closed are load balanced in combination with the predicted number of users and network speed.
[0092] Specifically, the method of judging whether the adjacent cells are load balanced after the target cell is closed based on the predicted number of users and the predicted network speed includes: determining the number of users after migration of each adjacent cell based on the sum of the number of users of each adjacent cell in the historical data and the corresponding predicted number of users; obtaining the network speed of each adjacent cell in the historical data; calculating the change value between the network speed of each adjacent cell in the historical data and the corresponding predicted network speed of each adjacent cell; respectively comparing the number of users after migration corresponding to each adjacent cell with a preset user number threshold, the predicted network speed corresponding to each adjacent cell with a preset network speed threshold, and the network speed change value corresponding to each adjacent cell with a preset network speed change threshold; if each adjacent cell satisfies the conditions that the number of users after migration is less than the preset user number threshold, the predicted network speed is greater than the preset network speed threshold, and the network speed change value is less than the preset network speed change threshold, then it is judged that the adjacent cells are load balanced after the target cell is closed; otherwise, it is judged that the load of each adjacent cell is unbalanced after the target cell is closed.
[0093] When performing cell energy saving operations, users will migrate to nearby neighboring cells that are not energy-saving. In order to ensure that these neighboring cells that are not energy-saving can still maintain good network performance and user experience after users migrate in, it is necessary to set a reasonable threshold for determining the load balancing status of the cell.
[0094] In the embodiment of the present application, the predicted number of users migrated to each adjacent cell is added to the original number of users in each adjacent cell in the historical data, so as to predict the corresponding number of users after migration in each adjacent cell. The predicted network speed of each adjacent cell is subtracted from the original network speed of each adjacent cell in the historical data, so as to predict the corresponding network speed change of each adjacent cell after the user migration.
[0095] In an embodiment of the present application, the three indicators of predicted number of users after migration, predicted network speed, and predicted change in network speed before and after migration can be combined to evaluate the load balancing of adjacent cells that do not save energy after user migration.
[0096] The following describes the steps for setting the thresholds for the above three indicators:
[0097] 1. User quantity threshold l threshold :It is an important indicator reflecting the cell load, which can be determined by the number of RRC connections in the corresponding cell. When the number of users reaches a certain number, it means that there are many connected users in the cell and the load is high, which may affect the user experience. By analyzing the historical RRC connection number data of the cell, the distribution of RRC connection numbers of the cell under different load conditions is found, and a critical value is determined as the user number threshold.
[0098] 2. Network speed threshold v threshold :Network speed is a key indicator that directly reflects user experience. When the network speed is lower than a certain value, the user experience will be significantly reduced, which may lead to reduced user satisfaction. By collecting and analyzing the historical data of cell network speed, the distribution of cell network speed under different network loads and configurations is determined, and the network speed value at which the user experience is significantly reduced is found as the network speed threshold.
[0099] 3. Network speed change threshold Δv threshold :Changes in network speed can reflect dynamic changes in network status. When network speed changes significantly, it may mean that the network load or configuration has changed significantly, and network parameters need to be adjusted in time to maintain a stable user experience.
[0100] In combination with the above thresholds, in one embodiment, the load balancing state of each cell may be determined based on the following conditions:
[0101] (l p <l threshold )&(vp >v threshold )&(Δv<Δv threshold ),
[0102] Among them, l p represents the number of users in the target adjacent cell after migration, v p represents the predicted network speed of the target neighboring cell, and Δv represents the predicted network speed change value of the target neighboring cell.
[0103] That is, after the target cell is turned off, the neighboring cells that are not energy-saving should meet the above conditions at the same time to be considered as being in a load-balanced state. Then the energy-saving strategy that currently uses the target cell as an energy-saving cell to be turned off can be issued. Otherwise, it means that the neighboring cells that are not energy-saving after the target cell is turned off are in a load-unbalanced state. The current energy-saving strategy that uses the target cell as an energy-saving cell to be turned off will cause load imbalance in other neighboring cells. In this case, it is necessary to delete the energy-saving strategy, or consider turning off other cells to formulate an energy-saving strategy to ensure the performance of the entire communication network and user perception.
[0104] In one embodiment, if the neighboring cells that are not energy-saving are in a load imbalance state after the target cell is turned off, the target cell can still be used as the energy-saving cell to be turned off, but the antenna transmission power of each neighboring cell that is not energy-saving is optimized. For example, by increasing the antenna transmission power of the neighboring cells that are not energy-saving, the corresponding neighboring cells are still in load balance after the target cell is turned off.
[0105] Optionally, after judging whether the adjacent cells are load-balanced after the target cell is closed based on the predicted number of users and the predicted network speed, the method further includes: if the loads of the adjacent cells are unbalanced, simulating the adjustment of the antenna transmission power of each adjacent cell; predicting the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmission power; updating the received signal strength in the historical data based on the predicted received signal strength; repeating the steps of predicting the number of users migrating from the target cell to each adjacent cell after the target cell is closed, to judging whether the adjacent cells are load-balanced after the target cell is closed, based on the updated historical data; if the loads of the adjacent cells are balanced, closing the target cell and adjusting the antenna transmission power of each adjacent cell to the simulated adjusted antenna transmission power.
[0106] The specific steps for achieving cell load balancing by optimizing the antenna transmission power of the adjacent cell that does not save energy are as follows:
[0107] a. Initialize antenna transmission power: First, read the existing network configuration data of each adjacent cell that is not energy-saving as the initialization antenna transmission power.
[0108] b. Predict the number of users after migration in each adjacent cell that does not save energy after the target cell is closed as an energy-saving cell. p and network speed v p .
[0109] c. Load balancing judgment: Use the preset user number threshold l threshold 、Network speed threshold v threshold , Network speed change threshold Δv threshold , determine whether shutting down the target cell will cause load balancing problems in other adjacent cells that are not energy-saving.
[0110] d. Optimize the antenna transmission power of each adjacent cell: If the above step c determines that a load imbalance state will occur, simulate and adjust the antenna transmission power of each adjacent cell that is not energy-saving, and predict the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmission power.
[0111] Specifically, the predicting of the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmit power includes: inputting the simulated adjusted antenna transmit power of the target adjacent cell into a pre-trained multilayer perceptron model to obtain the received signal strength of the target adjacent cell, and the multilayer perceptron model learns the relationship between the antenna transmit power and the received signal strength of the cell based on the antenna transmit power and the corresponding received signal strength of different cells, so as to predict the received signal strength of the cell.
[0112] In the embodiment of the present application, a multi-layer perceptron (MLP) model is constructed, which can learn the relationship between the antenna transmission power and the received signal strength of each adjacent cell based on the historical data of each adjacent cell (including the received signal strength of different users in each adjacent cell and the corresponding antenna transmission power), and is used to predict the received signal strength of each adjacent cell after the transmission power of each adjacent cell is adjusted. In this way, the antenna transmission power of the adjacent cell is better optimized, thereby improving the network performance.
[0113] After predicting the received signal strength after the antenna transmission power is adjusted, the received signal strength included in the historical data in steps 104 and 106 is updated respectively, so as to re-predict the number of users migrating from the target cell to each adjacent cell after the target cell is closed in step 104, and the network speed of the adjacent cell in step 106, and in step 108, based on the re-predicted number of users and the predicted network speed, determine whether the adjacent cells are load balanced after the target cell is closed.
[0114] If the adjacent cells are currently judged to be load balanced, the target cell is used as an energy-saving cell to be closed, and the antenna transmission power of each adjacent cell needs to be adjusted to the simulated adjusted antenna transmission power. The energy-saving strategy formulated in this way can achieve load-balanced cell energy saving under the premise of optimizing the adjacent cell antenna transmission power.
[0115] In one embodiment, after repeating the steps of predicting the number of users who will migrate from the target cell to each adjacent cell after the target cell is turned off and judging whether the adjacent cells are load balanced after the target cell is turned off based on the updated historical data, it also includes: if the loads of the adjacent cells are unbalanced, repeating the steps of simulating and adjusting the antenna transmit power of each adjacent cell and repeating the steps of predicting the number of users who will migrate from the target cell to each adjacent cell after the target cell is turned off and judging whether the adjacent cells are load balanced after the target cell is turned off, until the loads of the adjacent cells are balanced or the maximum number of iterations is reached.
[0116] That is to say, if optimizing the antenna transmission power of each adjacent cell once still cannot overcome the problem of unbalanced loads of other adjacent cells after the target cell is closed, the antenna transmission power of each adjacent cell can be optimized again. For example, the antenna transmission power of each adjacent cell can be adjusted successively according to the gradient. After each optimization adjustment, the above steps of simulating and adjusting the antenna transmission power of each adjacent cell, predicting the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmission power, updating the received signal strength in the historical data based on the predicted received signal strength, predicting the number of users who migrate from the target cell to each adjacent cell after the target cell is closed based on the updated historical data, predicting the network speed of each adjacent cell, and judging whether the adjacent cells are load balanced after the target cell is closed are repeated.
[0117] Until the loads of the adjacent cells are balanced, a corresponding energy-saving strategy is formulated based on the antenna transmission power of the adjacent cells adjusted by the final simulation and taking the target cell as an energy-saving cell.
[0118] Alternatively, the maximum number of iterations of the above repetitions is pre-set. If the set number of iterations has been reached and the antenna transmit power adjustment parameters that can solve the load imbalance problem of the adjacent cells are not found, the energy-saving strategy of the target cell as an energy-saving cell to be closed is deleted to ensure the performance of the entire communication network and user perception.
[0119] By optimizing the antenna transmission power, the network speed of the non-energy-saving cells can be made as close as possible to the overall average network speed, thereby achieving a load balancing effect.
[0120] In the above embodiment, the MLP model is used to predict the received signal strength after the antenna transmission power of the cell is adjusted, and combined with the predicted user migration situation after the target cell is closed, the load balancing state is judged, and the antenna transmission power is optimized through algorithm iteration to achieve network performance optimization and energy saving goals.
[0121] Step 110: If the loads of the adjacent cells are balanced, shut down the target cell.
[0122] If it is predicted that the load balancing of the adjacent cells can still be achieved after the users of the target cell are migrated to a suitable adjacent cell, an energy-saving strategy can be formulated to use the target cell as an energy-saving cell, and the strategy for shutting down the corresponding target cell can be issued. If the load balancing of the adjacent cells is also combined with the adjustment of the antenna transmit power parameters, the strategy for adjusting the antenna transmit power will be issued together.
[0123] Reference below Figure 2 , Figure 2 is an overall flow chart of the inter-cell load balancing method of the embodiment of the present application, such as Figure 2 As shown, the following steps are included:
[0124] Step 202, prepare historical data of each cell;
[0125] Step 204, calculating the migration adjacency matrix A according to the maximum received signal strength corresponding to the user in the historical data;
[0126] Step 206, based on the historical data, obtain the description vector representing the node feature X of each cell;
[0127] Step 208, after selecting the target cell as the energy-saving cell, predicting the number of users migrating into neighboring cells of the target cell that are not energy-saving, that is, the number of users migrating from the target cell to each neighboring cell;
[0128] Step 210, predicting the user perceived rate of the neighboring cell that is not energy-efficient, that is, the network speed;
[0129] Step 212, based on the threshold or value used by each adjacent cell to determine the load balancing state, determine whether each adjacent cell is in load balancing, if so, proceed to step 222, otherwise proceed to step 214;
[0130] Step 214, determining whether the above steps have been repeated for a preset number of iterations, if so, proceeding to step 220, otherwise proceeding to step 216;
[0131] Step 216, simulating the transmission power of the neighboring cell antenna to adjust;
[0132] Step 218, updating the RSRP value of the MR data corresponding to the received signal strength in the historical data, and returning to step 204, repeating steps 204 to 218 until a preset number of iterations is reached;
[0133] Step 220, deleting the energy-saving strategy of the target cell causing the load imbalance as an energy-saving cell;
[0134] Step 222, determine the energy-saving strategy to use the target cell as the energy-saving cell, and if there is antenna transmit power adjustment, determine the antenna transmit power parameter adjustment strategy of each adjacent cell;
[0135] Step 224: Send the corresponding strategy to execute the corresponding cell energy saving operation according to the strategy.
[0136] In an embodiment of the present application, historical data of a target cell and each adjacent cell of the target cell are obtained, wherein the historical data include the number of users in the cell, the received signal strength, and the network speed; a graph structure constructed based on the historical data and a pre-trained graph neural network model are used to predict the number of users migrating from the target cell to each adjacent cell after the target cell is closed. The graph structure uses the target cell and each adjacent cell as nodes, and the user migration ratio between the target cell and each adjacent cell as edges. The graph neural network model uses the graph structure constructed corresponding to the energy-saving cell and the corresponding non-energy-saving adjacent cell as a sample, and the number of users in the energy-saving cell to which each non-energy-saving adjacent cell is migrated after the energy-saving cell is closed as the sample. The quantity is obtained by label training, and the user migration ratio is related to the received signal strength of each adjacent cell corresponding to the user; based on the historical data and the predicted number of users, the network speed of each adjacent cell is predicted; based on the predicted number of users and the predicted network speed, it is judged whether the adjacent cells are load balanced after the target cell is closed; if the adjacent cells are load balanced, the target cell is closed, thereby the graph neural network can be used to accurately predict the number of user migrations and adjacent cells, and by analyzing the cell load balancing after user migration, the balanced distribution of network load is achieved, the load balancing of each cell is ensured, and the performance of the overall network is improved, thereby balancing the relationship between cell energy saving and load balancing well, and improving energy saving efficiency and network performance.
[0137] In addition, energy-saving and load-balancing strategies can be adjusted and optimized in real time. According to the real-time network status and user needs, energy-saving strategies and load-balancing solutions can be dynamically adjusted to ensure stable network operation and good user experience.
[0138] Alternatively, if Figure 3As shown, an embodiment of the present application also provides a small cell load balancing device 2000, including a processor 2400 and a memory 2200, and the memory 2200 stores a program or instruction that can be run on the processor 2400. When the program or instruction is executed by the processor 2400, the various steps of the above-mentioned small cell load balancing method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
[0139] The embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of any of the above-mentioned inter-cell load balancing method embodiments is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the readable storage medium includes a computer-readable storage medium, such as a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.
[0140] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. The computer program can be operated to enable a computer to implement each process of any of the above-mentioned inter-cell load balancing method embodiments when executed, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0141] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0142] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0143] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A method for load balancing between cells, characterized in that: include: Acquire historical data of a target cell and adjacent cells of the target cell, wherein the historical data includes the number of users in the cell, received signal strength, and network speed; Based on the graph structure constructed by the historical data and the pre-trained graph neural network model, the number of users migrated from the target cell to each adjacent cell after the target cell is closed is predicted, the graph structure uses the target cell and each adjacent cell as nodes, and the user migration ratio between the target cell and each adjacent cell as edges, the graph neural network model uses the graph structure constructed corresponding to the energy-saving cell and the corresponding non-energy-saving adjacent cell as samples, and the number of users in the energy-saving cell to migrate to each non-energy-saving adjacent cell after the energy-saving cell is closed as labels for training, and the user migration ratio is related to the received signal strength of each adjacent cell corresponding to the user; Based on the historical data and the predicted number of users, predict the network speed of each adjacent cell; Based on the predicted number of users and the predicted network speed, determining whether adjacent cells are load balanced after the target cell is shut down; If the loads of the adjacent cells are balanced, the target cell is shut down.
2. The method according to claim 1, characterized in that The graph structure constructed based on the historical data and the pre-trained graph neural network model predict the number of users migrating from the target cell to each adjacent cell after the target cell is closed, including: Determine, based on the maximum received signal strength of each user in the target cell corresponding to each adjacent cell in the historical data, a migration ratio of each user in the target cell to each adjacent cell after the target cell is closed; A graph structure is constructed with the target cell and each adjacent cell as nodes and the migration ratio between the target cell and each adjacent cell as edges, wherein the characteristics of each node include the number of users, received signal strength and network speed of the cell in the corresponding historical data, and the weight of each edge is the corresponding migration ratio; The graph structure is input into a pre-trained graph neural network model to obtain the number of users who migrate from the target cell to each adjacent cell after the target cell is closed.
3. The method according to claim 2, characterized in that The determining, based on the maximum received signal strength of each user in the target cell corresponding to each adjacent cell in the historical data, the migration ratio of each user in the target cell to each adjacent cell after the target cell is closed, includes: Determine the number of users with the maximum received signal strength in the target cell corresponding to the target adjacent cell based on the received signal strength in the historical data; The ratio of the number of users corresponding to the maximum received signal strength to the total number of users in the target cell is determined as the migration ratio of users migrating from the target cell to the target adjacent cell after the target cell is closed.
4. The method according to claim 1, characterized in that: The predicting the network speed of each adjacent cell based on the historical data and the predicted number of users includes: Respectively obtain characteristics of the number of users, received signal strength, network speed, and corresponding predicted number of users of the target adjacent cell in the historical data; The acquired features are combined and input into a convolutional neural network model to obtain the average network speed of the target adjacent cell. The convolutional neural network model predicts the average network speed of the cell based on the features of the cell including the number of users, received signal strength, network speed and the corresponding predicted number of users.
5. The method according to claim 1, characterized in that The determining, based on the predicted number of users and the predicted network speed, whether the adjacent cells are load balanced after the target cell is shut down includes: Determining the number of users after migration in each adjacent cell based on the sum of the number of users in each adjacent cell in the historical data and the corresponding predicted number of users; Obtaining the network speed of each adjacent cell in the historical data; Calculate the change between the network speed of each adjacent cell in the historical data and the corresponding predicted network speed of each adjacent cell; Respectively comparing the number of users after migration corresponding to each adjacent cell with a preset user number threshold, the predicted network speed corresponding to each adjacent cell with a preset network speed threshold, and the network speed change value corresponding to each adjacent cell with the preset network speed change threshold; If each adjacent cell satisfies the following conditions: the number of users after migration is less than a preset user number threshold, the predicted network speed is greater than a preset network speed threshold, and the network speed change value is less than a preset network speed change threshold, then it is determined that the load of each adjacent cell is balanced after the target cell is closed; Otherwise, it is determined that the loads of the adjacent cells are unbalanced after the target cell is shut down.
6. The method according to any one of claims 1 to 5, characterized in that After judging whether the adjacent cells are load balanced after the target cell is shut down based on the predicted number of users and the predicted network speed, the method further includes: If the loads of adjacent cells are unbalanced, adjust the antenna transmission power of each adjacent cell in a simulated manner; Predicting the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmit power; Based on the predicted received signal strength, updating the received signal strength in the historical data; Based on the updated historical data, repeating the steps of predicting the number of users migrating from the target cell to each adjacent cell after the target cell is shut down, to determining whether each adjacent cell is load balanced after the target cell is shut down; If the loads of the adjacent cells are balanced, the target cell is shut down and the antenna transmission power of the adjacent cells is adjusted to the simulated adjusted antenna transmission power.
7. The method according to claim 6, characterized in that The predicting of the received signal strength of each adjacent cell corresponding to each simulated adjusted antenna transmit power includes: The simulated adjusted antenna transmit power of the target adjacent cell is input into a pre-trained multilayer perceptron model to obtain the received signal strength of the target adjacent cell. The multilayer perceptron model learns the relationship between the antenna transmit power and the received signal strength of the cell based on the antenna transmit power of different cells and the corresponding received signal strength, so as to predict the received signal strength of the cell.
8. A load balancing device between cells, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.
9. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer program product, the computer program product comprising a non-transitory computer-readable storage medium storing a computer program, the computer program being operable to cause a computer to execute the steps of the method according to any one of claims 1 to 7.