Active mobility load balancing method and system based on interior point strategy optimization

Abstract

This invention relates to the field of communications, and more specifically, to a proactive mobility load balancing method and system based on in-point policy optimization. The method includes: constructing a signal-to-interference-plus-noise ratio (SINR) model; predicting trajectories using RS-BART to obtain two sets of UE-BS associations: reliable UE-BS associations and unreliable UE-BS associations; constructing a joint optimization problem of MLB and MRO with the goal of maximizing the Jain exponent; solving the joint optimization problem of MLB and MRO using an in-point policy to optimize the unreliable UE-BS associations, resulting in optimized UE-BS associations; determining all feasible UE-BS associations based on the optimized UE-BS associations and reliable UE-BS associations; calculating cell-specific offsets based on all feasible UE-BS associations and the allowable range of cell-specific offsets; and achieving user reassignment by adjusting the cell-specific offsets to complete load balancing. This method first uses RS-BART to predict trajectories and determine all feasible UE-BS associations, which reduces the selection range and improves balancing efficiency when performing mobility load balancing.

Description

Technical Field

[0001] The present invention relates to the field of communications, and more specifically, to an active mobility load balancing method and system based on interior point strategy optimization. Background Art

[0002] In ultra-dense networks (UDN), the distance between base stations is about 100 meters or even less. Due to the high density of wireless networks and the mobility of user devices, the handover rate increases and the cell load is prone to imbalance. This unbalanced network can lead to insufficient wireless resources, high handover failure rate, and even radio link failure. Therefore, it is a huge challenge for future B5G or 6G networks.

[0003] The prior art discloses an ultra-dense network switching method based on user mobile behavior prediction, including the following steps: A: obtaining pre-processed trajectory data of all base station user mobile behaviors in a set area; B: dividing into a base station new user data set and a base station old user data set; C: constructing a base station old user trajectory prediction model; D: obtaining a trained base station old user trajectory prediction model; E: obtaining the predicted mobile trajectory of the new base station user; F: when the base station user meets the A2 event triggering switching, find the target base station that meets the judgment condition as the final switching base station and switch. The present invention can predict future trajectory points through the user's historical trajectory sequence, which can effectively reduce the number of network switching times, optimize network switching performance and improve the robustness of the network. However, this method only reduces the number of network switching times through location prediction, and cannot solve the problem that the cell load is prone to imbalance, resulting in insufficient wireless resources and high switching failure rate. Summary of the invention

[0004] The purpose of the present invention is to disclose an active mobility load balancing method and system based on interior point strategy optimization with better performance.

[0005] In order to achieve the above object, the present invention provides an active mobility load balancing method based on interior point strategy optimization:

[0006] S1: construct a signal-to-interference-noise ratio model, and use RS-BART to predict trajectories to obtain two groups of UE-BS associations, namely, reliable UE-BS associations and unreliable UE-BS associations;

[0007] S2: construct a joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective, solve the joint optimization problem of MLB and MRO through the interior point strategy, optimize the unreliable UE-BS association, and obtain the optimized UE-BS association;

[0008] S3: Determine all feasible UE-BS associations according to the optimized UE-BS associations and the reliable UE-BS associations;

[0009] S4: Calculate the cell personality offset according to all feasible UE-BS associations and the allowed range of the cell personality offset;

[0010] S5: By adjusting the cell personality offset, the user is re-associated and load balancing is completed.

[0011] Further, in step S1, the signal to interference noise ratio model includes: a plurality of base stations and a plurality of users, the set of base stations is Where M is the total number of base stations; the user set is Where N is the total number of users; each base station can provide services to at least one user, U i represents the set of users served by the ith base station, v j is the minimum data rate requirement for the jth user, Let w ii,j is the spectrum efficiency SE of the link between the i-th base station and the j-th user. According to the heterogeneous wireless network, the load factor of the i-th base station is:

[0012]

[0013] Among them, K i represents the number of resource blocks allocated to all user equipments served by the i-th base station, and K is the maximum number of RBs per base station, where K ii Less than or equal to K, α i,j represents the association between the jth user and the i-th base station. When the i-th base station provides service for the jth user, α i,j =1, otherwise α i,j =0, spectral efficiency w i, j is:

[0014]

[0015] in, is the signal-to-interference-to-noise ratio, B and pi are the bandwidth and transmission power of the ith base station, respectively, and g i,j is the channel gain between the ith base station and the jth user, is the noise power.

[0016] Furthermore, in step S1, it includes:

[0017] S1.1: RS-BART model is Y=f(X1,…,X p )+ε,

[0018] where ε represents the error term, x=(x1,…,x p ) is feature information;

[0019] S1.2: Approximate by using the sum of Q regression trees, that is The specific definition is:

[0020]

[0021] Among them, T j represents the jth binary tree, and is a length b j A vector representing the tree T j Terminal node parameters, functions The role of is to distribute the input features where l∈1,…,b j to X;

[0022] S1.3: In RS-BART, we first sort the variables X, and then select the first r variables from a total of p variables, where r is much smaller than p, by defining This is an n×p matrix.

[0023] S1.4: Estimation of dimensionality reduction weights β p , based on the dimension reduction weight β p ,from Select r columns from the X i , X i is used as the input of the BART model, i.e.

[0024] S1.5: After the above step S1.4 is repeated B times, the output is expressed as: The next position y of the user is the position of the user in time slot t+1, ..., t+n, and the input X represents a set of features, including historical positions and p other features. The predicted trajectory is obtained through the output and then two sets of UE-BS associations are obtained, namely reliable UE-BS associations and unreliable UE-BS associations.

[0025] Furthermore, in step S2, constructing a joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective includes:

[0026] The Jain Index formula is as follows:

[0027] Where δ(η) is a value between 0 and 1, and there are three cases:

[0028] (1) If the loads of all base stations are the same, Then δ(η)=1,

[0029] (2) If there is one base station serving all users, then

[0030] (3) If there are k base stations, the load is The load of Mk base stations is 0, then

[0031] Based on the above description, the larger δ(η) is, the higher the fairness is. Therefore, in order to maximize the Jain index of the entire network, the joint optimization problem of MLB and MRO is constructed as follows:

[0032]

[0033] st

[0034]

[0035] The constructed joint optimization problem of MLB and MRO is equivalent to

[0036]

[0037] in is the sample coefficient of variation, μ η and σ η are the mean and variance of the load η, respectively.

[0038] Furthermore, in step S2, the joint optimization problem of MLB and MRO is solved by an interior point strategy to optimize the unreliable UE-BS association, including:

[0039] S2.1: Define the state vector as:

[0040] S2.2: Define the action vector as:

[0041] in, represents the set of neighboring base stations of the i-th base station, and the undetermined α i,j Those users are located in the overlapping area of ​​two adjacent base stations, and their movement direction prediction probability is not high;

[0042] S2.3: Set the long-term reward to

[0043] R(s t ,a t ,s t+1 )=δ(η).

[0044] S2.4: Select strategy π θ , maximize the cumulative discount reward set up For strategy π θThe expected discounted return, τ = (s0, a0, s1, a1…) is a trajectory, then the objective function is equivalent to:

[0045]

[0046] S2.5: By introducing a logarithmic barrier function in Then the objective function is restructured as:

[0047]

[0048] Among them, L CIPO (θ) is defined as:

[0049]

[0050] in, A t , are advantage function and clipping function respectively, r t The value interval of (θ) is [1-ε,1+ε]. Optimize unreliable UE-BS association.

[0051] Further, in step S2.4, the output strategy parameter θ is obtained by the following steps:

[0052] S2.4.1: Use parameter θ = θ0;

[0053] S2.4.2: Given θ k , sample N trajectories τ1,…,τ N , k = 0;

[0054] S2.4.3: Process the trajectory τ to calculate the learning rate;

[0055] S2.4.4: Update the policy parameter θ based on the following formula k : Where κ is the learning rate based on the trajectory in step S2.4.3;

[0056] S2.4.5: Determine whether k satisfies the condition. If so, output the strategy parameter θ=θ k+11 , otherwise k=k+1 and return to execute step S2.4.4.

[0057] Further, in step S4, calculating the cell individual offset includes: preliminarily calculating the base station individual offset formula as follows: Among them, P k and P i They represent the received signal strength of the user from the kth base station and the ith base station, respectively, and H iis the hysteresis value associated with the ith base station, is the cell individual offset between the kth base station and the ith base station, and P k After being replaced with transmit power and channel gain, the individual offset of the base station should satisfy:

[0058] Furthermore, the hysteresis value ensures that the signal strength from the adjacent k-th base station is sufficiently stronger than the signal strength from the ith base station; and the base station individual offset is used to balance the load between adjacent base stations.

[0059] Further, in step S4, the allowed range of the cell personality offset includes:

[0060] First, the SINR model is preliminarily balanced based on the individual base station offset. That is, a lightly loaded adjacent k-th base station of the i-th base station is selected to reduce the load of the i-th base station. Then the i-th base station will increase The value of and at the same time notify the kth base station to reduce value, adjust value to ensure that it is greater than the minimum and less than the maximum value These minimum and maximum values ​​should satisfy the following inequalities:

[0061]

[0062] get:

[0063] In addition, the individual offset of the base station should satisfy:

[0064]

[0065] So we get:

[0066]

[0067]

[0068] To prevent ping-pong switching, the following conditions should be met:

[0069]

[0070] Then we get:

[0071]

[0072] Finally we get:

[0073] in and They are and The adjustment value of Less than or equal to upper bound and Greater than or equal to lower bound are the minimum and maximum values ​​defined.

[0074] In addition, the present invention also provides an active mobility load balancing system based on interior point strategy optimization, comprising:

[0075] Prediction module: constructs a signal-to-interference-noise ratio model and uses RS-BART to predict trajectories to obtain two groups of UE-BS associations, namely reliable UE-BS associations and unreliable UE-BS associations;

[0076] Optimization module: constructs the joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective, solves the joint optimization problem of MLB and MRO through the interior point strategy, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association;

[0077] Determination module: Determine all feasible UE-BS associations according to the optimized UE-BS associations and reliable UE-BS associations;

[0078] Calculation module: Calculate the cell personality offset based on all feasible UE-BS associations and the allowed range of the cell personality offset;

[0079] Load balancing module: By adjusting the cell personality offset, user reassociation is achieved and load balancing is completed.

[0080] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0081] This method first uses RS-BART to predict the trajectory, and the UE-BS association divides a large number of UE-BS associations into reliable UE-BS associations and unreliable UE-BS associations, where UE is the user equipment and BS is the base station; when performing mobility load balancing, the selection range can be reduced to improve the balancing efficiency. This method solves the joint optimization problem of MLB and MRO through the interior point strategy, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association; then determines all feasible UE-BS associations; so that this embodiment avoids adverse consequences such as ping-pong switching or even wireless link failure without much prior and external knowledge; secondly, the interior point strategy is a first-order constraint optimization method, so its computational complexity is very low and has a wider range of applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a flow chart of the active mobility load balancing method based on interior point strategy optimization described in Example 1;

[0083] Figure 2 A block diagram of an active mobility load balancing system based on interior point strategy optimization described in Example 2;

[0084] Figure 3 The wireless link failure rate line graph described in the fourth embodiment;

[0085] Figure 4 The load variance comparison bar chart described in Example 4; DETAILED DESCRIPTION

[0086] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0087] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0088] Embodiment 1:

[0089] This embodiment provides Figure 1 An active mobility load balancing method based on interior point strategy optimization is shown, comprising the following steps:

[0090] S1: construct a signal-to-interference-noise ratio model, and use RS-BART to predict trajectories to obtain two groups of UE-BS associations, namely, reliable UE-BS associations and unreliable UE-BS associations;

[0091] S2: construct a joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective, solve the joint optimization problem of MLB and MRO through the interior point strategy, optimize the unreliable UE-BS association, and obtain the optimized UE-BS association;

[0092] S3: Determine all feasible UE-BS associations according to the optimized UE-BS associations and the reliable UE-BS associations;

[0093] S4: Calculate the cell personality offset according to all feasible UE-BS associations and the allowed range of the cell personality offset;

[0094] S5: By adjusting the cell personality offset, the user is re-associated and load balancing is completed.

[0095] This embodiment first uses RS-BART to predict the trajectory, and the UE-BS association divides a large number of UE-BS associations into reliable UE-BS associations and unreliable UE-BS associations, where the UE in the UE-BS association is the user equipment and the BS is the base station; when performing mobility load balancing, the selection range can be reduced to improve the balancing efficiency. This embodiment uses the interior point strategy to solve the joint optimization problem of MLB and MRO, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association; then determines all feasible UE-BS associations; so that this embodiment avoids adverse consequences such as ping-pong switching or even wireless link failure without much prior and external knowledge; secondly, the interior point strategy is a first-order constraint optimization method, so its computational complexity is very low and has a wider range of applications.

[0096] Embodiment 2:

[0097] This embodiment further discloses on the basis of the first embodiment:

[0098] Further, in step S1, the signal to interference noise ratio model includes: a plurality of base stations and a plurality of users, the set of base stations is Where M is the total number of base stations; the user set is Where N is the total number of users; each base station can provide services to at least one user, U i represents the set of users served by the ith base station, v j is the minimum data rate requirement for the jth user, Let w ii,j is the spectrum efficiency SE of the link between the i-th base station and the j-th user. According to the heterogeneous wireless network, the load factor of the i-th base station is:

[0099]

[0100] Among them, K i represents the number of resource blocks allocated to all user equipments served by the i-th base station, and K is the maximum number of RBs per base station, where K ii Less than or equal to K, α i,j represents the association between the jth user and the i-th base station. When the i-th base station provides service for the jth user, α i,j =1, otherwise α i,j =0, spectral efficiency w i,j for:

[0101]

[0102] in, is the signal-to-interference-noise ratio, B and p i are the bandwidth and transmission power of the i-th base station, gi,j is the channel gain between the ith base station and the jth user, is the noise power.

[0103] Furthermore, in step S1, it includes:

[0104] S1.1: RS-BART model is Y=f(X1,…,X p )+ε,

[0105] where ε represents the error term, x=(x1,…,x p ) is feature information;

[0106] S1.2: Approximate by using the sum of Q regression trees, that is The specific definition is:

[0107]

[0108] Among them, T j represents the jth binary tree, and is a length b j A vector representing the tree T j Terminal node parameters, functions The role of is to distribute the input features where l∈1,…,b j to X;

[0109] S1.3: In RS-BART, we first sort the variables X, and then select the first r variables from a total of p variables, where r is much smaller than p, by defining This is an n×p matrix.

[0110] S1.4: Estimation of dimensionality reduction weights β p , based on the dimension reduction weight β p ,from Select r columns from the X i , X i is used as the input of the BART model, i.e.

[0111] S1.5: After the above step S1.4 is repeated B times, the output is expressed as: The next position y of the user is the position of the user in time slot t+1, ..., t+n, and the input X represents a set of features, including historical positions and p other features. The predicted trajectory is obtained through the output and then two sets of UE-BS associations are obtained, namely reliable UE-BS associations and unreliable UE-BS associations.

[0112] Furthermore, in step S2, constructing a joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective includes:

[0113] The Jain Index formula is as follows:

[0114] Where δ(η) is a value between 0 and 1, and there are three cases:

[0115] (1) If the loads of all base stations are the same, Then δ(η)=1,

[0116] (2) If there is one base station serving all users, then

[0117] (3) If there are k base stations, the load is The load of Mk base stations is 0, then

[0118] Based on the above description, the larger δ(η) is, the higher the fairness is. Therefore, in order to maximize the Jain index of the entire network, the joint optimization problem of MLB and MRO is constructed as follows:

[0119]

[0120] st

[0121]

[0122] The constructed joint optimization problem of MLB and MRO is equivalent to

[0123]

[0124] in is the sample coefficient of variation, μ η and σ η are the mean and variance of the load η, respectively.

[0125] Furthermore, in step S2, the joint optimization problem of MLB and MRO is solved by an interior point strategy to optimize the unreliable UE-BS association, including:

[0126] S2.1: Define the state vector as:

[0127] S2.2: Define the action vector as:

[0128] in, represents the set of neighboring base stations of the i-th base station, and the undetermined α i,jThose users are located in the overlapping area of ​​two adjacent base stations, and their movement direction prediction probability is not high;

[0129] S2.3: Set the long-term reward to

[0130] R(s t ,a t ,s t+1 )=δ(η).

[0131] S2.4: Select strategy π θ , maximize the cumulative discount reward set up For strategy π θ The expected discounted return, τ = (s0, a0, s1, a1…) is a trajectory, then the objective function is equivalent to:

[0132]

[0133] S2.5: By introducing a logarithmic barrier function in Then the objective function is restructured as:

[0134]

[0135] Among them, L CIPO (θ) is defined as:

[0136]

[0137] in, A t , are advantage function and clipping function respectively, r t The value interval of (θ) is [1-ε,1+ε]. Optimize unreliable UE-BS association.

[0138] Further, in step S2.4, the output strategy parameter θ is obtained by the following steps:

[0139] S2.4.1: Use parameter θ = θ0;

[0140] S2.4.2: Given θ k , sample N trajectories τ1,…,τ N , k = 0;

[0141] S2.4.3: Process the trajectory τ to calculate the learning rate;

[0142] S2.4.4: Update the policy parameter θ based on the following formula k : Where κ is the learning rate based on the trajectory in step S2.4.3;

[0143] S2.4.5: Determine whether k satisfies the condition. If so, output the strategy parameter θ=θ k+11 , otherwise k=k+1 and return to execute step S2.4.4.

[0144] Further, in step S4, calculating the cell individual offset includes: preliminarily calculating the base station individual offset formula as follows: Among them, P k and P i They represent the received signal strength of the user from the kth base station and the ith base station, respectively, and H i is the hysteresis value associated with the ith base station, is the cell individual offset between the kth base station and the ith base station, and P k After being replaced with transmit power and channel gain, the individual offset of the base station should satisfy:

[0145] Furthermore, the hysteresis value ensures that the signal strength from the adjacent k-th base station is sufficiently stronger than the signal strength from the ith base station; and the base station individual offset is used to balance the load between adjacent base stations.

[0146] Further, in step S4, the allowed range of the cell personality offset includes:

[0147] First, the SINR model is preliminarily balanced based on the individual base station offset. That is, a lightly loaded adjacent k-th base station of the i-th base station is selected to reduce the load of the i-th base station. Then the i-th base station will increase The value of and at the same time notify the kth base station to reduce value, adjust value to ensure that it is greater than the minimum and less than the maximum value These minimum and maximum values ​​should satisfy the following inequalities:

[0148]

[0149] get:

[0150] In addition, the individual offset of the base station should satisfy:

[0151]

[0152] So we get:

[0153]

[0154]

[0155] To prevent ping-pong switching, the following conditions should be met:

[0156]

[0157] Then we get:

[0158]

[0159] Finally we get:

[0160] in and They are and The adjustment value of Less than or equal to upper bound and Greater than or equal to lower bound are the minimum and maximum values ​​defined.

[0161] This embodiment first uses RS-BART to predict the trajectory, and the UE-BS association divides a large number of UE-BS associations into reliable UE-BS associations and unreliable UE-BS associations, where the UE in the UE-BS association is the user equipment and the BS is the base station; when performing mobility load balancing, the selection range can be reduced to improve the balancing efficiency. This embodiment uses the interior point strategy to solve the joint optimization problem of MLB and MRO, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association; then determines all feasible UE-BS associations; so that this embodiment avoids adverse consequences such as ping-pong switching or even wireless link failure without much prior and external knowledge; secondly, the interior point strategy is a first-order constraint optimization method, so its computational complexity is very low and has a wider range of applications.

[0162] Embodiment three:

[0163] This embodiment also provides Figure 2 An active mobility load balancing system based on interior point strategy optimization is shown, comprising:

[0164] Prediction module: constructs a signal-to-interference-noise ratio model and uses RS-BART to predict trajectories to obtain two groups of UE-BS associations, namely reliable UE-BS associations and unreliable UE-BS associations;

[0165] Optimization module: constructs the joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective, solves the joint optimization problem of MLB and MRO through the interior point strategy, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association;

[0166] Determination module: Determine all feasible UE-BS associations according to the optimized UE-BS associations and reliable UE-BS associations;

[0167] Calculation module: Calculate the cell personality offset based on all feasible UE-BS associations and the allowed range of the cell personality offset;

[0168] Load balancing module: By adjusting the cell personality offset, user reassociation is achieved and load balancing is completed.

[0169] This embodiment first uses RS-BART to predict the trajectory, and the UE-BS association divides a large number of UE-BS associations into reliable UE-BS associations and unreliable UE-BS associations, where the UE in the UE-BS association is the user equipment and the BS is the base station; when performing mobility load balancing, the selection range can be reduced to improve the balancing efficiency. This embodiment uses the interior point strategy to solve the joint optimization problem of MLB and MRO, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association; then determines all feasible UE-BS associations; so that this embodiment avoids adverse consequences such as ping-pong switching or even wireless link failure without much prior and external knowledge; secondly, the interior point strategy is a first-order constraint optimization method, so its computational complexity is very low and has a wider range of applications.

[0170] Embodiment 4:

[0171] This embodiment simulates a network consisting of 20 macro cells and 60 micro cells. The locations of the base stations are extracted from a tested public dataset. The coverage radius of the macro cell is 2 km. In addition, the total coverage area of ​​all base stations is larger than the coverage area of ​​most geographical trajectories in the tested public dataset. The carrier center frequency and total bandwidth are 2.6 GHz and 100 MHz, respectively. The transmit power of the micro cell / macro cell is 26 / 43 dBm. The path loss between each user equipment and the base station it serves is calculated by 128.1+37.6log(max{d, 0.035}), where d is the distance in kilometers. The switching lag is 3 dB. In addition, in order to verify the performance of the proposed RS-BART, the public trajectory dataset in was used in the O-RAN based simulator.

[0172] For the dataset, 70% of the data is used for training, 10% for validation, and the remaining 20% ​​for testing. In the simulator, widely used preprocessing procedures are used to extract key information from the trajectory data, such as noise filtering and trajectory data compression.

[0173] This embodiment takes into account the prediction accuracy of the next serving cell and two indicators, namely, precision and recall, which are calculated as follows:

[0174]

[0175]

[0176] By definition, accuracy is a measure of the ratio of the number of correctly predicted next serving cells to the total number of next serving cell predictions. This indicator indicates whether the prediction of the next serving cell is acceptable. The higher the accuracy, the fewer unnecessary handovers, thereby improving the handover success rate.

[0177] Precision is appropriate when the cost of incorrectly predicting a negative example as a positive example is high. For example, in a lightly loaded cell, a user who is not heading to a heavily loaded neighboring cell should not be predicted to be moving to the neighboring cell and handing over. When this happens, it results in unnecessary handovers. For these cells, precision is an important metric.

[0178] Recall is a measure of how often the model correctly identifies a positive example among all actual positive examples. Recall is appropriate when it is costly to incorrectly predict a positive example as a negative example. For example, in a heavily loaded cell, a user heading to a lightly loaded neighboring cell should not be predicted as being served by the current cell. When this happens, it causes an imbalance in the network. For these cells, recall is an important metric.

[0179] Therefore, precision and recall measure the overall prediction ability of this embodiment in different situations. This embodiment is compared with previous methods. As shown in Table 1, when evaluated on public datasets, this embodiment consistently outperforms previous methods. This is because the spatial and temporal features of user locations are highly correlated, and users' behaviors have certain periodic patterns, such as people going to work every Monday. However, previous methods, Markov methods and RNNs, did not capture these features. This embodiment can process spatiotemporal features and extract periodic features. In addition, the intervals between user locations also help determine UE-BS associations, which also improves the robustness of MLB. Therefore, this embodiment is more suitable for the problem under consideration.

[0180] Table 1 Comparison between different prediction methods

[0181]

[0182] Next, the RLF (radio link failure) ratio and load deviation of this embodiment and other methods are compared in a typical scenario. Figure 3 As shown in the figure, the performance of this embodiment is obviously better than that of the prior art 1 and the prior art 2. Specifically, compared with the prior art 1 and the prior art 2, the RLF ratio of this embodiment is reduced by about 4 / 5 and 1 / 2. When the speed is lower than 60km / h, the RLF ratio is very low, lower than 1%. In addition, compared with the prior art 1 and the prior art 2, the growth trend of this embodiment is relatively gentle.

[0183] exist Figure 4 The same trend can be observed in . This also verifies the superiority of this embodiment. Generally speaking, lower load variation means that in this embodiment, the load is redistributed more evenly among the base stations.

[0184] The embodiment of the present invention provides an active mobility load balancing method and system based on interior point strategy optimization. The method includes: constructing a signal to noise ratio model, predicting trajectories through RS-BART to obtain two groups of UE-BS associations, which are reliable UE-BS associations and unreliable UE-BS associations; constructing a joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization target, solving the joint optimization problem of MLB and MRO through the interior point strategy, optimizing the unreliable UE-BS association, and obtaining the optimized UE-BS association; determining all feasible UE-BS associations according to the optimized UE-BS association and the reliable UE-BS association; calculating the cell personality offset according to all feasible UE-BS associations and the allowable range of the cell personality offset; and realizing user re-association by adjusting the cell personality offset to complete load balancing. The method first uses RS-BART to predict trajectories, and the UE-BS association divides a large number of UE-BS associations into reliable UE-BS associations and unreliable UE-BS associations, wherein the UE in the UE-BS association is the user equipment and the BS is the base station; when performing mobility load balancing, the selection range can be reduced to improve the balancing efficiency. This method solves the joint optimization problem of MLB and MRO through the interior point strategy, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association; then determines all feasible UE-BS associations; so that this embodiment avoids adverse consequences such as ping-pong switching and even wireless link failure without much prior and external knowledge; secondly, the interior point strategy is a first-order constrained optimization method, so its computational complexity is very low and its application range is wider.

[0185] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An active mobility load balancing method based on interior point strategy optimization, characterized in that: The following steps are involved: S1: construct a signal-to-interference-noise ratio model, and use RS-BART to predict trajectories to obtain two groups of UE-BS associations, namely, reliable UE-BS associations and unreliable UE-BS associations; S1.1: RS-BART model is in represents the error term, is feature information; S1.2: By using It is approximated by the sum of regression trees, that is, , specifically defined as: in, Indicates A binary tree, and is a length of A vector representing a tree Terminal node parameters, functions The role of is to distribute the input features ,in arrive ; S1.3: In RS-BART, firstly Sort by, and then from the total Select the first variable , among which Much smaller than , by definition , this is a The matrix of S1.4: Estimating dimensionality reduction weights , based on the dimensionality reduction weight ,from Select columns, which is represented as , is used as the input of the BART model, i.e. , S1.5: After the above step S1.4 is repeated B times, the output is expressed as: , the user's next location Is the user in the time slot , and input Represents a set of features, including historical locations and Other features are output to obtain the predicted trajectory and then two groups of UE-BS associations are obtained, namely reliable UE-BS associations and unreliable UE-BS associations; S2: constructing a joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective, solving the joint optimization problem of MLB and MRO through the interior point strategy, optimizing the unreliable UE-BS association, and obtaining the optimized UE-BS association; The joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective includes: The Jain Index formula is as follows: in It is a value between 0 and 1, with the following three cases: (1) If the loads of all base stations are the same, ,but , (2) If there is one base station serving all users, then , (3) If there is The load of a base station is ,and The load of each base station is 0, then , Based on the above description, The larger the value, the higher the fairness. Therefore, in order to maximize the Jain index of the entire network, the joint optimization problem of MLB and MRO is constructed as follows: The joint optimization problem of MLB and MRO is equivalent to in is the sample coefficient of variation, and The load The mean and variance of The interior point strategy is used to solve the joint optimization problem of MLB and MRO and optimize the unreliable UE-BS association, including: S2.1: Define the state vector as: S2.2: Define the action vector as: in, Indicates The set of neighboring base stations of base stations is undetermined The users are located in the overlapping area of ​​two adjacent base stations, and the predicted probability of their moving directions is not high; S2.3: Set the long-term reward to S2.4: Selection strategy , maximize the cumulative discount reward ,set up For strategy The expected discounted return, is a trajectory, then the objective function is equivalent to: S2.5: By introducing a logarithmic barrier function ,in , and then restructure the objective function as: in, Defined as: in, , , are advantage function and clipping function respectively, The value range of ,according to Optimize unreliable UE-BS association; S3: Determine all feasible UE-BS associations according to the optimized UE-BS associations and the reliable UE-BS associations; S4: Calculate the cell personality offset according to all feasible UE-BS associations and the allowed range of the cell personality offset; Calculating the cell individual offset includes: Preliminary calculation of the base station individual offset formula is as follows: in, and Respectively represent the user from Base stations and The received signal strength of each base station, It is with The hysteresis value associated with each base station, It is Base stations and The individual cell offsets between base stations will After being replaced with transmit power and channel gain, the individual offset of the base station should satisfy: ; The allowed range of zone personality offsets includes: First, the SINR model is preliminarily used for mobility load balancing according to the individual offset of the base station. A lightly loaded neighboring base station base stations to reduce the Base station load, then will increase , and notify the Base stations reduced value, adjust value to ensure that it is greater than the minimum And less than the maximum value , these minimum and maximum values ​​should satisfy the following inequalities: get: In addition, the individual offset of the base station should satisfy: , ， So we get: To prevent ping-pong switching, the following conditions should be met: Then we get: Finally we get: in and They are and The adjustment value of Less than or equal to upper bound and Greater than or equal to lower bound , , is at the lowest and highest values ​​defined; S5: By adjusting the cell personality offset, the user is re-associated and load balancing is completed.

2. The active mobility load balancing method based on interior point strategy optimization according to claim 1 is characterized in that: In step S1, the signal to interference noise ratio model includes: multiple base stations and multiple users, and the set of base stations is ,in is the total number of base stations; the user set is ;in is the total number of users; each base station can provide service to at least one user, Indicated by The set of users served by a base station, For the Minimum data rate requirement for each user, ,make For the Base stations and The spectral efficiency SE of the link between users is calculated based on the heterogeneous wireless network. The load factor of a base station is: ， in, Indicates the assignment to The number of resource blocks of all user equipments served by the base station, is the maximum number of RBs per base station, where Less than or equal to , Representative User and The association between base stations The base station is When providing services to a user, ,otherwise , spectral efficiency for: in, is the signal-to-interference-noise ratio, and They are The bandwidth and transmit power of each base station, It is Base stations and The channel gain between users is is the noise power.

3. The active mobility load balancing method based on interior point strategy optimization according to claim 1 is characterized in that: In step S2.4, the strategy parameters are output Obtained by the following steps: S2.4.1: Use parameters ; S2.4.2: Given ,sampling Tracks , k=0; S2.4.3: Processing Trajectory To calculate the learning rate; S2.4.4: Update the policy parameters based on the following formula : ,in, is the learning rate based on the trajectory in step S2.4.3; S2.4.5: Determine whether k meets the conditions, and if so, output the strategy parameters , otherwise k=k+1 and return to execute step S2.4.

4.

4. The active mobility load balancing method based on interior point strategy optimization according to claim 1 is characterized in that: The hysteresis value ensures that the The signal strength from the base station is strong enough to be stronger than that from the The base station individual offset is used to balance the load between adjacent base stations.

5. An active mobility load balancing system based on interior point strategy optimization of the active mobility load balancing method based on interior point strategy optimization according to claim 1, characterized in that: include: Prediction module: constructs a signal-to-interference-noise ratio model and uses RS-BART to predict trajectories to obtain two groups of UE-BS associations, namely reliable UE-BS associations and unreliable UE-BS associations; Optimization module: constructs the joint optimization problem of MLB and MRO with maximizing the Jain index as the optimization objective, solves the joint optimization problem of MLB and MRO through the interior point strategy, optimizes the unreliable UE-BS association, and obtains the optimized UE-BS association; Determination module: Determine all feasible UE-BS associations according to the optimized UE-BS associations and reliable UE-BS associations; Calculation module: Calculate the cell personality offset based on all feasible UE-BS associations and the allowed range of the cell personality offset; Load balancing module: By adjusting the cell personality offset, user reassociation is achieved and load balancing is completed.

Images