Adaptive base station energy consumption optimization method based on cross-operator cooperation

Abstract

The invention discloses a self-adaptive base station energy consumption optimization method based on cross-operator cooperation, which relates to the field of 5G network processing, and comprises the following steps: constructing a base station energy consumption model, and confirming an optimization objective of an optimization problem; in the COPE framework, user distribution is carried out based on the user distribution strategy of each base station by adopting a combined algorithm of itransform and Dlinear, and user distribution prediction is carried out; in the COPE framework, dynamically adjusting user distribution based on a predicted user distribution result by adopting an asynchronous dominant participant-commentator algorithm; in the COPE framework, a double-delay depth deterministic policy gradient algorithm is adopted, the base station energy use condition of each operator is evaluated, and an optimization target is optimized based on deep reinforcement learning; a cross-operator cooperation energy consumption optimization mechanism is adopted, a win-win cooperation framework is established, cooperation of telecom operators in a competitive environment is promoted, and the energy efficiency of the 5G infrastructure is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of 5G network processing, and in particular relates to an adaptive base station energy consumption optimization method based on cross-operator cooperation. Background Art

[0002] The development of 5G networks has brought faster network speeds, greater capacity, and lower latency, providing users with a better network experience. This has brought a network experience with higher speeds and lower latency for users, but it has also brought huge cost pressures to operators. Against the backdrop of increasingly fierce competition among operators, the same area will be repeatedly built with 5G base stations by each operator to ensure its own network services, which has doubled the energy consumption of 5G networks. In order to reduce operating costs and improve profitability, operators have begun to seek cooperation with enterprises.

[0003] However, operators are in a competitive relationship with each other, and it is difficult to generate cooperation without an interest drive. In a competitive environment involving resource sharing or cost sharing, each operator pays more attention to maximizing its own interests and is not willing to cooperate with competitors.

[0004] Secondly, the coverage areas of base stations under the jurisdiction of different operators are highly coupled, and the utilization rate of base stations is low. In order to provide services to their own users, the coverage areas of base stations of different operators highly overlap, resulting in multiple base stations in the same area, but with low utilization rates. The user distributions of different operators are uneven. Due to reasons such as the cooperation between operators and companies, there is a problem of highly concentrated users of a certain operator in some areas, resulting in a reduction in network service quality. The concentration of content requests will also lead to an increase in the load of base stations, and thus an increase in the energy consumption of base stations, which may lead to environmental problems, increased carbon emissions, and energy waste.

[0005] In addition, different operators transmit the same content to the same area, and data transmission is highly coupled. Due to the competitive relationship between operators and the lack of a cooperation mechanism, each operator deploys similar content transmission networks in the same area, resulting in duplication and redundancy in the data transmission process. Such a situation makes the network congested, wastes bandwidth, increases data transmission latency, and reduces the user experience.

[0006] Therefore, an adaptive base station energy consumption optimization method based on cross-operator cooperation is provided to solve the above problems. Summary of the Invention

[0007] The object of the present invention is to provide an adaptive base station energy consumption optimization method based on cross-operator cooperation, which adopts a cross-operator cooperation energy consumption optimization mechanism, establishes a win-win cooperation framework, promotes the cooperation of telecom operators in a competitive environment, effectively improves the energy efficiency of 5G infrastructure, and solves the key obstacles in the existing 5G network computing and processing.

[0008] To achieve the above object, the present invention provides an adaptive base station energy consumption optimization method based on cross-operator cooperation, including the following steps:

[0009] S1: Construct a base station energy consumption model and confirm the optimization objective of the optimization problem;

[0010] S2: In the COPE framework, adopt a combined algorithm of itransformer and Dlinear, perform user allocation based on the user allocation strategy of each base station, and predict the user distribution;

[0011] S3: In the COPE framework, adopt the asynchronous advantage actor-critic algorithm to dynamically adjust the user allocation based on the predicted user distribution results;

[0012] S4: In the COPE framework, adopt the twin-delayed deep deterministic policy gradient algorithm to evaluate the base station energy usage of each operator, optimize the optimization objective based on deep reinforcement learning, and decide to lease or shut down the base station.

[0013] Preferably, step S1 specifically includes the following steps:

[0014] S11: Construct a network model, set the O operators providing network services within the 5G network coverage area as O = {o n |n = 1,…,O}, and set the B 5G base stations constituting the operator's 5G cellular network as Set the U users within the network wireless coverage area as U n ={u n,j |j = 1,…,|U n |};

[0015] S12: Construct a base station load model, and set the base station load of each base station b n,i as the sum of the network request sizes of all users u assigned to the corresponding base station b n,i The base station load n,j is specifically expressed as:

[0016]

[0017]

[0018] where a n,i,j is a binary parameter representing the user association strategy between user u n,j and base station b n,i ;

[0018] S13: Construct a transmission rate model, and calculate the transmission rate of user u at base station b n,j based on the channel n,iTotal throughput in Total throughput Specifically expressed as:

[0019]

[0020] Among them, Represents the transmit power of base station b n,i The transmit power of Represents the transmit power of base station b n,l The transmit power of Represents user u n,j And base station b n,i On the channel Channel gain on Represents user u n,j And base station b n,l On the channel Channel gain on, N represents the noise power spectral density, Represents the channel Bandwidth of;

[0021] S14: Build a transmission time model and calculate the user request time Base station b n,i Channel working time on Base station b n,i In a time slot t m Total working time t within w (b n,i ) and sleep time t s (b n,i );

[0022] S15: Build an energy consumption model and set the total base station energy consumption En n To the sum of the energies consumed by all base stations b n,i During the active period, the total base station energy consumption En n Specifically expressed as:

[0023] En(b n,i ) = p w (b n,i )·(t a (b n,i ) + t r (b n,i )) + p s (b n,i )·t s (b n , i )

[0024]

[0025] Among them, En(b n,i) represents a base station b n,i The sum of the energy consumed during the active period, p w (b n,i ) represents the base station b n,i The power in the working state, p s (b n,i ) represents the base station b n,i The power in the sleep state, t r (b n,i ) represents the request response time;

[0026] S16: Construct a cost model and calculate the final cost Cost(n) of the operator O n ;

[0027] S17: Construct a cache policy model and set the cache management goal to maximize the cache hit rate CHR. Maximizing the cache hit rate CHR is specifically expressed as:

[0028]

[0029] Among them, D pre represents a set of data types requested, represents the user u n,j The request count within the time t;

[0030] S18: Set the optimization goal of the optimization problem to:

[0031]

[0032] Among them, cus represents the cache update policy, represents the energy consumption of the base station b n,i ; represents the user u n,j Served by the base station b n,i .

[0033] Preferably, in step S11, each base station b n,i is provided with C channels and an edge server with a cache capacity e i . The C channels are specifically set as Each channel The bandwidth is set to The edge server is specifically set as E = {e i | i = 1,..., |B|}, and the time interval of the base station Bn is set as T = {t m | m = 1,..., |T|}.

[0034] Preferably, in step S11, each user u n,j Within the specified time interval t mSet the size of the network requests randomly generated within as For each user u n,j Within the specified time interval t m Set the geographical coordinates of the network requests randomly generated within as

[0035] Preferably, step S14 specifically includes the following steps:

[0036] Step 1: According to the network transmission speed of user u n,j Set the user request time as:

[0037]

[0038] where represents the time required for the edge server to respond to user u n,i at base station b n,j t cloud represents the total time required for the cloud server to respond to the user request. The total time includes the communication time between the base station and the cloud server and the processing time of the cloud server;

[0039] Step 2: Based on each channel Set the channel working time n,i on base station b as the sum of the completion times of all requests on this channel. The channel working time n,i on base station b is specifically expressed as:

[0040]

[0041] Step 3: Set the total working time t n,i of base station b m within a time slot t w (b n,i ) as:

[0042]

[0043] where t a (b n,i ) represents the activation time of base station b n,i , represents the longest channel working time of base station b n,i ;

[0044] Step 4: Set the sleep time t n,i of base station b m within a time slot t s (b n,i ) as:

[0045] t s (b n,i ) = t m -t w (b n,i )。

[0046] Preferably, step S16 specifically includes the following steps:

[0047] Step 1: Set the total base station operation cost Cost n (n) of operator O to: El where En represents the total power consumption, q

[0048]

[0049] represents the upper limit of power consumption of the s-th layer, and C s represents the unit price corresponding to the s-th layer; s Step 2: Determine the additional cost generated during the rental or lease process according to the highest-tier price of the tiered fee. The rental price F

[0050] (n) and the lease-in price F out (n) are respectively expressed as: in F

[0051] (n) = Cost′ out (n) - Cost El (n) + ω El F

[0052] (n) = Cost in (n) - Cost′ El (n) + ω El where Cost′

[0053] (n) represents the electricity cost generated by operator O El after the lease ends, and ω represents the basic rent of the lease; n Step 3: Calculate the final cost Cost(n) of operator O

[0054] The final cost Cost(n) of operator O n is specifically expressed as: n Cost(n) = Cost

[0055] (n) - F El (n) + F in (n) + F out (n).

[0056] Preferably, in step S2, the input of the COPE framework is the network request data of user u n,j and base station b n,iThe load information and system parameters, where the system parameters include energy cost and latency threshold, and the output of the COPE framework is the user association policy a n,j between the user u n,i and the base station b n,i,j .

[0057] Preferably, in step S2, user distribution prediction is performed, and the current state of the time interval t m is set to the current state includes the historical locations and historical requests of each user u n,j within the defined time window, and the current state is specifically expressed as:

[0058]

[0059] where p represents the length of the historical window, represents the predicted location and predicted request of the user u n,j at the current time interval t m , represents the historical spatial coordinates of the user u n,j , represents the network request of the user u n,j .

[0060] Preferably, step S3 specifically includes the following steps:

[0061] S31: Set the state space of the base station b n,i to S i , and the state space S n,i of the base station b i includes the current base station load and the network requests within the coverage area of the base station b n,i , The state space S n,i of the base station b i is specifically expressed as:

[0062]

[0063] S32: Set the action space of the base station b n,i to A i , and the action space A n,i of the base station b i includes a set of user allocation operations, and the action space A n,i of the base station b i is specifically expressed as:

[0064]

[0065] S33: Through the reward function Ri Minimize base station b n,i energy consumption and latency of the allocation strategy, reward function R i Specifically expressed as:

[0066] R i = -α·En(b n,i ) - β·t bi

[0067] where t bi represents the service latency of base station b n,i , and both α and β represent adjustable parameters;

[0068] S34: Based on the asynchronous advantage actor-critic algorithm, obtain the optimal allocation strategy

[0069] Preferably, step S4 specifically includes the following steps:

[0070] S41: Set the state space of operator O n as S o , and the state space S n of operator O o includes the base station load under management and the base station energy state. The state space S n of operator O o is specifically expressed as:

[0071]

[0072] S42: Set the action space of operator O n as A o , and the action space A n of operator O o includes the available lease set and the deactivation decision set. The action space A n of operator O o is specifically expressed as:

[0073]

[0074] S43: Through the reward function R o motivate the cost and energy efficiency of operator O n , specifically expressed as:

[0075] R o = -Cost(n).

[0076] Therefore, the present invention adopts the above-mentioned adaptive base station energy consumption optimization method based on cross-operator cooperation, and has the following beneficial effects:

[0077] (1) The present invention adopts a cross-operator cooperative energy consumption optimization mechanism to establish a win-win COPE cooperation framework, which promotes the cooperation of telecom operators in a competitive environment, effectively improves the energy efficiency of 5G infrastructure, and solves the key obstacles in the existing 5G network computing and processing;

[0078] (2) The present invention proposes a unified prediction method, which improves the accuracy and responsiveness of user request prediction and realizes the active and effective management of network resources;

[0079] (3) The present invention introduces a collaborative energy optimization strategy, which refines user allocation through multi-angle coordination and improves network efficiency by balancing the operation perspective and load allocation.

[0080] The method solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0081] Figure 1 It is a flowchart of an adaptive base station energy consumption optimization method based on cross-operator cooperation of the present invention;

[0082] Figure 2 It is the motion trajectory prediction result of a single user in an embodiment of the present invention;

[0083] Figure 3 It is the total energy consumption of base stations running on the simulated data set and Geolife in an embodiment of the present invention;

[0084] Figure 4 It is the total delay of base stations running on the simulated data set and Geolife in an embodiment of the present invention;

[0085] Figure 5 It is the total cost of base stations running on the simulated data set and Geolife in an embodiment of the present invention. Detailed Embodiments

[0086] The method solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0087] Unless otherwise defined, the method terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the field to which the present invention belongs.

[0088] In the present invention, words such as "including" or "comprising" mean that the elements before this word cover the elements listed after this word, and do not exclude the possibility of also covering other elements. The orientation or positional relationship indicated by terms such as "inside", "outside", "above", "below", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation to the present invention. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. In the present invention, unless otherwise clearly specified and defined, terms such as "attached" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be directly connected, or indirectly connected through an intermediate medium. It can be the communication inside two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0089] Embodiment

[0090] As Figure 1 shown, the present invention provides an adaptive base station energy consumption optimization method based on cross-operator cooperation, including the following steps:

[0091] S1: Construct a base station energy consumption model and confirm the optimization goal of the optimization problem;

[0092] Step S1 specifically includes the following steps:

[0093] S11: Construct a network model, set O operators providing network services within the 5G network coverage area as O = {o n |n = 1,..., O}, and set B 5G base stations constituting the 5G cellular network of the operator as Set U users within the network wireless coverage area as U n = {u n,j |j = 1,..., |U n |}, and these users randomly generate network requests within a specified time interval;

[0094] In step S11, each base station b n,i is provided with C channels and an edge server with a cache capacity of e i The C channels are specifically set as Each channel The bandwidth is set as The edge server is specifically set as E = {e i |i = 1,..., |B|}, and the time interval of base station Bn is set as T = {t m|m=1,...,|T|}.

[0095] In step S11, each user u n,j At a specified time interval t m The size of the randomly generated network request is set to Each user u n,j At a specified time interval t m The randomly generated geo-coordinates of the network request are set to

[0096] S12: construct a base station load model, in which users located within the coverage area of ​​the base station are able to request data, and there is a one-to-many correlation between the base station and its users;

[0097] Each base station b n,i Base station load Set to be assigned to the corresponding base station b n,i All users of u n,j The size of the network request The sum of the base station load Specifically expressed as:

[0098]

[0099] Among them, a n,i,j It is a binary parameter, indicating that user u n,j With base station b n,i User association strategy between them;

[0100] S13: Constructing a transmission rate model based on the channel Calculate user u n,j At base station b n,i The total throughput in Total throughput Specifically expressed as:

[0101]

[0102] in, Indicates base station b n,i The transmission power, Indicates base station b n,l The transmission power, Represents user u n,j and base station b n,i In the channel The channel gain on Represents user u n,j and base station b n,l In the channel The channel gain on the channel, N represents the noise power spectral density, Indicates the channel bandwidth;

[0103] S14: Construct a transmission time model and calculate the user request time Base station b n,i channel working time on Base station b n,i in a time slot t m total working time t w (b n,i ) and sleep time t s (b n,i );

[0104] Step S14 specifically includes the following steps:

[0105] Step 1: According to the network transmission speed of user u n,j set the user request time as:

[0106]

[0107] where represents the time required for the edge server to respond to user u n,i at base station b n,j t cloud represents the total time required for the cloud server to respond to the user request. The total time includes the communication time between the base station and the cloud server and the processing time of the cloud server;

[0108] Step 2: All user requests on base station b n,i are randomly assigned to the channel for transmission. The transmission time can be considered as the sum of the transmission times of all user requests in the channel ;

[0109] Based on each channel set the channel working time n,i on base station b as the sum of the completion times of all requests on this channel. The channel working time n,i on base station b is specifically expressed as:

[0110]

[0111] Step 3: Before all data transmission requests are completed, base station b n,i remains in the working state, and then it will switch to the predetermined sleep state until the next time slot;

[0112] Set the total working time t n,i of base station b m in a time slot tw (b n,i ) is set to:

[0113]

[0114] where, t a (b n,i ) represents the activation time of base station b n,i , and represents the longest channel working time of base station b n,i ;

[0115] Step 4: Set the sleep time t n,i of base station b m within a time slot t s (b n,i ) is set to:

[0116] t s (b n,i ) = t m - t w (b n,i ).

[0117] S15: The energy consumption of 5G base stations is affected by various factors. One important factor is the load. At different loads, the working power of 5G base stations is also different, which means that the energy consumption will increase with the increase of the load. Therefore, when optimizing the energy consumption of 5G base stations, it is necessary to consider the load management strategy to ensure that the energy consumption can be effectively controlled even at high loads. 5G base stations have flexible energy management strategies. They can not only switch between the working and idle states, but also enter the sleep state to save energy during the inactive period. Compared with the traditional power-off method, entering the sleep mode can enable the base station to quickly wake up and enter the working state when needed, thus reducing the situation of network service quality degradation caused by the offline of the base station.

[0118] Build an energy consumption model. The energy consumption of the base station within a time slot can be divided into the energy consumption during the working period and the energy consumption during the sleep period. The time slots within the service time range of the base station in this area can be divided into the working time and the sleep time, and the power of various states also changes;

[0119] Set the total energy consumption En n of the base station as the sum of the energies consumed by all base stations b n,i during the active period. The total energy consumption En n of the base station is specifically expressed as:

[0120]

[0121] where, En(b n,i ) represents the sum of the energies consumed by a base station b n,i during the active period, pw (b n,i ) represents the power of base station b n,i in the working state, p s (b n,i ) represents the power of base station b n,i in the sleep state, t r (b n,i ) represents the request response time;

[0122] In the modeling of this embodiment, other energy consumption factors except transmission energy consumption are ignored. By focusing on the optimization of transmission energy consumption, the maximum energy consumption reduction benefit is obtained while keeping other factors unchanged;

[0123] S16: Build a cost model and calculate the final cost Cost(n) of operator O n ;

[0124] Step S16 specifically includes the following steps:

[0125] Step 1: The power consumption of base stations usually follows an industrial electricity pricing model, which is calculated based on hierarchical pricing. Set the total operating cost Cost n (n) of operator O El as:

[0126]

[0127] where En represents the total power consumption, q s represents the upper limit of power consumption in the s-th layer, and C s represents the unit price corresponding to the s-th layer;

[0128] Step 2: In this embodiment, the method to reduce the overall energy consumption of the operator is to achieve the optimal overall energy consumption through the lease and rental of base stations. That is, the lease behavior is only a necessary step for the operator to obtain benefits, rather than the ultimate goal. Therefore, the rent generated during the lease process only needs to be lower than the cost of operating the base station. Similarly, the income generated during the rental process only needs to be higher than the cost of serving additional users;

[0129] Determine the additional fees generated during the rental or lease process according to the highest-tier price of the ladder fees. The rental price F out (n) and the lease price F in (n) are respectively expressed as:

[0130] F out (n) = Cost′ El (n) - Cost El (n) + ω

[0131] F in (n) = CostEl (n)-Cost′ El (n)+ω

[0132] Among them, Cost′ El (n) represents the electricity cost incurred by operator O n after the lease ends, and ω represents the basic rent of the lease;

[0133] Step 3: Calculate the final cost Cost(n) of operator O n , and the final cost Cost(n) of operator O n is specifically expressed as:

[0134] Cost(n) = Cost El (n)-F in (n)+F out (n).

[0135] S17: Construct a caching policy model. Considering the limited storage capacity of edge caching, how to effectively utilize this caching to reduce the time required for the base station to satisfy user requests becomes a key aspect of the entire problem. The intelligently designed caching policy effectively utilizes the caching resources of the base station, thereby significantly improving energy efficiency and reducing latency;

[0136] First, each base station maintains a historical record of user requests in discrete time slots, capturing the data frequency and type of user requests. These historical data can be used as the basis for predicting future user requests. Under the guidance of historical user request patterns and prediction modeling, caching update decisions are made regularly.

[0137] Set the caching management objective to maximize the caching hit rate CHR. Maximizing the caching hit rate CHR is specifically expressed as:

[0138]

[0139] Among them, D pre represents a set of data types requested, represents user u n,j 's request count within time t;

[0140] The overall objective is to maximize the caching hit rate CHR while complying with the caching capacity constraint C cache , and the higher the caching hit rate, the higher the utilization efficiency of the caching resources, reducing the need to transmit data from the core network and improving the overall energy efficiency and user experience;

[0141] S18: Our goal is to optimize the energy efficiency of the global 5G network base stations on the premise of ensuring that the network services for all users are not affected, while ensuring the lowest network operation cost for each operator. Since operators will not accept loss-making transactions, it is also necessary to ensure that the revenue of all operators is positive. The optimization objective of the optimization problem is set as follows:

[0142]

[0143] Among them, cus represents the cache update policy, represents the energy consumption of base station b n,i s.t. represents user u n,j is served by base station b n,i ;

[0144] This optimization problem is subject to constraints such as the latency requirements of individual users, cache capacity limitations, and compliance with predefined scheduling of base station activities. The optimization objective is to find the most effective policies and strategies while meeting these limitations, and to obtain the optimal solution that the algorithm can provide when the load of the base station, cache content, and user requirements are in different states.

[0145] The COPE algorithm runs through three main stages: user behavior prediction, user allocation, and cross-operator collaboration to achieve energy efficiency and cost optimization. The algorithm selection in each stage addresses the specific challenges of user behavior prediction, resource allocation, and cooperative leasing in a complex multi-operator environment. This embodiment jointly minimizes the energy consumption and operation cost of the base stations in the entire 5G network;

[0146] S2: In the COPE framework, a combined algorithm of itransformer and Dlinear is adopted to perform user allocation based on the user allocation strategy of each base station and conduct user distribution prediction. Itransformer captures the time dependence of user mobility, while Dlinear enhances demand prediction by considering fluctuations. By integrating the two models, the prediction accuracy is improved to provide a more accurate user behavior prediction for proactive network resource allocation;

[0147] In step S2, the COPE algorithm utilizes the comprehensive advantages of time series prediction and two DRLs in three stages to predict user requests in sequence, optimize user allocation, and promote cross-operator collaboration. By defining clear mathematical representations for the state space, action space, and reward function in each stage, the algorithm establishes a framework that can adapt dynamically and in real time. The input of the COPE framework is the network request data of user u n,j and the load information of base station b n,i , and the system parameters include energy cost and latency threshold. The output of the COPE framework is user u n,j and base station bn,i User association strategy between n,i,j .

[0148] In step S2, user distribution prediction is performed, and the time interval t m The current state is set to Current Status Including each user u n,j Historical positions and historical requests within a defined time window, current status Specifically expressed as:

[0149]

[0150] Where p represents the length of the history window, Represents user u n,j At the current time interval t m The predicted location and prediction request, Represents user u n,j The historical spatial coordinates of Represents user u n,j network request.

[0151] These predictions provide the necessary real-time input for user allocation decisions, ensuring accurate and demand-consistent allocation of network resources.

[0152] S3: In the COPE framework, an asynchronous advantage actor-critic algorithm is used to dynamically adjust user allocation based on the predicted user distribution results;

[0153] Step S3 specifically includes the following steps:

[0154] S31: Base station b n,i The state space is set to S i , base station b n,i The state space S i Including current base station load and base station b n,i Network requests within the coverage area Base station b n,i The state space S i Specifically expressed as:

[0155]

[0156] S32: Base station b n,i The action space is set to A i , base station b n,i The action space A i Including a set of user allocation operations, base station b n,i The action space A i Specifically expressed as:

[0157]

[0158] S33: Minimize the energy consumption of base station b i and the latency of the allocation strategy through the reward function R n,i The reward function R i is specifically expressed as:

[0159] R i = -α·En(b n,i ) - β·t bi

[0160] where t bi represents the service latency of base station b n,i and both α and β represent adjustable parameters;

[0161] By improving the reward structure and considering both energy consumption and service quality, the asynchronous advantage actor-critic algorithm is enhanced, enabling the algorithm model to better balance energy efficiency and user satisfaction in large-scale 5G networks;

[0162] S34: Based on the asynchronous advantage actor-critic algorithm, obtain the optimal allocation strategy This strategy can minimize the load imbalance between base stations while optimizing energy efficiency.

[0163] S4: In the COPE framework, adopt the twin-delayed deep deterministic policy gradient algorithm to evaluate the base station energy usage of each operator, optimize the optimization objective based on deep reinforcement learning, and decide to lease or shut down base stations.

[0164] Step S4 specifically includes the following steps:

[0165] S41: Set the state space of operator O n as S o The state space S n of operator O o includes the base station load under management and the base station energy status. The state space S n of operator O o is specifically expressed as:

[0166]

[0167] S42: Set the action space of operator O n as A o The action space A n of operator O o includes the available lease set and the deactivation decision set. The action space A n of operator Oo Specifically expressed as:

[0168]

[0169] S43: Through the reward function R o Motivate the operator O n 's cost and energy efficiency, specifically expressed as:

[0170] R o = -Cost(n).

[0171] Improve the Twin Delayed Deep Deterministic Policy Gradient algorithm by modifying its reward function, while optimizing energy conservation and cost reduction to stabilize the decision-making process, improve the leasing strategy, balance the load distribution, and maximize the operator's energy efficiency.

[0172] This embodiment conducts experiments using simulation datasets and the Geolife dataset respectively;

[0173] As Figure 2 shown, run the user location prediction method on the Geolife dataset, predict the user's upcoming next location based on the user's action data from the previous day, and count the prediction results. When the prediction accuracy is one percent, the accuracy rate is 96.701%;

[0174] As Figure 3 shown, conduct energy conservation experiments on the simulation dataset and the Geolife dataset respectively. Compared with the greedy method used in most regions currently, the method of this embodiment greatly increases the energy efficiency. To prove the enhancing effect of prediction and leasing on energy efficiency, this embodiment tests the results without user action prediction and without leasing behavior respectively. The experimental results show that both of these actions enhance the energy efficiency of the method. As Figure 4 shown, the leasing behavior will not have a negative impact on the network service quality, and there is still a slight improvement compared with the situation without leasing behavior;

[0175] As Figure 5 shown, analyze the cost of each operator's network operation for one day before and after the leasing behavior. Compared with the situation without leasing behavior, each operator has a cost savings of 8% to 10%. Generally speaking, on the premise of the win-win situation of all operators, the overall network operation cost drops by about 9%.

[0176] Therefore, the present invention adopts the above-mentioned adaptive base station energy consumption optimization method based on cross-operator cooperation, establishes a win-win COPE cooperation framework, promotes the cooperation of telecom operators in a competitive environment, and effectively improves the energy efficiency of 5G infrastructure; proposes a unified prediction method to improve the accuracy and responsiveness of user request prediction, and realizes the proactive and effective management of network resources; introduces a collaborative energy optimization strategy, refines user allocation through multi-angle coordination, and improves network efficiency by balancing the operation perspective and load allocation.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the method of the present invention, and these modifications or equivalent replacements cannot make the modified method deviate from the spirit and scope of the method of the present invention.

Claims

1. An adaptive base station energy consumption optimization method based on cross-operator cooperation, characterized in that: The following steps are involved: S1: Build a base station energy consumption model and confirm the optimization target of the optimization problem; S2: In the COPE framework, a combination of itransformer and Dlinear algorithms is used to perform user allocation based on the user allocation strategy of each base station and predict user distribution; S3: In the COPE framework, an asynchronous advantage actor-critic algorithm is used to dynamically adjust user allocation based on the predicted user distribution results; S4: In the COPE framework, a double-delayed deep deterministic policy gradient algorithm is used to evaluate the energy usage of each operator's base station, optimize the optimization target based on deep reinforcement learning, and decide whether to lease or shut down the base station.

2. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11: Build a network model, set the O operators providing network services in the 5G network coverage area as O={o n |n=1,…,O}, set the B 5G base stations constituting the operator’s 5G cellular network to Set U users within the network wireless coverage area to U n = {u n,j |j=1,…,|U n |}; S12: Build a base station load model and convert each base station b n,i Base station load Set to be assigned to the corresponding base station b n,i All users of u n,j The size of the network request The sum of the base station load Specifically expressed as: Among them, a n,i,j It is a binary parameter, indicating user u n,j With base station b n,i User association strategy between S13: Constructing a transmission rate model based on the channel Calculate user u n,j At base station b n,i Total throughput in Total throughput Specifically expressed as: in, Indicates base station b n,i The transmission power, Indicates base station b n,l The transmission power, Represents user u n,j and base station b n,i In the channel The channel gain on Represents user u n,j and base station b n,l In the channel The channel gain on the channel, N represents the noise power spectral density, Indicates channel bandwidth; S14: Build a transmission time model to calculate user request time Base station b n,i Channel working time Base station b n,i In a time slot t m Total working time in t w (b n,i ) and sleep time t s (b n,i ); S15: Construct an energy consumption model and convert the total energy consumption of the base station En n Set to all base stations b n,i The sum of the energy consumed during the active period, the total energy consumption of the base station En n Specifically expressed as: Among them, En(b n,i ) represents a base station b n,i The sum of energy consumed during the activity period, p w (b n,i ) represents base station b n,i Power in working state, p s (b n,i ) represents base station b n,i Power in sleep state, t r (b n,i ) indicates the request response time; S16: Build cost model and calculate operator O n The final cost Cost(n); S17: Build a cache strategy model and set the cache management target to maximize the cache hit rate CHR. The maximum cache hit rate CHR is specifically expressed as: Among them, D pre Indicates a set of data types being requested, Represents user u n,j Request count within time t; S18: The optimization objective of the optimization problem is set to: Among them, cus represents the cache update strategy. Indicates base station b n,i Energy consumption, Represents user u n,j By base station b n,i Serve.

3. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 2, characterized in that: In step S11, each base station b n,i There are C channels and a buffer with capacity e i The edge server of C channels is specifically set as Each channel The bandwidth is set to The edge server is specifically set to E = {e i |i=1,...,|B|}, base station B n The time interval is set to T = {t m |m=1,...,|T|}.

4. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 2, characterized in that: In step S11, each user u n,j At a specified time interval t m The size of the randomly generated network request is set to Each user u n,j At a specified time interval t m The randomly generated geo-coordinates of the network request are set to 5. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 2, characterized in that: Step S14 specifically includes the following steps: Step 1: According to user u n,j The network transmission speed will reduce the user request time Set to: in, Indicates that the edge server is at base station b n,i Respond to user u n,j The time required, t cloud Indicates the total time required for the cloud server to respond to user requests. The total time includes the communication time between the base station and the cloud server and the processing time of the cloud server. Step 2: On a per channel basis The base station b n,i Channel working time Set to the sum of all request completion times on the channel, base station b n,i Channel working time Specifically expressed as: Step 3: Set up base station b n,i In a time slot t m Total working time in t w (b n,i ) is set to: Among them, t a (b n,i ) represents base station b n,i The activation time, Indicates base station b n,i Maximum channel working time; Step 4: Set up base station b n,i In a time slot t m Sleep time t s (b n,i ) is set to: t s (b n,i )=t m -t w (b n,i )。 6. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 2, characterized in that: Step S16 specifically includes the following steps: Step 1: Change operator O n Total base station operating cost El (n) is set to: Among them, En represents the total power consumption, q s represents the upper limit of power consumption of the sth layer, C s Indicates the unit price corresponding to the sth layer; Step 2: Determine the additional costs incurred during the rental or leasing process based on the highest level of the tiered fee. out (n) and the rental price F in (n) are respectively expressed as: F out (n)=Cost′ El (n)-Cost El (n)+ω F in (n)=Cost El (n)-Cost′ El (n)+ω Among them, Cost′ El (n) represents operator O n The electricity cost incurred after the lease ends, ω represents the basic rent of the lease; Step 3: Calculate the operator O n The final cost Cost(n), operator O n The final cost Cost(n) is specifically expressed as: Cost(n)=Cost El (n)-F in (n)+F out (n)。 7. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 1, characterized in that: In step S2, the input of the COPE framework is user u n,j Network request data, base station b n,i The load information and system parameters of the COPE framework include energy cost and delay threshold. n,j With base station b n,i User association strategy between n,i,j .

8. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 1, characterized in that: In step S2, user distribution prediction is performed, and the time interval t m The current state is set to Current Status Including each user u n,j Historical positions and historical requests within a defined time window, current status Specifically expressed as: Where p represents the length of the history window, Represents user u n,j At the current time interval t m The predicted location and prediction request, Represents user u n,j The historical spatial coordinates of Represents user u n,j network request.

9. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 1, characterized in that: Step S3 specifically includes the following steps: S31: Base station b n,i The state space is set to S i , base station b n,i The state space S i Including current base station load and base station b n,i Network requests within the coverage area Base station b n,i The state space S i Specifically expressed as: S32: Base station b n,i The action space is set to A i , base station b n,i The action space A i It includes a set of user allocation operations, base station b n,i The action space A i Specifically expressed as: S33: Through the reward function R i Minimize base station b n,i The energy consumption and delay of the allocation strategy, the reward function R i Specifically expressed as: R i =-α·En(b n,i )-β·t bi Among them, t bi Indicates base station b n,i The service delay of , α and β are both adjustable parameters; S34: Based on the asynchronous dominant actor-critic algorithm, the optimal allocation strategy is obtained 10. The method for adaptive base station energy consumption optimization based on cross-operator cooperation according to claim 1, characterized in that: Step S4 The specific steps include: S41: Operator O n The state space is set to S o , operator O n The state space S o Including the base station load under management and base station energy status, operator O n The state space S o Specifically expressed as: S42: Operator O n The action space is set to A o , operator O n The action space A o The available lease set and deactivation decision set included, operator O n The action space A o Specifically expressed as: S43: Through the reward function R o Incentive Operator O n The cost and energy efficiency are expressed as: R o =-Cost(n)。

Images