Joint user-base station association and resource optimization configuration method

Abstract

A joint user-base station association and resource optimization configuration method comprises the following steps: establishing a wireless network micro base station system scene, a system energy consumption model, a renewable energy output model and a two-way transaction model of a micro base station and a smart power grid, predicting a user mobility trajectory based on a neural network for the established model, and optimizing; an optimal user-base station association decision is obtained through a multi-attribute decision, the problem of long-term average operation cost minimization is solved through a random dual gradient method, resource scheduling configuration is performed through a distributed online power control algorithm, and the service quality is ensured while the long-term average cost is minimized. According to the method, an instant decision can be dynamically made in advance on the premise of minimizing the total cost of the system, and statistical data of random processes such as any channel, renewable energy sources and electricity price changes do not need to be known priori.

Description

Technical Field

[0001] The present invention relates to a technology in the field of wireless communication, in particular to a method for joint user-base station association and resource optimization configuration. Background Art

[0002] 5G networks will tend to be mainly in the form of small base stations. We optimize the wireless micro base station system in conjunction with user-base station association and resource allocation to meet user service needs while effectively reducing system costs. This will become a necessary idea for achieving sustainable development and a new field of basic research. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention proposes a joint user-base station association and resource optimization configuration method, which can dynamically make instant decisions in advance under the premise of minimizing the total system cost without the need to a priori know the statistical data of any random processes such as channels, renewable energy and electricity price changes.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to a joint user-base station association and resource optimization configuration method. By establishing a wireless network micro base station system scenario, a system energy consumption model, a renewable energy output model, and a two-way transaction model between the micro base station and the smart grid, the constructed model is optimized by predicting the user mobility trajectory based on a neural network, and then the optimal user-base station association decision is obtained through multi-attribute decision-making. The long-term average operating cost minimization problem is solved by a random dual subgradient method, and resource scheduling and configuration are performed through a distributed online power control algorithm to achieve the minimum long-term average cost while ensuring service quality.

[0006] The wireless network micro base station system scenario includes: a multi-micro base station wireless network system powered by a smart grid consisting of I = {1, 2, ..., I} micro base stations and K = {1, 2, ..., K} mobile users, wherein each micro base station and user is equipped with M and 1 antenna respectively.

[0007] The micro base station is provided with an energy collection device and a rechargeable battery. The micro base station can collect renewable energy (such as wind energy, solar energy, etc.) from the environment through the energy collection device to reduce dependence on the smart grid. Compared with the existing energy consumption, the acquisition of this renewable energy, assuming it is free, brings a more economical and efficient option to the system. In addition, the micro base station can also conduct two-way energy transactions with the smart grid based on dynamic electricity prices to further achieve the goal of reducing system overhead. Assuming that the duration of each time slot is a unit time, the concepts of energy and power can be used interchangeably.

[0008] The system energy consumption model includes: the energy consumption of each micro base station is It includes circuit, heat dissipation cooling, backup battery and transmission power of and static power P c,i , and meet in: is the maximum energy consumption allowed for each micro base station, and the beamforming vector of the micro base station connected to user k in time slot t is When user k is associated with micro base station i, the binary variable a i,k =1, otherwise a i,k = 0. Transmitting power For Select the corresponding row and column in to form the transmission beamforming vector of the i-th micro base station. The operating energy consumption of each micro base station in time slot t is

[0009] The renewable energy output model includes: the renewable energy obtained in time slot t And following the independent and identical distribution, the battery's charge and discharge capacity satisfies: The battery capacity meets the following requirements: The dynamic equation of the battery level at the base station side is: in: is the charge (discharge) capacity of the battery in time slot t. When charging the battery, When , the battery is discharged; and are the maximum amount of battery discharge and charge, respectively. is the energy state of the base station at the beginning of time slot t, is the minimum allowed battery energy level, is the maximum allowed battery energy level.

[0010] The two-way transaction model between the micro base station and the smart grid includes: the energy purchased by the micro base station is obtained from the two-way energy transaction mechanism: The energy sold by the micro base station is The cost of a micro base station is: Where: The price of the grid energy unit to purchase renewable energy at time slot t α t , the price of purchasing existing energy β t , electricity selling price γ t , α t , β t and γ t Always satisfy α t ≥β t ≥γ t>0, m∈[0,1] is the proportion of renewable energy in the purchased energy, (1-m) is the proportion of existing energy in the purchased energy; the purchase amount of existing energy satisfies: Where: μ is the carbon emission coefficient of existing energy (unit), the maximum time average carbon footprint

[0011] The optimization mentioned above refers to: constructing and decoupling the user-base station association and resource allocation problems, that is, after determining the user-base station association in advance, minimizing the long-term average cost of the system while satisfying the system computing resources, battery charging and discharging, and capacity level, specifically including:

[0012] 1) Constructing the user-base station association and resource allocation problem of mixed integer nonlinear problem The constraints include: Where: The price of purchasing renewable energy α t , the price of purchasing existing energy β t , electricity selling price γ t All are random; and are the maximum amount of battery discharge and charge, respectively, and is the charge (discharge) capacity of the battery in time slot t. When charging the battery, When , the battery is discharged; is the energy state of the base station at the beginning of time slot t, is the minimum allowed battery energy level, is the maximum battery energy level allowed. At the same time, an auxiliary variable is introduced and Cost of micro base stations

[0013] 2) Apply user mobility prediction to optimize user association performance and resource allocation, and then decouple the joint user association and resource allocation problem into two sub-problems: predict the user's mobility trajectory through the flashback architecture neural network, and determine the user-base station association variable a through multi-attribute decision making. i,k Finally, for a given user-base station association relationship, the stochastic subgradient method is used to solve the long-term average minimization problem of system cost.

[0014] The flashback architecture neural network includes: an embedding layer, a graph convolution (GCN) layer, an aggregation layer and a prediction layer, wherein: the embedding layer uses one-hot encoding to encode user and POI information; the GCN layer encodes the semantic information and spatial relationship of the POI transfer matrix into a vector representation to be integrated into the original POI Embeding; the aggregation layer inputs the POIEWmbedding and user preferences updated by the output of the GCN layer into the aggregation layer, uses RNN to obtain hidden layer information, and uses time, space information and user preference information to update the RNN hidden layer; the prediction layer merges the output of each time step of the aggregation layer and the user Embedding, enters the full connection to obtain the final output, and uses the cross entropy loss function.

[0015] The multi-attribute decision-making method is to use the distance between user base stations, the energy transaction cost of each base station and the green energy output of the base station as decision attributes, and use the Euclidean distance and grey correlation degree to evaluate the association schemes, and select the association scheme with the best attributes. i,k , specifically including:

[0016] i) Trajectory prediction: The sparse user trajectory is used as a training set, and the user's movement trajectory is predicted through flashback neural network training to calculate the distance between the user and the micro base station.

[0017] ii) Multi-attribute decision model: The calculated distance, energy transaction cost of each base station and green energy output of the base station are taken as attributes to construct an associated decision matrix, which is then combined with weights and normalized.

[0018] iii) Best association scheme selection: Determine the Euclidean distance and grey association degree between all association schemes, and comprehensively select the best association scheme to obtain a i,k .

[0019] The long-term average operating cost minimization problem is solved by the stochastic dual subgradient method to obtain distributed online power control and resource scheduling configuration, which specifically includes:

[0020] a) The system cost optimization problem is expressed as The constraints include: Among them: Based on neural network and multi-attribute decision making, select the best attribute association scheme a i,k .

[0021] b) Adopt offline control strategy and relax the constraints in step a): Using the semidefinite programming (SDP) technique to relax the constraints, problem P2 can be relaxed to The constraints are in:

[0022] c) The dual problem of problem P3 is expressed by the Lagrange dual function, specifically: Where: Lagrange dual function The set of all variables that satisfy the constraints for each time slot t is the Lagrange multiplier associated with the constraint.

[0023] d) The stochastic dual subgradient method is used to solve the problem, that is, iterative calculation is performed time slot by time slot: Where: η>0 is the appropriate step size, And meet

[0024] Preferably, the following iterations are performed: Where: η>0 is the appropriate step size, For variables A random estimate of By using Solve for λ Get, that is Specifically: Minimize L t (x t ,λ) to get the decision of each time slot And use this decision The optimized variable values ​​obtained in iterative update

[0025] The distributed online power control algorithm (DOCA) specifically includes:

[0026] ① Initialization: First select a suitable step size η and initialize the appropriate

[0027] ②Solve the problem: Solve the minimization objective problem L in each time slot t t (x t ,λ): The optimal beamforming optimization and energy trading decision for each micro base station are obtained, respectively. and

[0028] ③Execution decision: Based on the obtained Micro base stations perform beamforming; based on Carry out two-way energy trading between micro base stations and power grids, as well as control the charging and discharging of micro base station batteries.

[0029] ④ Lagrange multiplier update: At each time slot t, as and Known, update the Lagrange multiplier

[0030] The present invention relates to a system for implementing the above method, comprising: a user-base station association optimization unit, a micro base station beamforming optimization unit and a two-way energy trading unit, wherein: the user-base station association optimization unit predicts the user's movement trajectory through a neural network and selects the optimal association scheme in each time slot by combining multiple attribute decisions; the micro base station beamforming optimization unit calculates the power consumption caused by the data transmission of each user according to the beamforming decision of its antenna in each time slot; the two-way energy trading unit determines the amount of energy traded with the smart grid through the current market price of the smart grid and the renewable energy collected by the micro base station at the end of each time slot, and controls the charging and discharging of the battery while updating the energy queue of the micro base station at the end of each time slot. Technical Effects

[0031] The present invention combines the wireless network micro base station system scenario of the smart grid, constructs a model that minimizes the long-term average total cost of the system, introduces neural networks and multi-attribute decision-making to determine the user-base station association scheme in advance, and introduces the random dual sub-gradient method to solve the problem after relaxation. The long-term average operating cost sub-problem makes an optimization decision online in each time slot without prior information. Compared with the existing wireless network micro base station system, the wireless network micro base station system of the present invention introduces the smart grid technology, combines the two-way energy trading mechanism and the dynamic output characteristics of renewable energy, and jointly optimizes user-base station association, resource allocation and energy management. The present invention can deploy renewable energy under the minimum total cost of the system, and only requires the system information of the current time slot to realize distributed online control decision-making, and minimize the long-term average cost of the system while ensuring the user QoS requirements. The distributed online control method in the present invention can obtain feasible and asymptotically optimal results, and can dynamically make instant decisions without prior knowledge of any channel, renewable energy and electricity price changes and other random process statistics. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the present invention;

[0033] Figure 2 Scenario diagram for wireless network systems powered by energy harvesting and smart grids;

[0034] Figure 3 It is a flowchart of the joint algorithm of neural network and multi-attribute decision making;

[0035] Figure 4 It is a schematic diagram of the DOCA algorithm flow;

[0036] Figure 5 This is a schematic diagram comparing the long-term average cost of the system under different algorithms. DETAILED DESCRIPTION

[0037] like Figure 1 As shown, this embodiment relates to a joint user-base station association and resource optimization configuration method, including:

[0038] Step 1: Establish implementation system model: Price of purchasing green energy α t Generated by the folded normal distribution, the price of purchasing existing energy β t and the price of selling energy γ t Satisfy β t = wα t and γ t =vα t , where w = 0.8, v = 0.6. Number of micro base stations I = 3, number of users K = 10. Energy harvesting The wind speed is fitted according to the Weibull distribution. The capacity of each battery is limited to At the same time, the initial battery energy level The energy queue value of the micro base station is initialized at the given initial time slot, and the appropriate η value is selected.

[0039] Step 2: User mobility prediction: Use the sparse user trajectory data as a training set to predict future user mobility trajectories and calculate the distance between the user and the micro base station.

[0040] Step 3: Determine the user-base station association decision: Take the calculated distance, base station green energy output and base station energy transaction cost as the attributes of the multi-attribute decision model and select the best association decision a i,k .

[0041] Step 4: Decision based on the best association i,k , execute the DOCA algorithm to realize the online operation of the micro base station network, that is, the two-way energy transaction with the smart grid at the same time, including:

[0042] 4.1) Micro-base station long-term average cost minimization sub-problem: At each time slot t, solve the beamforming problem and energy trading problem to obtain the optimal beamforming decision and energy trading decisions

[0043] 4.2) Solve the optimization problem: At the same time in: From α t >β t >γ t >0, we get and ψ t >φ t >0, after the above transformation, it can be clearly seen that About and The objective function is also about and The convex function of , the optimal beamforming and energy trading decision is obtained through the convex solver

[0044] Step 5: Execution decision: based on Micro base stations perform antenna beamforming; based on Micro base stations perform battery charging and discharging and conduct two-way energy transactions with smart grids.

[0045] Step 6: Lagrange multiplier update: At each time slot t, update the micro base station battery status and update the Lagrange multiplier accordingly, specifically:

[0046] After specific practical experiments, the software was used to model the wireless network micro base station system powered by the smart grid, and the user-base station association algorithm (UA) and the distributed online control algorithm (DOCA) were used to configure the system resources, and compared with the online greedy method (Greedy algorithm), the DOCA-noUA method that does not consider the user-base station association, and the DOCA-noRES method that does not consider renewable energy. Figure 5 The results of the long-term average cost of the system under different algorithms within 500 time slots are shown. It can be seen that the use of the DOCA method to configure the smart grid-driven wireless network micro base station system can achieve a smaller long-term average cost of the system while ensuring battery stability and user QoS.

[0047] In summary, compared with the prior art, the present invention is based on a wireless network micro base station system driven by a smart grid, and models energy consumption, batteries, and two-way energy trading. The goal is to minimize the long-term average cost of the system while ensuring user QoS. Through neural networks and multi-attribute decision-making methods, a user-base station association scheme is obtained in advance; through the random dual subgradient method, a distributed online resource allocation method is obtained, which is a resource optimization configuration method, and solves the resource allocation and energy management optimization problems of a wireless network micro base station system driven by a smart grid; the long-term average cost of the system brought about by the present invention can be more similar to the optimal solution obtained by known a priori future system information, and decisions can be made in advance and immediately to modulate and optimize wireless resources, thereby reducing the time consumption caused by decision-making.

[0048] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A method for joint user-base station association and resource optimization configuration, characterized in that: By establishing a wireless network micro base station system scenario, a system energy consumption model, a renewable energy output model, and a two-way transaction model between micro base stations and smart grids, the constructed model is optimized by predicting user mobility trajectories based on a neural network, and then the optimal user-base station association decision is obtained through multi-attribute decision-making. The long-term average operating cost minimization problem is solved by the stochastic dual subgradient method, and resource scheduling and configuration are performed through a distributed online power control algorithm to minimize the long-term average cost while ensuring service quality. The wireless network micro base station system scenario includes: a multi-micro base station wireless network system powered by a smart grid composed of I = {1, 2, ..., I} micro base stations and K = {1, 2, ..., K} mobile users, wherein each micro base station and user is equipped with M and 1 antennas respectively; The micro base station is provided with an energy collection device and a rechargeable battery. The micro base station collects renewable energy from the environment through the energy collection device to reduce dependence on the smart grid; The system energy consumption model includes: the energy consumption of each micro base station is It includes circuit, heat dissipation cooling, backup battery and transmission power of and static power P c,i , and meet in: is the maximum energy consumption allowed for each micro base station, and the beamforming vector of the micro base station connected to user k in time slot t is When user k is associated with micro base station i, the binary variable a i,k =1, otherwise a i,k =0, transmit power For Select the corresponding rows and columns in order to form the transmission beamforming vector of the i-th micro base station. The operating energy consumption of each micro base station in time slot t is The two-way transaction model between the micro base station and the smart grid includes: the energy purchased by the micro base station is obtained from the two-way energy transaction mechanism: The energy sold by the micro base station is The cost of a micro base station is: Where: The price of the grid energy unit to purchase renewable energy at time slot t α t , the price of purchasing existing energy β t , electricity selling price γ t , α t , β t and γ t Always satisfy α t ≥β t ≥γ t >0, m∈[0,1] is the proportion of renewable energy in the purchased energy, (1-m) is the proportion of existing energy in the purchased energy; the purchase amount of existing energy satisfies: Where: μ is the carbon emission coefficient of existing energy (unit), the maximum time average carbon footprint 2. The method for joint user-base station association and resource optimization configuration according to claim 1, characterized in that: The renewable energy output model includes: the renewable energy obtained in time slot t And following the independent and identical distribution, the battery's charge and discharge capacity satisfies: The battery capacity meets the following requirements: The dynamic equation of the battery level at the base station side is: in: is the charge (discharge) capacity of the battery in time slot t. When charging the battery, When , the battery is discharged; and are the maximum amount of battery discharge and charge, respectively. is the energy state of the base station at the beginning of time slot t, is the minimum allowed battery energy level, is the maximum allowed battery energy level.

3. The method for joint user-base station association and resource optimization configuration according to claim 1, characterized in that: The optimization mentioned above refers to: constructing and decoupling the user-base station association and resource allocation problems, that is, after determining the user-base station association in advance, minimizing the long-term average cost of the system while satisfying the system computing resources, battery charging and discharging, and capacity level, specifically including: 1) Constructing the user-base station association and resource allocation problem of mixed integer nonlinear problem P1: The constraints include: Where: The price of purchasing renewable energy α t , the price of purchasing existing energy β t , electricity selling price γ t All are random; and are the maximum amount of battery discharge and charge, respectively, and is the charge (discharge) capacity of the battery in time slot t. When charging the battery, When , the battery is discharged; is the energy state of the base station at the beginning of time slot t, is the minimum allowed battery energy level, is the maximum allowed battery energy level, and an auxiliary variable is introduced and Cost of micro base stations 2) Apply user mobility prediction to optimize user association performance and resource allocation, and then decouple the joint user association and resource allocation problem into two sub-problems: predict the user's mobility trajectory through the flashback architecture neural network, and determine the user-base station association variable a through multi-attribute decision making. i,k Finally, for a given user-base station association relationship, the stochastic subgradient method is used to solve the long-term average minimization problem of system cost.

4. The method for joint user-base station association and resource optimization configuration according to claim 3, characterized in that: The flashback architecture neural network includes: an embedding layer, a graph convolution (GCN) layer, an aggregation layer and a prediction layer, wherein: the embedding layer uses one-hot encoding to encode user and POI information; the GCN layer encodes the semantic information and spatial relationship of the POI transfer matrix into a vector representation to be integrated into the original POIEmbeding; the aggregation layer inputs the POIEWmbedding and user preferences updated by the output of the GCN layer into the aggregation layer, uses RNN to obtain hidden layer information, and uses time, space information and user preference information to update the RNN hidden layer; the prediction layer merges the output of each time step of the aggregation layer and the user Embedding, enters the full connection to obtain the final output, and uses the cross entropy loss function.

5. The method for joint user-base station association and resource optimization configuration according to claim 1, characterized in that: The multi-attribute decision-making method is to use the distance between user base stations, the energy transaction cost of each base station and the green energy output of the base station as decision attributes, and use the Euclidean distance and grey correlation degree to comprehensively evaluate the association schemes, and select the association scheme with the best attributes. i,k , specifically including: i) Trajectory prediction: Using sparse user trajectories as training sets, the user's movement trajectory is predicted through flashback neural network training, and the distance between the user and the micro base station is calculated; ii) Multi-attribute decision model: taking the calculated distance, energy transaction cost of each base station and green energy output of the base station as attributes, the associated decision matrix is ​​constructed, followed by combination weighting and normalization; iii) Best association scheme selection: Determine the Euclidean distance and grey association degree between all association schemes, and comprehensively select the best association scheme to obtain a i,k .

6. The method for joint user-base station association and resource optimization configuration according to claim 1, characterized in that: The long-term average operating cost minimization problem is solved by the stochastic dual subgradient method to obtain distributed online power control and resource scheduling configuration, which specifically includes: a) Formulate the system cost optimization problem as P2: The constraints include: Among them: Based on neural network and multi-attribute decision making, select the best attribute association scheme a i,k ; b) Adopt offline control strategy and relax the constraints in step a): The constraints are relaxed using the semi-definite programming (SDP) technique to relax problem P2 to P3: The constraints are in: c) The dual problem of problem P3 is expressed by the Lagrange dual function, specifically: Where: Lagrange dual function The set of all variables that satisfy the constraints for each time slot t is the Lagrange multiplier associated with the constraint; d) The stochastic dual subgradient method is used to solve the problem, that is, iterative calculation is performed time slot by time slot: Where: η>0 is the appropriate step size, And meet 7. The method for joint user-base station association and resource optimization configuration according to claim 6, characterized in that: Perform the following iterations: Where: η>0 is the appropriate step size, For variables A random estimate of By using Solve for λ Get, that is Specifically: Minimize L t (X t ,λ) to get the decision of each time slot And use this decision The optimized variable values ​​obtained in iterative update 8. The method for joint user-base station association and resource optimization configuration according to claim 1, characterized in that: The distributed online power control algorithm (DOCA) specifically includes: ① Initialization: First select a suitable step size η and initialize the appropriate ②Solve the problem: Solve the minimization objective problem L in each time slot t t (X t ,λ): The optimal beamforming optimization and energy trading decision for each micro base station are obtained, respectively. and ③Execution decision: Based on the obtained Micro base stations perform beamforming; based on Carry out two-way energy transactions between micro base stations and power grids, as well as control the charging and discharging of micro base station batteries; ④ Lagrange multiplier update: At each time slot t, as and Known, update the Lagrange multiplier 9. A system for joint user-base station association and resource optimization configuration for implementing any of the methods described in claims 1-8, characterized in that: include: The user-base station association optimization unit, the micro base station beamforming optimization unit and the two-way energy trading unit, wherein: the user-base station association optimization unit predicts the user's movement trajectory through a neural network and selects the optimal association scheme through a joint multi-attribute decision-making in each time slot; the micro base station beamforming optimization unit calculates the power consumption caused by the data transmission of each user according to the beamforming decision of its antenna in each time slot; the two-way energy trading unit determines the amount of energy traded with the smart grid at the end of each time slot through the current market price of the smart grid and the renewable energy collected by the micro base station, and controls the charging and discharging of the battery while updating the energy queue of the micro base station at the end of each time slot.

Images