Flow prediction method based on deep learning and 5G heterogeneous BSMG cooperative dormancy and energy sharing method and system based on improved MOEAD algorithm

Abstract

The invention discloses a 5G heterogeneous BSMG traffic prediction method and system based on deep learning. The method comprises the following steps: S1, obtaining current traffic data of each base station microgrid; and S2, inputting into a deep learning model of CNN-DAM-BiLSTM, and outputting to obtain flow data of a next time period. The invention also discloses a 5G heterogeneous BSMG cooperative dormancy and energy sharing method and system, and the method comprises the steps: T1, carrying out the dormancy operation when a base station microgrid is lower than a dormancy load threshold value based on the predicted flow data and the dormancy load threshold value; and T2, obtaining the maximum demand rate, the energy sharing amount, the purchased energy, the dormancy state and the active state of the user, inputting into the 5G heterogeneous BSMG collaborative dormancy and energy sharing model, and outputting to obtain a dormancy and energy sharing strategy. According to the method, future flow data can be accurately predicted, the communication quality of users can be ensured to a great extent, and meanwhile, the energy efficiency of the mobile cellular network is improved.

Description

Technical Field

[0001] The present invention relates to the field of smart grid information engineering. More specifically, the present invention relates to a 5G heterogeneous BSMG collaborative sleep and energy sharing method and system based on a deep learning-based traffic prediction method and an improved MOEAD algorithm. Background Art

[0002] With the continuous development of information and communication technology (ICT) on the fifth generation communication technology (5G) and more advanced mobile networks, the number of mobile users and IoT devices will explode. In order to meet the requirements of low latency and high reliability of future wireless devices, 5G heterogeneous networks can make network management more efficient and flexible through the dense deployment of micro base stations and macro base stations. At the same time, the energy consumption generated by large-scale base station deployment and high-load operation and its impact on the environment are becoming increasingly significant, which means that ICT will become the main source of global energy consumption and greenhouse gas emissions in the future. Therefore, building a green 5G heterogeneous network is crucial to the development of the ICT industry.

[0003] To meet these challenges, the application of sustainable renewable energy (such as solar and wind energy) can effectively reduce the electricity bills and carbon emissions of base stations. However, traditional power grids may not be able to fully absorb green energy and eventually burden the environment. Microgrids are an important part of smart grids. The interconnected heterogeneous base station microgrid (BSMG) system formed by accessing renewable energy is conducive to the two-way flow of information and energy, and has become a promising solution for realizing green communication networks in the future. Base station sleep aims to switch the base station to sleep mode at an appropriate time, which will reduce its excessive energy consumption to a certain extent. Users within its coverage will be connected to other active base stations, which may reduce the user's quality of service (QoS). But on the other hand, considering the randomness and intermittency of traffic and renewable energy in the base station network, it is necessary to ensure accurate prediction of BSMG traffic demand and green energy arrival to solve potential information and energy collaborative management problems. Summary of the invention

[0004] An object of the present invention is to solve at least the above problems and to provide at least the advantages which will be described hereinafter.

[0005] In order to achieve these purposes and other advantages according to the present invention, a traffic prediction method for 5G heterogeneous BSMG based on deep learning is provided, comprising the following steps:

[0006] S1. Obtain current flow data of each base station microgrid, wherein the base station microgrid includes a central base station microgrid and a plurality of sub-base station microgrids within the coverage of the central base station microgrid;

[0007] S2. Input the current flow data into a trained flow prediction model, and output the predicted flow data of each base station microgrid in the next time period, wherein the flow prediction model is a CNN-DAM-BiLSTM deep learning model, and the method for predicting the flow data of the next time period by the flow prediction model specifically includes:

[0008] Extract the flow features of the flow data, and use the convolutional block attention mechanism (CBAM) to perform channel feature weighting and spatial feature weighting on the flow features to obtain the CBAM-weighted flow feature representation;

[0009] The traffic feature representation weighted by CBAM is used as the input of the bidirectional long short-term memory network BiLSTM, and the multi-head attention mechanism MHAM is used to weight the output of the bidirectional long short-term memory network BiLSTM and then map and output the traffic data of the next time period.

[0010] Preferably, the convolutional block attention mechanism CBAM is specifically as follows: first, a global average pooling operation of the channel attention features is performed on the flow features extracted by the convolutional layer to obtain the channel attention weights, and then a maximum pooling operation of the spatial attention features is performed to obtain the spatial attention weights, and finally, a Sigmoid activation function and feature multiplication are used to obtain the spatial attention weight vector of each channel and the weighted feature representation, i.e., the CBAM weighted flow feature representation.

[0011] A 5G heterogeneous BSMG cooperative sleep and energy sharing method based on an improved MOEA / D algorithm is provided, comprising the following steps:

[0012] T1. Based on the flow data of each base station microgrid in the next time period predicted by the flow prediction method, according to the preset sleep load threshold, the base station microgrid is in a no-load state or is lower than the sleep load threshold, and a sleep base station microgrid set and an active base station microgrid set are obtained;

[0013] T2, obtain the user's maximum demand rate, energy sharing amount and energy purchased from the smart grid, as well as the dormant state and active state of the base station microgrid obtained in step T1, and input them into a trained 5G heterogeneous BSMG collaborative dormancy and energy sharing model, and output the dormant base station microgrid set, active base station microgrid set and energy sharing strategy for the next time period;

[0014] Among them, the 5G heterogeneous BSMG collaborative sleep and energy sharing model takes the BSMG system benefit and the user's QoS benefit as the objective function, the user's maximum demand rate, the sleep state and active state of the base station microgrid, the energy sharing amount and the energy purchased from the smart grid as the decision variables, and the user's signal-to-noise ratio SINR threshold limit, the base station microgrid overload limit, the energy storage capacity limit, the energy sharing and the smart grid energy usage limit as the constraints. The improved MOEA / D algorithm is used to solve the energy sharing strategy, wherein the improved MOEA / D algorithm uses the quantum local search method to adaptively adjust the neighborhood value size.

[0015] Preferably, when energy is shared, the rule of energy flow is to prioritize green energy sharing between base station microgrids, and the base station microgrid with surplus green energy transmits green energy to the base station microgrid with energy shortage;

[0016] When each base station microgrid cannot meet the energy demand of the base station microgrid which is in an energy shortage state, energy is purchased from the smart grid to provide it with energy.

[0017] Preferably, the green energy is electric energy provided by a photovoltaic power generation system.

[0018] A traffic prediction system based on the traffic prediction method of 5G heterogeneous BSMG based on deep learning is provided, comprising:

[0019] A data acquisition module, which is used to obtain the current flow data of each base station microgrid, wherein the base station microgrid includes a central base station microgrid and multiple sub-base station microgrids within the coverage area of ​​the central base station microgrid;

[0020] The flow prediction module is used to input the current flow data into a trained flow prediction model, and output the predicted flow data of each base station microgrid in the next time period, wherein the flow prediction model is a CNN-DAM-BiLSTM deep learning model, and the method for predicting the flow data of the next time period by the flow prediction model specifically includes:

[0021] Extract the flow features of the flow data, and use the convolutional block attention mechanism (CBAM) to perform channel feature weighting and spatial feature weighting on the flow features to obtain the CBAM-weighted flow feature representation;

[0022] The traffic feature representation weighted by CBAM is used as the input of the bidirectional long short-term memory network BiLSTM, and the multi-head attention mechanism MHAM is used to weight the output of the bidirectional long short-term memory network BiLSTM and then map and output the traffic data of the next time period.

[0023] Provided is a collaborative sleep and energy sharing system based on the 5G heterogeneous BSMG collaborative sleep and energy sharing method, comprising:

[0024] A sleep module, which is used to perform a sleep operation when the base station microgrid is idle or below the sleep load threshold based on the flow data of each base station microgrid in the next time period predicted by the flow prediction method, and obtain a sleep base station microgrid set and an active base station microgrid set;

[0025] An energy sharing module is used to obtain the user's maximum demand rate, energy sharing amount, and energy purchased from the smart grid, as well as the dormant state and active state of the base station microgrid obtained in step T1, and input them into a trained 5G heterogeneous BSMG collaborative dormancy and energy sharing model, and output the dormant base station microgrid set, active base station microgrid set, and energy sharing strategy for the next time period;

[0026] Among them, the 5G heterogeneous BSMG collaborative sleep and energy sharing model takes the BSMG system benefit and the user's QoS benefit as the objective function, the user's maximum demand rate, the sleep state and active state of the base station microgrid, the energy sharing amount and the energy purchased from the smart grid as the decision variables, and the user's signal-to-noise ratio SINR threshold limit, the base station microgrid overload limit, the energy storage capacity limit, the energy sharing and the smart grid energy usage limit as the constraints. The improved MOEA / D algorithm is used to solve the energy sharing strategy, wherein the improved MOEA / D algorithm uses the quantum local search method to adaptively adjust the neighborhood value size.

[0027] Preferably, each central base station microgrid is configured with a corresponding photovoltaic power generation system and a smart grid power supply system, and each sub-base station microgrid is configured with a corresponding photovoltaic power generation system.

[0028] The present invention provides an electronic device, comprising:

[0029] one or more processors;

[0030] And a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the traffic prediction method or the 5G heterogeneous BSMG collaborative sleep and energy sharing method.

[0031] A computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the traffic prediction method or the 5G heterogeneous BSMG collaborative sleep and energy sharing method is implemented.

[0032] The present invention has at least the following beneficial effects:

[0033] First, the convolutional block attention mechanism CBAM and the multi-head attention mechanism MHAM are used to form a dual attention mechanism to optimize the convolutional neural network CNN and the bidirectional long short-term memory network BiLSTM, and a 5G heterogeneous BSMG traffic prediction model based on CNN-DAM-BiLSTM is established. While increasing the weight of important channel features, more feature information can be learned in different spaces, ultimately improving the accuracy of traffic data prediction.

[0034] Second, the proposed 5G heterogeneous BSMG collaborative sleep and energy sharing model fully considers the user experience, maximizing the benefits of the BSMG system while improving the network energy efficiency of the BSMG system.

[0035] Third, the MOEA / D algorithm is improved through quantum local search and adaptive mutation strategies to improve its convergence speed and global search capability, and finally the two-way flow of information and energy in the 5G heterogeneous BSMG system is realized.

[0036] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a structural diagram of a 5G heterogeneous BSMG collaborative sleep and energy sharing method based on deep learning and improved MOEA / D algorithm according to Embodiment 1 of the present invention;

[0038] Figure 2 This is a flow chart of a traffic prediction model for 5G heterogeneous BSMG based on CNN-DAM-BiLSTM according to Example 2 of the present invention;

[0039] Figure 3 A comparison chart of flow prediction curves of different flow prediction models within 24 hours in Example 2 of the present invention;

[0040] Figure 4 The flowchart of the improved MOEA / D algorithm based on quantum local search and adaptive mutation strategy of the present invention;

[0041] Figure 5 A comparison chart of the average system benefits of 5G heterogeneous BSMGs using different multi-objective optimization algorithms provided in Example 3 of the present invention;

[0042] Figure 6 A comparison chart of system benefits of a 5G heterogeneous BSMG system under five different energy-saving algorithms provided in Example 3 of the present invention;

[0043] Figure 7 A comparison chart of network energy efficiency of a 5G heterogeneous BSMG system under five different energy-saving algorithms provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0045] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the orientation or positional relationship indicated by the terms is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0046] like Figures 1 to 7 As shown, the present invention provides a traffic prediction method for 5G heterogeneous BSMG based on deep learning, comprising the following steps:

[0047] S1. Obtain current flow data of each base station microgrid, wherein the base station microgrid includes a central base station microgrid and a plurality of sub-base station microgrids within the coverage of the central base station microgrid;

[0048] S2. Input the current flow data into a trained flow prediction model, and output the predicted flow data of each base station microgrid in the next time period, wherein the flow prediction model is a CNN-DAM-BiLSTM deep learning model, and the method for predicting the flow data of the next time period by the flow prediction model specifically includes:

[0049] Extract the flow features of the flow data, and use the convolutional block attention mechanism (CBAM) to perform channel feature weighting and spatial feature weighting on the flow features to obtain the CBAM-weighted flow feature representation;

[0050] The traffic feature representation weighted by CBAM is used as the input of the bidirectional long short-term memory network BiLSTM, and the multi-head attention mechanism MHAM is used to weight the output of the bidirectional long short-term memory network BiLSTM and then map and output the traffic data of the next time period.

[0051] In the above technical scheme, a W-layer 5G heterogeneous BSMG system is constructed. The first layer is the central base station microgrid (CBSMG), and the other W-1 layers are sub-base station microgrids (SBSMG). The base station network is microcells, millicells, etc., and the base station transmission power of different layers is different. The present invention mainly studies the downlink transmission system with W=2, and realizes the bidirectional flow of information and energy between each BSMG through a centralized control unit.

[0052] The main difference between the central base station microgrid CBSMG and the sub-base station microgrid SBSMG is that CBSMG can provide low-speed services and capacity, while SBSMG is deployed in areas with large user traffic within the coverage of CBSMG to enhance network capacity and provide high-speed services. Each BSMG is equipped with renewable energy power generation devices and energy storage devices. In addition, CBSMG will serve as an information and energy management unit, that is, a control unit, to manage multiple SBSMGs and mobile users in its coverage area. CBSMG is powered by its own green energy power generation equipment and an external smart grid. The green energy power generation equipment includes photovoltaic modules and energy storage devices. When green energy is insufficient, it is powered by an external smart grid to meet the needs. While SBSMG has its own green energy power generation equipment, its energy management is controlled by CBSMG. When energy is insufficient, the external smart grid can power each SBSMG through the CBSMG system to ensure the user's QoS requirements. Since CBSMG also has communication units such as AAU and BBU, it can also be associated with mobile users. On the other hand, when the SBSMG is under heavy load and its own green energy is insufficient, it can cooperate with other SBSMGs to fully utilize green energy; when the SBSMG is idle or under low load, it can be switched to sleep mode to save energy.

[0053] Considering the differences in BSMG traffic in different time periods, a deep learning model of CNN-DAM-BiLSTM was proposed based on convolutional neural network (CNN) and bidirectional long short-term memory network (BiLSTM) as a traffic prediction model. The convolutional block attention mechanism and multi-head attention mechanism were added to enhance the ability to focus on and capture key information in time series.

[0054] Specifically, the structure of the convolutional neural network CNN is mainly composed of convolutional layers, pooling layers and fully connected layers. However, when processing time series prediction models, traditional convolutional neural networks CNN may face the problem of information loss or local features that cannot fully reflect the overall sequence trend. The attention mechanism refers to the use of limited computing resources as fully as possible when processing large amounts of data, and using them to process more important data. Based on this, the present application applies the convolutional block attention mechanism (CBAM) to the convolutional neural network CNN. By considering information interaction on different time scales and introducing a weight distribution mechanism, the convolutional neural network CNN can more effectively improve its learning ability for important channel feature information, thereby improving the prediction accuracy of the traffic prediction model. The convolutional block attention mechanism CBAM consists of two parts: a channel attention module (CAM) and a spatial attention module (SAM). The channel attention module CAM first performs global average pooling on the traffic features extracted by the convolution layer, then connects them through the fully connected layer, and then performs maximum pooling through the spatial attention module SAM. Finally, the sigmoid activation function and feature multiplication are used to obtain the spatial attention weight vector of each channel and the weighted feature representation. The channel attention module can enhance the correlation between different feature channels and eliminate the redundant information in the channel, which helps the traffic prediction model focus on important feature channels. The spatial attention module SAM performs maximum pooling on the channel dimension of the features after weighted processing by the channel attention module CAM to obtain the maximum value of the channel response at each spatial position, thereby strengthening important features. It can enhance different spatial positions in the features and help the traffic prediction model focus on important spatial areas. Therefore, after integrating the channel attention module and the spatial attention module into each convolution block of the convolutional neural network CNN, the output features of each convolution block first pass through the channel attention module and then pass through the spatial attention module. In this way, the features of each convolution block are enhanced to be more focused on the representation of important channels and spatial positions, that is, the CBAM-weighted traffic feature representation.

[0055] Specifically, the state unit of the bidirectional long short-term memory network will update and save the historical state, so the output of each step is closely related to the current input and historical input. If the input time series is long, it will be limited by the storage space of the state unit, which may cause the loss of information at important time nodes. Based on this, when predicting the traffic of BSMG, this application introduces a multi-head attention mechanism (MHAM) into the bidirectional long short-term memory network BiLSTM to enhance the ability to pay attention to and capture different aspects of the time series. The multi-head attention mechanism can independently focus on different aspects by calculating multiple attention heads in parallel, and adopt their own weight distribution methods to analyze the input information from different angles. The multi-head attention mechanism converts the vector output by the bidirectional long short-term memory network BiLSTM into three input matrices through different mapping operations: query q (Query), key k (Key) and value v (Value), and the corresponding weights obtained through a large amount of training are w respectively. Q 、w K 、w V , from which the allocation weight of each hidden layer vector is obtained as:

[0056] Q=w Q q (1)

[0057] K=w K k (2)

[0058] V=w V v (3)

[0059] The multi-head attention mechanism divides the time series into L subspaces. Each head performs self-attention calculation on the subspace, and then obtains multiple heads and concatenates each head. The attention output matrix is ​​normalized by the softmax function as follows:

[0060]

[0061] where d k It is the feature dimension of each key, used for weight scaling, and is normalized to the [0,1] interval by Softmax.

[0062] MultiHead(Q,K,V)=Concat(head1,...,head l ) l (5)

[0063] In formula (5), w l is the weight of the linear transformation, head l represents the Lth head of the multi-head attention module, and Concat represents the concatenation operation.

[0064] Specifically, the CBAM weighted traffic feature representation obtained by the output of the convolutional neural network CNN is used as the input of the bidirectional long short-term memory network BiLSTM. The output of the bidirectional long short-term memory network BiLSTM is first weighted by the multi-head attention mechanism, and then the comprehensive feature information is passed to the fully connected layer to realize the prediction of traffic in the future.

[0065] The present invention combines the convolutional block attention mechanism CBAM and the multi-head attention mechanism MHAM to optimize the convolutional neural network CNN and the bidirectional long short-term memory network BiLSTM respectively, and establishes a flow prediction model of the base station microgrid BSMG based on CNN-DAM-BiLSTM. In the feature extraction stage, by introducing the CBAM mechanism in CNN, the channel attention module and the spatial attention module are integrated after each convolution block of CNN. The output feature map of each convolution block first passes through the channel attention module, and then passes through the spatial attention module to increase the weight of important channel features, and finally passes through the pooling layer and the fully connected layer. Since BiLSTM has the problem of longitudinal time series information loss, in order to better capture the correlation between long time series data, multi-head attention is added to the BiLSTM model to form the MHAM-BiLSTM model. The BiLSTM network extracts features with long-term correlation characteristics from the traffic feature vector extracted by the optimized convolutional neural network CNN through learning, and based on this, combined with the multi-head attention mechanism, weights the output of the BiLSTM network, so that it can learn more feature information in different spaces and pass the weighted comprehensive feature information to the fully connected layer to realize the prediction of traffic in the future.

[0066] The present invention also provides a 5G heterogeneous BSMG collaborative sleep and energy sharing method based on the traffic prediction method and the improved MOEA / D algorithm, comprising the following steps:

[0067] T1, based on the flow data of each base station microgrid in the next time period predicted by the flow prediction method, according to the preset sleep load threshold, perform a sleep operation when the base station microgrid is in no-load or below the sleep load threshold, and obtain a sleep base station microgrid set and an active base station microgrid set;

[0068] T2, obtain the user's maximum demand rate, energy sharing amount and energy purchased from the smart grid, as well as the dormant state and active state of the base station microgrid obtained in step T1, and input them into a trained 5G heterogeneous BSMG collaborative dormancy and energy sharing model, and output the dormant base station microgrid set, active base station microgrid set and energy sharing strategy for the next time period;

[0069] Among them, the 5G heterogeneous BSMG collaborative sleep and energy sharing model takes the BSMG system benefit and the user's QoS benefit as the objective function, the user's maximum demand rate, the sleep state and active state of the base station microgrid, the energy sharing amount and the energy purchased from the smart grid as the decision variables, and the user's signal-to-noise ratio SINR threshold limit, the base station microgrid overload limit, the energy storage capacity limit, the energy sharing and the smart grid energy usage limit as the constraints. The improved MOEA / D algorithm is used to solve the energy sharing strategy, wherein the improved MOEA / D algorithm uses the quantum local search method to adaptively adjust the neighborhood value size.

[0070] Specifically, in the above technical solution, first define set B n = {0, 1, 2, ..., N} to represent a central base station microgrid CBSMG and N sub-base station microgrids SBSMG in the 5G heterogeneous BSMG system. is a collection of users of the 5G heterogeneous BSMG system. The clusters do not interfere with each other. The state of the 5G heterogeneous BSMG system changes with the traffic load and energy of the base station within the time slot length ΔT.

[0071] For the n User j∈U m , its signal-to-noise ratio (SINR) is:

[0072]

[0073] Where W i and τ i are the transmit power and channel gain of BSMGi, W s and τ s is the transmit power and channel gain of the sth interference source. Finally, is the power of the noise.

[0074] According to Shannon's formula, the average service rate available to user j on BSMGi in time slot t can be calculated as follows:

[0075] c i,j (t) = μ0(t)log2(1+SINR i,j ) (7)

[0076] Wherein, μ0(t) is the bandwidth provided by a resource block of BSMGi at time t.

[0077] Assume that a user is associated with only one BSMG during time slot t, and the current set of users associated with BSMGi is U i,j , and the maximum required rate when user j is served is r i,j(t), from which the normalized load of BSMGi in time slot t is obtained as:

[0078]

[0079] Among them, ψ i is the number of resource blocks that can be allocated to BSMGi. For simplicity, it is assumed that BSMG will not operate in an overloaded state, and the BSMG operating state is divided into active state and sleep state. CBSMG is always in the active state, and due to the high flexibility of SBSMG, SBSMG can choose to sleep. Based on the above background, the energy consumption of CBSMG and SBSMG can be calculated as:

[0080] P c =P0+ρ i (θP max +P f ) (9)

[0081]

[0082] Energy consumption of CBSMG c It includes static energy consumption and dynamic energy consumption. P0 is the fixed energy consumption in static state, while the dynamic energy consumption depends on the real-time load state of BSMG. max and P f are the maximum transmission power and circuit power of CBSMG, and θ is the power loss coefficient caused by hardware loss. In SBSMG, as the service load changes, its energy consumption remains basically unchanged, so P A and P SL To represent the energy consumption of SBSMG in active and dormant states. According to formulas (9) and (10), the total energy consumption of 5G heterogeneous BSMG system is:

[0083]

[0084] The calculation method for network energy efficiency is as follows:

[0085]

[0086] Specifically, based on the traffic data predicted by the traffic prediction model, when the SBSMG is idle or ρ i When (t) is small, the BSMG can be selectively placed in sleep mode to save grid energy or renewable energy. i (t) determines the state of SBSMG λ(t). When the small cell becomes idle (ρ i (t)=0), they switch to the dormant state and remain in the dormant state while the traffic load remains below the preset load threshold (ρi (t)<ρ th ).

[0087] The sets of active BSMG and dormant BSMG are defined respectively.

[0088] is Γ active (t) = {i∈BSMG n ∩i≠0λ(t)=1} and Γ sleep (t) = {i∈BSMG n ∩i≠0λ(t)=0}, let R i =maxr i,j (t),j∈U j It represents the maximum required rate set of users during the low-load period, and then introduces the following constraints for active and dormant SBSMGs, so that BSMGs can guarantee the user's Qos through rate indicators and do not allow any BSMG to be overloaded.

[0089]

[0090] Among them, SINR th is the threshold of the user signal-to-noise ratio. Based on the aforementioned traffic prediction algorithm, the traffic data is obtained, thereby obtaining the candidate sleep set of BSMG. Specifically, the energy consumption of a single BSMG i can be expressed as:

[0091]

[0092] And, considering the user experience, the network energy efficiency obtained for the maximum required rate defines the user QoS gain function as:

[0093]

[0094] Then, we continue to determine the dormancy and energy sharing scheme through energy sharing between BSMGs. Specifically,

[0095] When renewable energy is insufficient, BSMG will also purchase energy from the smart grid to ensure the user's Qos. For simplicity, solar energy is used as the only source of renewable energy for BSMG. At time slot t, BSMG's renewable energy generation can be expressed as g(t), and the renewable energy generation can be predicted based on historical photovoltaic power generation data and statistical methods. The predicted value is:

[0096]

[0097] Where: g p (t) is the power generation of the photovoltaic power generation system; Under standard test conditions (G STC =1000W / m2 , T STC =25℃) rated power generation; G is the light radiation intensity; ET is the ambient temperature; χ is the power coefficient, assuming that the energy consumption of the renewable energy power generation process is 0.

[0098] Each BSMG is equipped with energy storage equipment, which can not only cope with future high load demand, but also reduce the dependence on smart grid energy, thereby making full use of renewable energy. Assuming that the initial energy storage is 0, the energy storage value of BSMG during time slot t is expressed as e s (t), which is subject to the maximum energy storage capacity E s The energy sharing between BSMGs is based on the difference in energy consumption due to the change in traffic load in time and space, and the different renewable energy reserves due to geographical diversity. According to the current load and renewable energy harvesting status of BSMGi, the energy available for energy sharing can be expressed as:

[0099]

[0100] Since the actual distance between each BSMG is small, assuming that the energy sharing transmission loss can be ignored, it means that BSMGi has surplus energy to share with other BSMGs with energy shortage. * The actual energy sharing strategy between when i=i * represents the energy added by BSMGi from the smart grid, where u max is the maximum value of energy sharing and satisfies the following constraints:

[0101] 0≤u i (t)≤u max (18)

[0102] To encourage energy sharing among heterogeneous BSMGs, the cost of energy sharing among BSMGs is set to be higher than the cost of energy purchased from the smart grid. * Compared with the method of relying entirely on traditional power grid power supply, the benefits of the BSMG system after adopting cooperative sleep and energy sharing can be obtained:

[0103]

[0104] Then the BSMG dormant strategy Γ active (t) and Γ sleep (t), energy purchased from the smart grid u ii (t), energy sharing between BSMG BSMG loads are jointly optimized to minimize the cost of purchasing energy from the external smart grid based on user QoS, so as to ultimately achieve the best energy saving effect. The optimization problem can be expressed as:

[0105]

[0106] sqF i ≥log2(1+SINR th )

[0107] 0≤ρ i (t)≤1

[0108] 0≤u i (t)≤u max

[0109] e s (t)≤E s (20)

[0110] Analysis shows that this optimization problem is a typical NP-hard problem, and its objective function is to maximize the BSMG system benefit and the user's QoS benefit function. The decision variables include the user's maximum demand rate, the active and dormant states of the BSMG, the amount of energy shared, and the energy purchased from the smart grid. The constraints are the user's SINR threshold limit, the BSMG overload limit, the energy storage capacity limit, and the energy sharing and smart grid energy usage limits. Based on this, the MOEA / D algorithm (multi-objective evolutionary algorithm based on decomposition) is used to solve the BSMG collaborative dormancy and energy sharing strategy that maximizes the BSMG system benefit and the user's QoS benefit function.

[0111] Specifically, the MOEA / D algorithm first generates a set of uniform weight vectors, divides the multi-objective problem into many single-objective sub-problems, assigns corresponding weights and the neighborhood of related population points to each sub-problem, and generates new solutions for each sub-problem through crossover mutation. Finally, the Chebyshev method is used to aggregate the multi-objective problem into a single-objective problem, update the parent population, and optimize the solution of the corresponding neighborhood for each sub-problem, repeating the above steps until the termination condition is reached. The MOEA / D algorithm has good performance in dealing with continuous multi-objective problems, but its neighborhood value is fixed, which will slow down the convergence speed, and because it only uses a fixed mutation probability, it will result in weak global search capabilities. Based on this, the present application proposes a method for adaptively adjusting the size of neighborhood values ​​based on quantum local search.

[0112] First calculate the adaptive expansion coefficient:

[0113]

[0114] It max is the maximum number of iterations, γmax and γ min is a preset fixed value, and then the attraction point S is generated according to the individual historical optimal position and the group historical optimal position:

[0115]

[0116] Where φ is a random matrix that obeys uniform distribution between [0,1]. Then, based on the δ potential well, assuming that the neighborhood position vector has quantum behavior, the wave function φ is used to describe the state of the vector. By solving the Schrödinger equation of the one-dimensional δ potential well, the probability density of the vector appearing at a certain point in space is obtained:

[0117]

[0118] In the formula, is the new neighborhood position vector, L is the position vector after the individual expansion coefficient changes, and finally the new position vector equation is obtained through Monte Carlo random simulation:

[0119]

[0120] In addition, considering the information of individuals in the population, an adaptive mutation probability υ(t+1) is generated based on the number of iterations and Gaussian mutation to enhance the ability of global search, as shown in formula (25):

[0121]

[0122] in, is the rate adjustment factor, and Ga(0,1) is a standard normal distribution, which has better global search capability without reducing the convergence speed.

[0123] In the above technical solution, by constructing a 5G heterogeneous BSMG collaborative sleep and energy sharing model, the user experience is fully considered, and the network energy efficiency of the BSMG system is improved while maximizing the benefits of the BSMG system. And by improving the MOEA / D algorithm through quantum local search and adaptive mutation strategies, its convergence speed and global search capabilities can be improved, and finally the two-way flow of information and energy in the 5G heterogeneous BSMG system can be realized.

[0124] Based on the same inventive concept, the present invention also provides a traffic prediction system based on the traffic prediction method of 5G heterogeneous BSMG based on deep learning, comprising:

[0125] A data acquisition module, which is used to obtain the current flow data of each base station microgrid, wherein the base station microgrid includes a central base station microgrid and multiple sub-base station microgrids within the coverage area of ​​the central base station microgrid;

[0126] The flow prediction module is used to input the current flow data into a trained flow prediction model, and output the predicted flow data of each base station microgrid in the next time period, wherein the flow prediction model is a CNN-DAM-BiLSTM deep learning model, and the method for predicting the flow data of the next time period by the flow prediction model specifically includes:

[0127] Extract the flow features of the flow data, and use the convolutional block attention mechanism (CBAM) to perform channel feature weighting and spatial feature weighting on the flow features to obtain the CBAM-weighted flow feature representation;

[0128] The traffic feature representation weighted by CBAM is used as the input of the bidirectional long short-term memory network BiLSTM, and the multi-head attention mechanism MHAM is used to weight the output of the bidirectional long short-term memory network BiLSTM and then map and output the traffic data of the next time period.

[0129] In the above technical solution, the convolutional block attention mechanism CBAM and the multi-head attention mechanism MHAM form a dual attention mechanism to optimize the convolutional neural network CNN and the bidirectional long short-term memory network BiLSTM respectively, and establish a 5G heterogeneous BSMG traffic prediction model based on CNN-DAM-BiLSTM. While increasing the weight of important channel features, it can also learn more feature information in different spaces, ultimately improving the accuracy of traffic data prediction.

[0130] Based on the same inventive concept, the present invention also provides a collaborative sleep and energy sharing system based on the 5G heterogeneous BSMG collaborative sleep and energy sharing method, including:

[0131] A sleep module, which is used to perform a sleep operation when the base station microgrid is idle or below the sleep load threshold based on the flow data of each base station microgrid in the next time period predicted by the flow prediction method, and obtain a sleep base station microgrid set and an active base station microgrid set;

[0132] An energy sharing module is used to obtain the user's maximum demand rate, energy sharing amount, and energy purchased from the smart grid, as well as the dormant state and active state of the base station microgrid obtained in step T1, and input them into a trained 5G heterogeneous BSMG collaborative dormancy and energy sharing model, and output the dormant base station microgrid set, active base station microgrid set, and energy sharing strategy for the next time period;

[0133] Among them, the 5G heterogeneous BSMG collaborative sleep and energy sharing model takes the BSMG system benefit and the user's QoS benefit as the objective function, the user's maximum demand rate, the sleep state and active state of the base station microgrid, the energy sharing amount and the energy purchased from the smart grid as the decision variables, and the user's signal-to-noise ratio SINR threshold limit, the base station microgrid overload limit, the energy storage capacity limit, the energy sharing and the smart grid energy usage limit as the constraints. The improved MOEA / D algorithm is used to solve the energy sharing strategy, wherein the improved MOEA / D algorithm uses the quantum local search method to adaptively adjust the neighborhood value size.

[0134] In the above technical solution, the proposed 5G heterogeneous BSMG collaborative sleep and energy sharing model fully considers the user experience, maximizes the benefits of the BSMG system and improves the network energy efficiency of the BSMG system. And through quantum local search and adaptive mutation strategies, the MOEA / D algorithm is improved to improve its convergence speed and global search capabilities, and finally realizes the two-way flow of information and energy in the 5G heterogeneous BSMG system.

[0135] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present invention may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0136] The present invention also provides an electronic device, comprising:

[0137] one or more processors;

[0138] and a memory in communication connection with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the traffic prediction method or the 5G heterogeneous BSMG collaborative sleep and energy sharing method. The electronic device may be any terminal device including a mobile phone, a laptop computer, a desktop computer, a tablet computer, a PDA (Personal Digital Assistant), a POS (Point of Sales), a vehicle-mounted computer, etc.

[0139] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the traffic prediction method or the 5G heterogeneous BSMG collaborative sleep and energy sharing method is implemented.

[0140] Through the description of the above implementation mode, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present invention, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0141] <Example 1>

[0142] like Figure 1 As shown, a 5G heterogeneous BSMG cooperative sleep and energy sharing method based on deep learning and improved MOEA / D algorithm includes the following steps:

[0143] Step 1: Build a 5G heterogeneous BSMG system consisting of a central base station microgrid and sub-base station microgrids as a simulation scenario, and realize the bidirectional flow of information and energy between each BSMG through a centralized control unit;

[0144] Step 2: Build a deep learning model based on CNN-DAM-BiLSTM, and use historical traffic data to train the deep learning model to obtain a trained deep learning model, namely the traffic prediction model of CNN-CBAM-BiLSTM;

[0145] Then, the traffic data of each BSMG in the current 5G heterogeneous BSMG system is input into the trained deep learning model, and the predicted traffic data of each BSMG in the future is output;

[0146] Step 3: According to the sleep load threshold ρ th and predicted traffic data of each BSMG in the future, and perform sleep operation when the BSMG is in no-load or low-load state;

[0147] Step 4: Construct two objective functions: BSMG system benefit and user QoS benefit. Take the user's maximum demand rate, the active and dormant status of BSMG, the energy sharing amount and the energy purchased from the smart grid as decision variables. Take the user's SINR threshold limit, BSMG overload limit, energy storage capacity limit, energy sharing and smart grid energy usage limit as constraints. Solve them through the improved MOEA / D algorithm and finally realize energy sharing of 5G heterogeneous BSMG system.

[0148] The rule of energy flow is to prioritize energy sharing between BSMGs, with BSMGs with surplus energy delivering energy to BSMGs with energy shortages. If each BSMG is unable to meet the needs of a BSMG with energy shortages, the external power grid will provide energy for it.

[0149] <Example 2>

[0150] like Figure 2 and 3 As shown, building and comparing the deep learning model based on CNN-DAM-BiLSTM mentioned in Example 1 includes the following steps:

[0151] Step 1: Obtain the data set required for training the deep learning model. The data set contains the cellular traffic values ​​and base station coordinate data recorded at 1-hour intervals in a city in China.

[0152] Step 2: Build four different traffic prediction models, including CNN-BiLSTM, CNN-BiGRU, CNN-SA-BiLSTM and the CNN-DAM-BiLSTM model of this application, and use the data of the data set in step 1 to train the four traffic prediction models respectively, and output their respective prediction results; where BiGRU represents a bidirectional gated recurrent unit (Bidirectional Gated Recurrent Unit), and SA-BiLSTM represents a self-attention bidirectional long short-term memory network (Self-Attention Bidirectional Long Short-Term Memory);

[0153] Step 3: Compare the predicted values ​​of traffic data from the four traffic prediction models with the actual values.

[0154] The results show that CNN-DAM-BiLSTM has the best prediction performance. Therefore, when performing BSMG sleep operation, based on the comparison results, the CNN-DAM-BiLSTM model is selected for traffic prediction.

[0155] <Example 3>

[0156] The improved MOEA / D algorithm proposed in the present invention is simulated in simulation software, and its performance is compared with multi-objective optimization algorithms such as MOSSA, NSGA-Ⅱ, and MOPSO. Figure 5 The figures show the comparison of the average system benefit of the improved MOEA / D algorithm and the other three algorithms in the optimization problem involved in Example 1 when the number of iterations is set to 400. It can be seen from the figure that the improved MOEA / D algorithm converges faster than other algorithms to obtain the optimal solution, and has a stronger ability to jump out of the local optimum after introducing the mutation operation. The optimal solution obtained is closer to the theoretical optimum than other algorithms.

[0157] like Figure 6 , 7 As shown, the cooperative sleep and energy sharing method (IMOEA / D-CSESA) proposed in Example 1 is compared with energy-saving algorithms in other documents, wherein the energy-saving algorithms in different comparative documents are introduced as follows:

[0158] Comparative Algorithm 1: Randomly select active SBSMGs in the network to sleep based on the distance between CBSMGs;

[0159] Comparative Algorithm 2: Without considering user QoS, base station hibernation is performed based on traffic prediction results of deep learning.

[0160] Comparative Algorithm 3: Solve the optimization problem involved in Example 1 by using reinforcement learning methods such as Q learning;

[0161] Comparative Algorithm 4: Solve the BSMG collaborative sleep and energy sharing problem involved in Example 1 through dynamic evolutionary game.

[0162] Figure 6 , 7 The strategies solved by the above five algorithms are compared in terms of system benefits and network energy efficiency within one day. It can be seen that the IMOEA / D-CSESA strategy proposed in the present invention has excellent performance in both aspects, which can effectively reduce the energy consumption of cellular networks and ensure user QoS.

[0163] The significance of the present invention lies in that it can not only guarantee the communication quality of users in the system to a large extent, but also adapt to the current explosive growth of user traffic, and at the same time improve the energy efficiency of mobile cellular networks. It plays an important role in realizing the two-way flow of BSMG information and energy, reducing carbon emissions, achieving sustainable development, and building a low-carbon and intelligent cellular network in the future.

[0164] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A traffic prediction method for 5G heterogeneous BSMG based on deep learning, characterized in that: The following steps are involved: S1. Obtain current flow data of each base station microgrid, wherein the base station microgrid includes a central base station microgrid and a plurality of sub-base station microgrids within the coverage of the central base station microgrid; S2. Input the current flow data into a trained flow prediction model, and output the predicted flow data of each base station microgrid in the next time period, wherein the flow prediction model is a CNN-DAM-BiLSTM deep learning model, and the method for predicting the flow data of the next time period by the flow prediction model specifically includes: Extract the flow features of the flow data, and use the convolutional block attention mechanism (CBAM) to perform channel feature weighting and spatial feature weighting on the flow features to obtain the CBAM-weighted flow feature representation; The traffic feature representation weighted by CBAM is used as the input of the bidirectional long short-term memory network BiLSTM, and the multi-head attention mechanism MHAM is used to weight the output of the bidirectional long short-term memory network BiLSTM and then map and output the traffic data of the next time period.

2. The traffic prediction method for 5G heterogeneous BSMG based on deep learning as claimed in claim 1, characterized in that: The convolutional block attention mechanism CBAM is specifically as follows: first, a global average pooling operation of the channel attention features is performed on the flow features extracted by the convolutional layer to obtain the channel attention weights, and then a maximum pooling operation of the spatial attention features is performed to obtain the spatial attention weights. Finally, the Sigmoid activation function and feature multiplication are used to obtain the spatial attention weight vector of each channel and the weighted feature representation, that is, the CBAM weighted flow feature representation.

3. A 5G heterogeneous BSMG collaborative sleep and energy sharing method based on an improved MOEA / D algorithm, characterized in that: The following steps are involved: T1. Based on the flow data of each base station microgrid in the next time period predicted by the flow prediction method according to claim 1, according to the preset sleep load threshold, perform a sleep operation when the base station microgrid is in no-load or below the sleep load threshold, and obtain a sleep base station microgrid set and an active base station microgrid set; T2, obtain the user's maximum demand rate, energy sharing amount and energy purchased from the smart grid, as well as the dormant state and active state of the base station microgrid obtained in step T1, and input them into a trained 5G heterogeneous BSMG collaborative dormancy and energy sharing model, and output the dormant base station microgrid set, active base station microgrid set and energy sharing strategy for the next time period; Among them, the 5G heterogeneous BSMG collaborative sleep and energy sharing model takes the BSMG system benefit and the user's QoS benefit as the objective function, the user's maximum demand rate, the sleep state and active state of the base station microgrid, the energy sharing amount and the energy purchased from the smart grid as the decision variables, and the user's signal-to-noise ratio SINR threshold limit, the base station microgrid overload limit, the energy storage capacity limit, the energy sharing and the smart grid energy usage limit as the constraints. The improved MOEA / D algorithm is used to solve the energy sharing strategy, wherein the improved MOEA / D algorithm uses the quantum local search method to adaptively adjust the neighborhood value size.

4. The 5G heterogeneous BSMG cooperative sleep and energy sharing method based on the improved MOEA / D algorithm as claimed in claim 3 is characterized in that: When sharing energy, the rule of energy flow is to prioritize green energy sharing between base station microgrids, and the base station microgrid with surplus green energy will transmit green energy to the base station microgrid with energy shortage; When each base station microgrid cannot meet the energy demand of the base station microgrid which is in an energy shortage state, energy is purchased from the smart grid to provide it with energy.

5. The 5G heterogeneous BSMG cooperative sleep and energy sharing method based on the improved MOEA / D algorithm as claimed in claim 4, characterized in that: The green energy is the electric energy provided by the photovoltaic power generation system.

6. A traffic prediction system based on the traffic prediction method of 5G heterogeneous BSMG based on deep learning according to claim 1 or 2, characterized in that: include: A data acquisition module, which is used to obtain the current flow data of each base station microgrid, wherein the base station microgrid includes a central base station microgrid and multiple sub-base station microgrids within the coverage area of ​​the central base station microgrid; The flow prediction module is used to input the current flow data into a trained flow prediction model, and output the predicted flow data of each base station microgrid in the next time period, wherein the flow prediction model is a CNN-DAM-BiLSTM deep learning model, and the method for predicting the flow data of the next time period by the flow prediction model specifically includes: Extract the flow features of the flow data, and use the convolutional block attention mechanism (CBAM) to perform channel feature weighting and spatial feature weighting on the flow features to obtain the CBAM-weighted flow feature representation; The traffic feature representation weighted by CBAM is used as the input of the bidirectional long short-term memory network BiLSTM, and the multi-head attention mechanism MHAM is used to weight the output of the bidirectional long short-term memory network BiLSTM and then map and output the traffic data of the next time period.

7. A collaborative sleep and energy sharing system based on the 5G heterogeneous BSMG collaborative sleep and energy sharing method according to any one of claims 3 to 5, characterized in that: include: A sleep module, which is used to perform a sleep operation when the base station microgrid is idle or below the sleep load threshold based on the flow data of each base station microgrid in the next time period predicted by the flow prediction method, and obtain a sleep base station microgrid set and an active base station microgrid set; The energy sharing module is used to obtain the user's maximum demand rate, energy sharing amount and energy purchased from the smart grid, as well as the dormant state and active state of the base station microgrid obtained in step T1, and input them into a trained 5G heterogeneous BSMG collaborative dormancy and energy sharing model, and output the dormant base station microgrid set, active base station microgrid set and energy sharing strategy for the next time period; Among them, the 5G heterogeneous BSMG collaborative sleep and energy sharing model takes the BSMG system benefit and the user's QoS benefit as the objective function, the user's maximum demand rate, the sleep state and active state of the base station microgrid, the energy sharing amount and the energy purchased from the smart grid as the decision variables, and the user's signal-to-noise ratio SINR threshold limit, the base station microgrid overload limit, the energy storage capacity limit, the energy sharing and the smart grid energy usage limit as the constraints. The improved MOEA / D algorithm is used to solve the energy sharing strategy, wherein the improved MOEA / D algorithm uses the quantum local search method to adaptively adjust the neighborhood value size.

8. The cooperative sleep and energy sharing system according to claim 7, characterized in that: Each central base station microgrid is equipped with a corresponding photovoltaic power generation system and a smart grid power supply system, and each sub-base station microgrid is equipped with a corresponding photovoltaic power generation system.

9. An electronic device, characterized in that include: one or more processors; And a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the traffic prediction method described in claim 1 or 2 or the 5G heterogeneous BSMG collaborative sleep and energy sharing method described in any one of claims 3 to 5.

10. A computer-readable storage medium, characterized in that A computer program is stored on the storage medium. When the program is executed by the processor, it implements the traffic prediction method described in claim 1 or 2 or the 5G heterogeneous BSMG collaborative sleep and energy sharing method described in any one of claims 3 to 5.

Images