Resource allocation method, apparatus and device, medium and program product

Abstract

The application provides a resource allocation method, device and equipment, medium and program product, applied to the technical field of communication, the method comprises: obtaining resource scheduling priority weight information and network state information, the resource scheduling priority weight information comprises m*n scheduling priority weight indexes; based on the resource scheduling priority weight information, m total scheduling priority weight indexes corresponding to m users are calculated; the m total scheduling priority weight indexes and the network state information are input into a first model to obtain an output result, the output result comprises transmission power and channel association of the m users; based on the output result, network resources are allocated to the m users. Since the method of the application can better process high-dimensional and variable information, the resource allocation mode obtained by the method according to the embodiment of the application can improve the network resource utilization rate compared with the mode of the prior art.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a resource allocation method, apparatus and equipment, medium and program product. Background Technology

[0002] Existing network service scheduling priority metrics primarily focus on three indicators: 5G QoS Identifier (5QI), Allocation / Retention Priority (ARP), and Radio Network / Frequency Selection Priority (RAT / RFSP). The existing network's scheduling priority weight aggregation strategy employs a "highest value for similar indicators, product of dissimilar indicators" approach to obtain the overall scheduling priority weight metric that aggregates 5QI, ARP, and RFSP. By adjusting the 5QI, ARP, and RFSP parameters for various user services and utilizing existing network radio resource scheduling, allocation, and management algorithms, the wireless network performance (e.g., transmission rate, transmission latency) for each user service is ensured to generally meet the overall scheduling priority weight metric while satisfying the communication requirements of each service.

[0003] Because the current wireless network environment is dynamic and complex, the network state information corresponding to complex networks is high-dimensional and variable. Existing resource allocation algorithms are difficult to handle high-dimensional, high-frequency continuous decision-making tasks. Therefore, allocating resources according to existing resource allocation algorithms will lead to low network resource utilization. Summary of the Invention

[0004] This application provides a resource allocation method, apparatus and equipment, medium and program product that can solve the problem of low network resource utilization.

[0005] In a first aspect, embodiments of this application provide a resource allocation method, the method comprising:

[0006] Obtain resource scheduling priority weight information and network status information. The resource scheduling priority weight information includes m×n scheduling priority weight indicators, where m and n are both positive integers. m is used to indicate the number of users, and n is used to indicate the types of scheduling priority weight indicators.

[0007] Based on the resource scheduling priority weight information, calculate m total scheduling priority weight indicators corresponding to m users. The total scheduling priority weight indicators are used to indicate the single indicator determined by the corresponding user by combining n scheduling priority weight indicators.

[0008] The m total scheduling priority weight indicators and the network status information are input into the first model to obtain the output results, which include the transmission power and channel correlation of the m users.

[0009] Based on the output results, network resources are allocated to the m users.

[0010] Secondly, embodiments of this application also provide a resource allocation device, the resource allocation device comprising:

[0011] The acquisition module is used to acquire resource scheduling priority weight information and network status information. The resource scheduling priority weight information includes m×n scheduling priority weight indicators, where m and n are both positive integers, m is used to indicate the number of users, and n is used to indicate the types of scheduling priority weight indicators.

[0012] The calculation module is used to calculate m total scheduling priority weight indicators corresponding to m users one by one based on the resource scheduling priority weight information. The total scheduling priority weight indicators are used to indicate the single indicator determined by the corresponding user by combining n scheduling priority weight indicators.

[0013] The input module is used to input the m total scheduling priority weight indicators and the network status information into the first model to obtain the output results, which include the transmission power and channel correlation of the m users;

[0014] The allocation module is used to allocate network resources to the m users based on the output results.

[0015] Thirdly, embodiments of this application also provide an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the resource allocation method as described in the first aspect.

[0016] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the resource allocation method as described in the first aspect.

[0017] Fifthly, a computer program product is provided, including computer instructions that, when executed by a processor, implement the steps of the resource allocation method as described in the first aspect.

[0018] In this embodiment, based on the resource scheduling priority weight information, m total scheduling priority weight indicators are calculated, corresponding to each of the m users. Even if more new scheduling priority weight indicators are subsequently introduced (increasing the dimensionality of the resource scheduling priority weight information), they can still be uniformly represented by the total scheduling priority weight indicators. Therefore, this method can effectively handle high-dimensional information. Furthermore, the trained first model outputs the corresponding output results (i.e., resource allocation methods) with the m total scheduling priority weight indicators and network status information, namely the transmission power and channel association of the m users. Therefore, this method can handle real-time changing network status information. In summary, since the method of this embodiment can better handle high-dimensional and variable information, the resource allocation method obtained by the method of this embodiment is more in line with the requirements of resource scheduling priority weight information and network status information than the prior art. Therefore, allocating network resources to the m users according to the resource allocation method can improve network resource utilization. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the resource allocation method provided in an embodiment of this application;

[0021] Figure 2 This is a schematic diagram of a wireless network scenario model provided in an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of the structure of a resource allocation device provided in an embodiment of this application;

[0023] Figure 4 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] To better understand, the following describes the technical terms that may be involved in the existing embodiments of this application, specifically as follows:

[0026] As shown in Table 1:

[0027]

[0028] Table 1

[0029] In existing technologies, wireless network resources are scheduled individually based on pre-configured priority weight indicators. This is suitable for current service assurance and resource allocation systems that only focus on three indicators: 5QI, ARP, and RFSP. However, with the introduction of more priority scheduling indicators, this system cannot guarantee the correct execution of other service priority indicators, nor can it adjust priority indicators not yet set in the current network. Current wireless network environments are dynamic and complex. Existing wireless network resource allocation algorithms struggle to capture potential cooperative and competitive relationships in dynamic and complex network environments. Furthermore, the network state information corresponding to dynamic and complex networks is high-dimensional and constantly changing, resulting in significant single-decision latency for existing wireless network resource allocation algorithms when processing high-dimensional data. Moreover, due to real-time changes in network state, existing wireless network resource allocation algorithms struggle to handle high-dimensional, high-frequency continuous decision-making tasks.

[0030] This application provides a resource allocation method.

[0031] See Figure 1 , Figure 1 This is a flowchart of the resource allocation method provided in the embodiments of this application, such as... Figure 1 As shown, it includes the following steps:

[0032] Step 101: Obtain resource scheduling priority weight information and network status information. The resource scheduling priority weight information includes m×n scheduling priority weight indicators, where m and n are both positive integers. m is used to indicate the number of users, and n is used to indicate the types of scheduling priority weight indicators.

[0033] In this step, such as Figure 2 The wireless network scenario model shown illustrates a user who can be a user equipment or terminal. Each user can only be associated with one base station but can be associated with multiple channels. The m×n scheduling priority weight indicators represent m users, each with n scheduling priority weight indicators. These indicators can include 5QI, ARP, and RFSP, among others. Network status information includes the channel gain of users and base stations in the network.

[0034] Step 102: Based on the resource scheduling priority weight information, calculate m total scheduling priority weight indicators corresponding to m users. The total scheduling priority weight indicators are used to indicate the single indicator determined by the corresponding user by combining n scheduling priority weight indicators.

[0035] In this step, each user has n scheduling priority weight indicators. Based on the n scheduling priority weight indicators, the total scheduling priority weight indicator for each user can be calculated. Using a single indicator to indicate the n scheduling priority weight indicators reduces the amount of data that needs to be considered when allocating resources based on the priority weight indicators.

[0036] Step 103: Input the m total scheduling priority weight indicators and the network status information into the first model to obtain the output results, which include the transmission power and channel association of the m users.

[0037] In this step, the m total scheduling priority weights corresponding to each of the m users in the network, along with network state information including the channel gain of users and base stations, are input into the first model. The first model is pre-trained and can output the network's resource allocation method, i.e., a model of the transmission power and channel correlation of all users in the network, based on the total scheduling priority weights of all users in the network and the channel gain of users and base stations.

[0038] Step 104: Based on the output results, allocate network resources to the m users.

[0039] In this step, network resources are allocated to the m users according to the output of the first model, so that the transmission power and channel association of the m users in the network after allocation are consistent with the situation indicated by the output.

[0040] In the resource allocation method provided in this application embodiment, based on the resource scheduling priority weight information, m total scheduling priority weight indicators corresponding to m users are calculated one-to-one. Even if more new scheduling priority weight indicators are subsequently introduced (increasing the dimension of resource scheduling priority weight information), they can still be uniformly represented by the total scheduling priority weight indicators. Therefore, the above method can handle high-dimensional information well. Furthermore, the trained first model outputs the output results (i.e., the resource allocation method) corresponding to the m total scheduling priority weight indicators and network status information, namely the transmission power and channel association of the m users. Therefore, the above method can handle real-time changing network status information. In summary, since the method of this application embodiment can better handle high-dimensional and variable information, the resource allocation method obtained by the method of this application embodiment is more in line with the requirements of resource scheduling priority weight information and network status information than the prior art. Therefore, allocating network resources to m users according to the resource allocation method can improve network resource utilization.

[0041] Optionally, the step of calculating m total scheduling priority weight indicators corresponding to m users one-to-one based on the resource scheduling priority weight information includes:

[0042] Based on n scheduling priority weight indicators, a fuzzy consistency matrix is ​​calculated. The fuzzy consistency matrix is ​​used to indicate the degree of importance of comparison between any two scheduling priority weight indicators among the n scheduling priority weight indicators.

[0043] Based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators, m total scheduling priority weight indicators corresponding one-to-one with m users are determined.

[0044] The following example illustrates how to calculate a fuzzy consistency matrix:

[0045] Given n scheduling priority weight indicators, by comparing the importance of each scheduling priority weight indicator pairwise, a fuzzy consistency matrix can be obtained:

[0046] A = (a ij ) n×n ,

[0047] Among them, a ij This indicates the importance of the i-th scheduling priority weight indicator relative to the j-th scheduling priority weight indicator. ij The possible values ​​for are shown in Table 2:

[0048] Indicators i and j are equally important 0.5 Indicator i is generally more important than j 0.6 Indicator i is significantly more important than j. 0.7 Indicator i is significantly more important than j 0.8 Indicator i is particularly important than j 0.9 Indicator i is extremely important than j 0.95 Indicator i is more important than others. Take the value in (0.5, 0.95). Indicator i is less important than j <![CDATA[a ij =1-a ji ]]>

[0049] Table 2

[0050] After obtaining the fuzzy consistency matrix, the total scheduling priority weight index for each user can be determined by combining the n scheduling priority weight indices for each user.

[0051] In this embodiment, even if a new scheduling priority weight index is introduced during the subsequent network resource allocation process, a new fuzzy consistency matrix can be calculated to quickly integrate the new scheduling priority weight index, giving more important evaluation weights to the more important scheduling priority weight index. The method of this embodiment has higher adaptability to high-dimensional information, thereby improving the resource utilization rate of subsequent resource allocation methods.

[0052] Optionally, determining the m total scheduling priority weight indicators corresponding one-to-one with the m users based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators includes:

[0053] The m×n scheduling priority weight indicators are normalized one by one to obtain m×n normalized scheduling priority weight indicators.

[0054] Based on the fuzzy consistency matrix and the preset scaling factor, calculate the n weight aggregation coefficients corresponding to the n scheduling priority weight indicators;

[0055] The n weight aggregation coefficients are normalized to obtain n normalized scheduling priority aggregation parameters corresponding to the n scheduling priority weight indices.

[0056] Based on the n normalized scheduling priority aggregation parameters and the m×n normalized scheduling priority weight indicators, m total scheduling priority weight indicators are determined.

[0057] The following example illustrates how to calculate the overall scheduling priority weight index:

[0058] The m×n scheduling priority weight indicators can form a scheduling priority weight matrix, denoted as: D=(d ij ) m×n .

[0059] Where, d ij This represents the j-th scheduling priority weight index for the i-th user.

[0060] For D=(d ij ) m×n By performing linear normalization on each item, we obtain m×n normalized scheduling priority weight indices, which can form a normalized scheduling priority weight matrix D. n =(d ij n ) m×n .

[0061] Where, d ij n This represents the j-th normalized scheduling priority weight index for the i-th user.

[0062] For a set of data X = [x1, x2, ..., x...] n The method for linear normalization is shown in the following formula (1):

[0063]

[0064] Where, x i n Let X be the linearly normalized value of the i-th item in the data set X, and max(X) and min(X) represent the maximum and minimum values ​​in the data set X, respectively.

[0065] Based on the fuzzy consistency matrix and the preset scaling factor, calculate the n weight aggregation coefficients corresponding to the n scheduling priority weight indicators one by one, as shown in the following formula (2):

[0066]

[0067] Where n represents the number of scheduling priority weight indicators, a ij The fuzzy consistency matrix represents the importance relationship between indicators i and j, where τ is a preset scaling factor. The larger τ is, the greater the difference in weight between each indicator, and the greater the computational cost. In one embodiment, τ = 2 yields a more suitable weight allocation, and the increase in computational cost remains within an acceptable range.

[0068] The weight aggregation coefficients are normalized, and the calculation method is shown in the following formula (3):

[0069]

[0070] Among them, w i n This represents the normalized scheduling priority aggregation parameter corresponding to the i-th scheduling priority weight index.

[0071] Finally, based on the n normalized scheduling priority aggregation parameters and the normalized scheduling priority weight index of each user, m total scheduling priority weight indices are determined.

[0072] In this embodiment, by performing multiple normalization processes, the instability of numerical calculation caused by the large differences in the value ranges among the n scheduling priority weight indicators can be reduced, making it easier to process high-dimensional data. Furthermore, when a new scheduling priority weight indicator is added, the newly added indicator can be better generalized, thereby improving the ability of the method in this embodiment to process high-dimensional, real-time changing data.

[0073] Optionally, determining m total scheduling priority weight indicators based on the n normalized scheduling priority aggregation parameters and the m×n normalized scheduling priority weight indicators includes:

[0074] Obtain n first parameters for each user, wherein the n first parameters are n products of the n normalized scheduling priority aggregation parameters and the n normalized scheduling priority weight indicators of the corresponding user;

[0075] The sum of the n first parameters for each user is used to determine the m total scheduling priority weight indicators.

[0076] In this embodiment, the calculation method of the total scheduling priority weight index of the i-th user is as shown in the following formula (4):

[0077]

[0078] Among them, w j n This represents the normalized aggregated scheduling priority parameter of the j-th scheduling priority weight index.

[0079] The method described in this embodiment can calculate the total scheduling priority weight index for each user. By using a single total scheduling priority weight index to represent the n types of scheduling priority weight indices for each user, as well as the additional types of scheduling priority weight indices added later, the dimensionality of the data is reduced, which can improve the processing capability of the method in this embodiment for high-dimensional data.

[0080] Optionally, the first model is a model obtained by training a preset model based on multiple historical data.

[0081] Each piece of historical data includes corresponding historical total scheduling priority weight indicators, historical network status information, historical transmission power of m users, and historical channel association information of m users;

[0082] The preset model includes a first network and a second network. The first network is used to output the transmission power and channel association of m users based on the input total scheduling priority weight index and network status information. The second network is used to calculate the total network transmission rate under the output condition of the first network.

[0083] The process of training the preset model based on multiple historical data includes:

[0084] Based on the historical total scheduling priority weight index, the historical transmission power of the m users, and the historical channel association of the m users, the historical total network transmission rate is determined;

[0085] The historical total scheduling priority weight index and the historical network status information are input into the preset model to obtain the first output result of the first network and the second output result of the second network determined based on the first output result. The parameters of the first network are adjusted based on the second output result and the parameters of the first network are adjusted based on the second output result and the historical total transmission rate of the network until the preset model meets the target condition in the (L+1)th training. The preset model after the Lth training is used as the first model, where L is a positive integer.

[0086] The target conditions include:

[0087] The second output result obtained from the (L+1)th training of the preset model is greater than or equal to the first preset parameter, and the difference between the second output result obtained from the (L+1)th training and the corresponding historical total network transmission rate is less than or equal to the second preset parameter.

[0088] Before utilizing the resource allocation method output by the first model, a preset model needs to be trained to obtain the first model. In an optional embodiment, the preset model (also referred to as the agent) can employ the DDPG algorithm. This is a deep reinforcement learning algorithm. Reinforcement learning is a common method for solving decision-making problems. Reinforcement learning includes two basic elements: states and actions. Executing an action in a given state is the policy. The agent needs to continuously explore and learn to obtain a good policy. Traditional reinforcement learning algorithms use tables for storage, and the state space is limited by the table's dimensionality. If the agent has many states and actions, the table size will continuously increase, even exceeding computer memory, and searching in each update phase may also take a long time. However, real-world tasks often have large state and action spaces, so traditional reinforcement learning algorithms perform poorly in complex scenarios. Deep learning exhibits good performance in handling high-dimensional data. Deep reinforcement learning combines deep learning with reinforcement learning, using DNNs to replace the original table storage for fitting and calculating the action value function.

[0089] DDPG is a deep reinforcement learning algorithm for continuous action spaces, combining deterministic policy gradient methods and deep learning techniques. DDPG is an improvement upon DQN, specifically addressing the continuous action space problem. DDPG first introduces a deterministic Actor network (the first network in this embodiment) and a Critic network (the second network in this embodiment). The Actor network learns a deterministic action policy, i.e., the probability of choosing the optimal action given a state. The Critic network evaluates the value of the action chosen by the Actor network. Both networks are parameterized using deep neural networks and trained using the backpropagation algorithm. DDPG uses the deterministic policy gradient method to update the parameters of the Actor network by maximizing the gradient of the action value function, and uses TD error to update the parameters of the Critic network. Equation (5) is the Actor policy update formula. Equation (6) is the loss function for Critic value update.

[0090]

[0091] Where J represents the objective function of the first network (usually set as the expected value of the cumulative reward). The objective function of the first network is given by the policy network parameters θ. π The gradient of N is given by the equation, where N represents the number of experiences stored in the experience pool. Indicates that in state s n Action a=π(s n The action-value function Q(s, a|θ) is as follows: Q Regarding the gradient of action a, Indicates that in state s n Lower policy network π(s|θ) π The output of the policy network parameter θ π The gradient.

[0092]

[0093] Where N represents the number of experiences stored in the experience pool, r n (s n a n ) indicates that in state s n The following action a is adopted n The obtained reward value, γ is the reward discount factor, and γ takes the value [0, 1]. The closer it is to 0, the more it emphasizes immediate rewards, and the closer it is to 1, the more it emphasizes future rewards. Q′(s n+1 ,π′(s n+1 |θ π′ )) represents the target Q-network's response to the next state s n+1 and target policy network parameters θ π′ The output value, Q(s) n a n ) represents the predicted state s of the Q-network. n Action a n Output value.

[0094] DDPG's training process is based on the ideas of experience replay and a target network. Experience replay is a technique for storing and reusing past experiences to break down correlations between samples and improve training efficiency. The target network is used to stabilize the training process by delaying the update of its parameters to reduce oscillations and instabilities during training. DDPG's goal is to maximize the action value function, enabling the policy to find the optimal action strategy in the environment.

[0095] In a wireless network scenario, the basic elements of reinforcement learning are set as follows: The state of the reinforcement learning agent consists of the user's total scheduling priority set weight and the channel gain with each base station. The action of the reinforcement learning agent is set as the user's transmission power and channel association status. The reward for reinforcement learning is set as the historical total network transmission rate.

[0096] Please refer to the above. Figure 2 The network consists of B base stations and U users, denoted as BS = {1, 2, ..., B} and UE = {1, 2, ..., U}, respectively. The smallest divisible unit of spectrum resource is defined as the bandwidth of a channel, with a bandwidth of w. There are K channels in total, denoted as C = {1, 2, ..., K}.

[0097] The historical data is explained as follows:

[0098] Historical data includes corresponding historical overall scheduling priority weight indicators, historical network status information, historical transmission power of m users, and historical channel association information of m users. The historical network status information includes the channel gain g of user u and base station b. u b The historical transmission power of m users includes the transmission power of user u on channel k, denoted as P. u k The historical channel correlation data for m users includes the correlation coefficient Y between user u and base station b. u b Y u b =1 indicates that user u and base station b are associated, Y u b =0 indicates that user u and base station b are not correlated. The correlation coefficient between user u and channel k is Z. u k Z u k =1 indicates that user u and channel k are associated, Z u k =0 indicates that user u and channel k are not associated. A user can only be associated with one base station, but can be associated with multiple channels.

[0099] Based on network status information, user transmission power, and channel correlation, the transmission rate of users in the network is determined as follows:

[0100] During communication, a user may experience interference from other users transmitting on the same channel. The interference experienced by user u on channel k from other users is denoted as I. u k As shown in equation (7):

[0101] I u k =∑ b∈BS ∑ u′∈UE，u′≠u Y u b z u k z u′ k g u′ b P u′ k (7),

[0102] Where u′ represents other users besides user u.

[0103] After determining that user u is experiencing interference from other users on channel k, the signal-to-noise ratio γ of user u on channel k can be further obtained. uk As shown in equation (8):

[0104]

[0105] Where, σ 2 To add Gaussian white noise power.

[0106] According to Shannon's channel formula, the transmission rate R of user u on channel k can be calculated. u k As shown in equation (9):

[0107] R u k =Wlog2(1+γ) u k (9),

[0108] The total transmission rate of user u is denoted as R. u As shown in equation (10):

[0109] R u =∑ k∈C R u k (10),

[0110] User u has a minimum transmission rate constraint R u ,in It needs to satisfy R u ≥R u min Transmission rate is a network performance indicator that quantifies resource scheduling and service assurance. To reflect the matching relationship between resource allocation results and service scheduling priorities, this application uses normalized Euclidean distance to quantify the matching degree.

[0111] In the U-dimensional Euclidean space, the overall scheduling priority weight of each user's service and the transmission rate of each user can be considered as two points (α1, α2, ..., α...). U (R1, R2, ..., R) U The normalized point of the original two points is located at two points 1 away from the origin along the ray direction from the origin to these two points, denoted as (α1). n α2 n , …, α U n ) and (R1 n R2 n , ..., R U n The distance L between these two is the matching degree, as shown in equation (11):

[0112]

[0113] L≤L max L max This represents the maximum tolerance for a perfect match.

[0114] The user's maximum uplink transmission power is P. max It is necessary to satisfy ∑P u k ≤P max For channel gain, the channel fading models considered include large-scale fading and small-scale fading. Large-scale fading includes path loss and shadowing fading. The expression for path loss is δ(d) = 37 + 30log(d), where d is the distance between the base station and the user, and the standard deviation of shadowing fading is set to 8dB. Small-scale fading uses the Rayleigh fading model with a mean of 0dB and a variance of 8dB.

[0115] The optimization objective of wireless network resource allocation is to maximize the total network transmission rate. The optimization problem can be expressed as follows:

[0116]

[0117] ∑P u k ≤P max (12d),

[0118]

[0119] L≤L max (12f).

[0120] Equation (12a) represents the total network transmission rate; equations (12b) and (12c) are binary parameter constraints and a constraint that a user can only be associated with one base station; equation (12d) is the maximum total transmission power constraint; equation (12e) is the minimum transmission rate constraint for a user; and equation (12f) is the tolerance constraint for the matching degree between resource allocation results and total scheduling priority weights.

[0121] Due to continuous variables (such as P) u k ) and binary variables (such as Y) u b Given the nonlinear constraints, optimization problem (12a) is a non-convex and non-concave optimization problem. Furthermore, the joint power selection and channel allocation problems in complex communication networks are NP-hard problems. Deep reinforcement learning can be used to solve these types of non-convex optimization problems.

[0122] The historical total scheduling priority weight index and the historical network status information are input into a preset model to obtain the first output result of the first network. The first output result includes the user's transmission power and channel correlation. The second network determines the total network transmission rate based on the transmission power and channel correlation, calculated as shown in equations (7) to (10) and (12a) above. The second output result is compared with the historical total network transmission rate determined based on the historical total scheduling priority weight index, the historical transmission power of the m users, and the historical channel correlation of the m users, allowing for parameter tuning of the second network. The first network can also be tuned based on the second output result.

[0123] In this embodiment, the first model is obtained by training the preset model in the above manner, which can improve the first model's ability to handle network resource allocation scenarios, thereby enabling the output of the first model to improve network resource utilization.

[0124] Optionally, determining the historical total network transmission rate based on the historical total scheduling priority weight index, the historical transmission power of the m users, and the historical channel association of the m users includes:

[0125] Based on the historical transmission power of the m users and the historical channel association of the m users, determine m×k transmission rates, where k is a positive integer and k is used to indicate the number of channels;

[0126] Based on the m second parameters corresponding to the m users and the m total scheduling priority weight indicators, the target matching tolerance between the network's communication requirements, transmission performance and the m total scheduling priority weight indicators is determined, where the second parameter is the sum of the k transmission rates of the corresponding user.

[0127] The product of the third, fourth, and fifth parameters is determined as the total historical network transmission rate;

[0128] The fifth parameter is the sum of the m×k transmission rates; if any of the k transmission rates is less than the minimum transmission rate constraint, the third parameter is 0; if there is no transmission rate less than the minimum transmission rate constraint among the k transmission rates, the third parameter is 1; if the target matching tolerance is greater than or equal to the maximum matching tolerance, the fourth parameter is 0; if the target matching tolerance is less than the maximum matching tolerance, the fourth parameter is 1.

[0129] In this embodiment, during the training of the preset model, the historical total network transmission rate (i.e., reward function) used to tune the second network by comparing it with the second output result is calculated as follows:

[0130] r = f(P) u k Z u k )g(P u k Z u k )∑ u∈UE R u (13),

[0131] Where, f(P) u k Z u k ) is the transmission rate constraint threshold function, i.e., the third parameter, g(P) u k Z u k ) represents the matching tolerance constraint threshold function, i.e., the fourth parameter, as shown in equations (14) and (15), respectively. The expression for the Sigmoid function is shown in equation (16):

[0132] f(P u k Z u k )=∏ u∈UE S(R u -R u min (14),

[0133] g(P u k Z u k )=S(L max -L) (15),

[0134]

[0135] It should be noted that the calculation method of the second parameter is shown in Equation (10), the calculation method of the fifth parameter is shown in Equation (12a), and the calculation method of the target matching tolerance is shown in Equation (11), which will not be repeated here.

[0136] In this embodiment, two constraints are considered in the reward function: a transmission rate constraint and a matching tolerance constraint. If the constraints are not met, a penalty of zero reward value is applied directly. In this way, cases where a reward value still exists even when the constraints are not met can be excluded, making the reward function more closely reflect reality and thus improving the accuracy of model training.

[0137] See Figure 3 , Figure 3This is a schematic diagram of the structure of a resource allocation device provided in an embodiment of this application. For example... Figure 3 As shown, the device 300 includes:

[0138] The acquisition module 301 is used to acquire resource scheduling priority weight information and network status information. The resource scheduling priority weight information includes m×n scheduling priority weight indicators, where m and n are both positive integers. m is used to indicate the number of users, and n is used to indicate the types of scheduling priority weight indicators.

[0139] The calculation module 302 is used to calculate m total scheduling priority weight indicators corresponding to m users one by one based on the resource scheduling priority weight information. The total scheduling priority weight indicators are used to indicate the single indicator determined by the corresponding user by combining n scheduling priority weight indicators.

[0140] The input module 303 is used to input the m total scheduling priority weight indicators and the network status information into the first model to obtain the output results, which include the transmission power and channel correlation of the m users.

[0141] The allocation module 304 is used to allocate network resources to the m users based on the output results.

[0142] Optionally, the computing module 302 is further configured to:

[0143] Based on n scheduling priority weight indicators, a fuzzy consistency matrix is ​​calculated. The fuzzy consistency matrix is ​​used to indicate the degree of importance of comparison between any two scheduling priority weight indicators among the n scheduling priority weight indicators.

[0144] Based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators, m total scheduling priority weight indicators corresponding one-to-one with m users are determined.

[0145] Optionally, the computing module 302 is further configured to:

[0146] The step of determining m total scheduling priority weight indicators corresponding one-to-one with m users based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators includes:

[0147] The m×n scheduling priority weight indicators are normalized one by one to obtain m×n normalized scheduling priority weight indicators.

[0148] Based on the fuzzy consistency matrix and the preset scaling factor, calculate the n weight aggregation coefficients corresponding to the n scheduling priority weight indicators;

[0149] The n weight aggregation coefficients are normalized to obtain n normalized scheduling priority aggregation parameters corresponding to the n scheduling priority weight indices.

[0150] Based on the n normalized scheduling priority aggregation parameters and the m×n normalized scheduling priority weight indicators, m total scheduling priority weight indicators are determined.

[0151] Optionally, the computing module 302 is further configured to:

[0152] Obtain n first parameters for each user, wherein the n first parameters are n products of the n normalized scheduling priority aggregation parameters and the n normalized scheduling priority weight indicators of the corresponding user;

[0153] The sum of the n first parameters for each user is determined as the m total scheduling priority weight indicators.

[0154] Optionally, the first model is a model obtained by training a preset model based on multiple historical data.

[0155] Each piece of historical data includes corresponding historical total scheduling priority weight indicators, historical network status information, historical transmission power of m users, and historical channel association information of m users;

[0156] The preset model includes a first network and a second network. The first network is used to output the transmission power and channel association of m users based on the input total scheduling priority weight index and network status information. The second network is used to calculate the total network transmission rate under the output condition of the first network.

[0157] The process of training the preset model based on multiple historical data includes:

[0158] Based on the historical total scheduling priority weight index, the historical transmission power of the m users, and the historical channel association of the m users, the historical total network transmission rate is determined;

[0159] The historical total scheduling priority weight index and the historical network status information are input into the preset model to obtain the first output result of the first network and the second output result of the second network determined based on the first output result. The parameters of the first network are adjusted based on the second output result and the parameters of the first network are adjusted based on the second output result and the historical total transmission rate of the network until the preset model meets the target condition in the (L+1)th training. The preset model after the Lth training is used as the first model, where L is a positive integer.

[0160] The target conditions include:

[0161] The second output result obtained from the (L+1)th training of the preset model is greater than or equal to the first preset parameter, and the difference between the second output result obtained from the (L+1)th training and the corresponding historical total network transmission rate is less than or equal to the second preset parameter.

[0162] Optionally, determining the historical total network transmission rate based on the historical total scheduling priority weight index, the historical transmission power of the m users, and the historical channel association of the m users includes:

[0163] Based on the historical transmission power of the m users and the historical channel association of the m users, determine m×k transmission rates, where k is a positive integer and k is used to indicate the number of channels;

[0164] Based on the m second parameters corresponding to the m users and the m total scheduling priority weight indicators, the target matching tolerance between the network's communication requirements, transmission performance and the m total scheduling priority weight indicators is determined, where the second parameter is the sum of the k transmission rates of the corresponding user.

[0165] The product of the third, fourth, and fifth parameters is determined as the total historical network transmission rate;

[0166] The fifth parameter is the sum of the m×k transmission rates; if any of the k transmission rates is less than the minimum transmission rate constraint, the third parameter is 0; if there is no transmission rate less than the minimum transmission rate constraint among the k transmission rates, the third parameter is 1; if the target matching tolerance is greater than or equal to the maximum matching tolerance, the fourth parameter is 0; if the target matching tolerance is less than the maximum matching tolerance, the fourth parameter is 1.

[0167] It should be noted that the resource allocation device 300 provided in this application embodiment can achieve the following: Figure 1 The entire technical process of the resource allocation method shown in the embodiment, and the same technical effect, will not be described again here to avoid repetition.

[0168] The data transmission device in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. Non-mobile electronic devices can also be servers, network attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not specifically limit the specific devices.

[0169] Optionally, such as Figure 4 As shown, this application embodiment also provides an electronic device 400, including a processor 401 and a memory 402. The memory 402 stores a program or instructions that can run on the processor 401. When the program or instructions are executed by the processor 401, they implement the various steps of the above-described resource allocation method embodiment and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0170] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0171] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described resource allocation method embodiments and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0172] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0173] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above-described... Figure 1 The various processes of the resource allocation method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0174] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0175] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0176] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A resource allocation method, characterized in that, The method includes: Obtain resource scheduling priority weight information and network status information. The resource scheduling priority weight information includes m×n scheduling priority weight indicators, where m and n are both positive integers. m is used to indicate the number of users, and n is used to indicate the types of scheduling priority weight indicators. Based on the resource scheduling priority weight information, calculate m total scheduling priority weight indicators corresponding to m users. The total scheduling priority weight indicators are used to indicate the single indicator determined by the corresponding user by combining n scheduling priority weight indicators. The m total scheduling priority weight indicators and the network status information are input into the first model to obtain the output results, which include the transmission power and channel correlation of the m users. Based on the output results, network resources are allocated to the m users; The step of calculating m total scheduling priority weight indicators corresponding to m users one-to-one based on the resource scheduling priority weight information includes: Based on n scheduling priority weight indicators, a fuzzy consistency matrix is ​​calculated. The fuzzy consistency matrix is ​​used to indicate the degree of importance of comparison between any two scheduling priority weight indicators among the n scheduling priority weight indicators. Based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators, determine the m total scheduling priority weight indicators that correspond one-to-one with the m users. The process of determining m total scheduling priority weight indicators corresponding one-to-one with m users based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators includes: The m×n scheduling priority weight indicators are normalized one by one to obtain m×n normalized scheduling priority weight indicators. Based on the fuzzy consistency matrix and the preset scaling factor, calculate the n weight aggregation coefficients corresponding to the n scheduling priority weight indicators; The n weight aggregation coefficients are normalized to obtain n normalized scheduling priority aggregation parameters corresponding to the n scheduling priority weight indices. Based on the n normalized scheduling priority aggregation parameters and the m×n normalized scheduling priority weight indicators, m total scheduling priority weight indicators are determined. The determination of m total scheduling priority weight indicators based on the n normalized scheduling priority aggregation parameters and the m×n normalized scheduling priority weight indicators includes: Obtain n first parameters for each user, wherein the n first parameters are n products of the n normalized scheduling priority aggregation parameters and the n normalized scheduling priority weight indicators of the corresponding user; The sum of the n first parameters for each user is used to determine the m total scheduling priority weight indicators.

2. The method according to claim 1, characterized in that, The first model is a model obtained by training a preset model based on multiple historical data. Each piece of historical data includes corresponding historical total scheduling priority weight indicators, historical network status information, historical transmission power of m users, and historical channel association information of m users; The preset model includes a first network and a second network. The first network is used to output the transmission power and channel association of m users based on the input total scheduling priority weight index and network status information. The second network is used to calculate the total network transmission rate under the output condition of the first network. The process of training the preset model based on multiple historical data includes: Based on the historical total scheduling priority weight index, the historical transmission power of the m users, and the historical channel association of the m users, the historical total network transmission rate is determined; The historical total scheduling priority weight index and the historical network status information are input into the preset model to obtain the first output result of the first network and the second output result of the second network determined based on the first output result. The parameters of the first network are adjusted based on the second output result and the parameters of the first network are adjusted based on the second output result and the historical total transmission rate of the network until the preset model meets the target condition in the (L+1)th training. The preset model after the Lth training is used as the first model, where L is a positive integer. The target conditions include: The second output result obtained from the (L+1)th training of the preset model is greater than or equal to the first preset parameter, and the difference between the second output result obtained from the (L+1)th training and the corresponding historical total network transmission rate is less than or equal to the second preset parameter.

3. The method according to claim 2, characterized in that, The determination of the historical total network transmission rate based on the historical total scheduling priority weight index, the historical transmission power of the m users, and the historical channel association of the m users includes: Based on the historical transmission power of the m users and the historical channel association of the m users, determine m×k transmission rates, where k is a positive integer and k is used to indicate the number of channels; Based on the m second parameters corresponding to the m users and the m total scheduling priority weight indicators, the target matching tolerance between the network's communication requirements, transmission performance and the m total scheduling priority weight indicators is determined, where the second parameter is the sum of the k transmission rates of the corresponding user. The product of the third, fourth, and fifth parameters is determined as the total historical network transmission rate; The fifth parameter is the sum of the m×k transmission rates; if any of the k transmission rates is less than the minimum transmission rate constraint, the third parameter is 0; if there is no transmission rate less than the minimum transmission rate constraint among the k transmission rates, the third parameter is 1; if the target matching tolerance is greater than or equal to the maximum matching tolerance, the fourth parameter is 0; if the target matching tolerance is less than the maximum matching tolerance, the fourth parameter is 1.

4. A resource allocation device, characterized in that, The resource allocation device includes: The acquisition module is used to acquire resource scheduling priority weight information and network status information. The resource scheduling priority weight information includes m×n scheduling priority weight indicators, where m and n are both positive integers, m is used to indicate the number of users, and n is used to indicate the types of scheduling priority weight indicators. The calculation module is used to calculate m total scheduling priority weight indicators corresponding to m users one by one based on the resource scheduling priority weight information. The total scheduling priority weight indicators are used to indicate the single indicator determined by the corresponding user by combining n scheduling priority weight indicators. The input module is used to input the m total scheduling priority weight indicators and the network status information into the first model to obtain the output results, which include the transmission power and channel correlation of the m users; The allocation module is used to allocate network resources to the m users based on the output results; The computing module is also used for: Based on n scheduling priority weight indicators, a fuzzy consistency matrix is ​​calculated. The fuzzy consistency matrix is ​​used to indicate the degree of importance of comparison between any two scheduling priority weight indicators among the n scheduling priority weight indicators. Based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators, determine the m total scheduling priority weight indicators that correspond one-to-one with the m users. The computing module is also used for: The process of determining m total scheduling priority weight indicators corresponding one-to-one with m users based on the fuzzy consistency matrix and the m×n scheduling priority weight indicators includes: The m×n scheduling priority weight indicators are normalized one by one to obtain m×n normalized scheduling priority weight indicators. Based on the fuzzy consistency matrix and the preset scaling factor, calculate the n weight aggregation coefficients corresponding to the n scheduling priority weight indicators; The n weight aggregation coefficients are normalized to obtain n normalized scheduling priority aggregation parameters corresponding to the n scheduling priority weight indices. Based on the n normalized scheduling priority aggregation parameters and the m×n normalized scheduling priority weight indicators, m total scheduling priority weight indicators are determined. The computing module is also used for: Obtain n first parameters for each user, wherein the n first parameters are n products of the n normalized scheduling priority aggregation parameters and the n normalized scheduling priority weight indicators of the corresponding user; The sum of the n first parameters for each user is used to determine the m total scheduling priority weight indicators.

5. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the resource allocation method as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the resource allocation method as described in any one of claims 1 to 3.

7. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the resource allocation method as described in any one of claims 1 to 3.

Images