Traffic adaptive carrier aggregation and resource scheduling method, system and device for 230MHz electric power authorized frequency band and storage medium

Abstract

The invention discloses a flow adaptive carrier aggregation and resource scheduling method, system and device for a 230MHz electric power authorized frequency band and a storage medium, and the method is based on a 230MHz electric power dedicated frequency spectrum, and comprises the following steps: obtaining data arrival time and data size of different users; predicting the data arrival time and the data size of different users based on an LSTM neural network; according to the predicted arrival time and size of the data, allocating different resources to users with different requirements by using a subcarrier activation and deactivation algorithm based on Q-Learning; according to the invention, the service rate requirement of a user can be met, and data transmission can be carried out with the lowest energy consumption; the subcarriers can be dynamically activated or stopped, and the utilization efficiency of resources is improved.

Description

Technical Field

[0001] The present invention relates to the field of power wireless communication technology, and in particular to a method, system, device and storage medium for traffic adaptive carrier aggregation and resource scheduling in a 230 MHz power authorized frequency band. Background Art

[0002] Although TD-LTE-based wireless private networks have been deployed in the 230MHz frequency band allocated specifically for the power system, the frequency band utilization is not high and the support for new services in the power distribution field is insufficient. The main reason is that the frequency points in the 230MHz frequency band are relatively scattered and the bandwidth of a single frequency point is also low. Although the existing system uses technologies such as carrier aggregation and adaptive modulation to improve spectrum utilization, the effect is not obvious due to the business characteristics of the distribution area. On the one hand, a large number of terminals are deployed in the distribution area, and these terminals have greater requirements for uplinks. On the other hand, these heterogeneous terminals have very different requirements for speed, latency and other factors.

[0003] Existing carrier aggregation and resource scheduling solutions do not take this into account, especially the significant demand differences of heterogeneous terminals. In order to meet the growing demand for high-frequency data collection and interaction in the power distribution field, it is necessary to study new carrier aggregation and resource scheduling methods. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method, system, device and storage medium for traffic adaptive carrier aggregation and resource scheduling in the 230MHz power licensed frequency band, which adaptively performs carrier aggregation and resource scheduling based on the prediction of data arrival time and data size of different users (terminals), while meeting user needs, maximizing spectrum utilization and reducing terminal power consumption.

[0005] Technical solution: The method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorized frequency band described in the present invention is based on the 230MHz power dedicated spectrum and includes the following steps:

[0006] Get the data arrival time and data size of different users;

[0007] Predict the data arrival time and data size of different users based on LSTM neural network;

[0008] According to the predicted data arrival time and size, a Q-Learning-based subcarrier activation and deactivation algorithm is used to allocate different resources to users with different needs.

[0009] Furthermore, the data arrival time and data size of different users are obtained as follows:

[0010] The N historical data packet arrival time intervals of each user among the K users are constructed as a K×N random matrix I K×N , where K is the number of rows in the matrix, representing K users; N is the length of the time interval sequence of the user's data packets;

[0011] The N historical data packet sizes of each user among the K users are constructed as a K×N random matrix S K×N , where K is the number of rows in the matrix, representing K users; N is the length of the user's data packet size sequence;

[0012] The random matrix I K×N , S K×N , divide the training set and test set.

[0013] Furthermore, the data arrival time and data size of different users are predicted based on the LSTM neural network as follows:

[0014] Set neural network training parameters;

[0015] The test set and training set are input into the LSTM neural network for training and prediction respectively.

[0016] Furthermore, using a Q-Learning-based subcarrier activation and deactivation algorithm, allocating different resources to users with different needs includes the following steps:

[0017] Define the total number of users as K and the total number of subcarriers as M; represent each user as an agent with its CCA instance: define the CCA observation vector of the kth user as C k ; Define the average rate of the kth user in the last CCA interval as Define the user's service quality requirements as

[0018] According to the predicted data arrival size and data arrival time; according to the defined total number of users K, total number of subcarriers M and agent of CCA instance, the action equation and state function of Q-Learning are set, and then the reward function of the corresponding action under the state is calculated according to the action equation and state function; the action with the highest reward is selected, and the total rate of user k is calculated.

[0019] Furthermore, the total number of users is defined as K and the total number of subcarriers is defined as M as follows:

[0020] Assume that there are K users in total. The network uses carrier aggregation technology CCA to perform carrier aggregation on M subcarriers CC, where each subcarrier contains N resource block groups, and a resource block group consists of multiple resource blocks; the maximum number of CCs activated by the kth user is M k, including a primary subcarrier PCC and multiple secondary subcarriers SCC; PCC is always activated for users.

[0021] Furthermore, each user is represented as an agent with its CCA instance as follows:

[0022] The CCA observation vector of the kth user is defined as C k =[φ k ,q k ], where φ k =[φ k,1 ,φ k,2 ,…,φ k,M-1 ] represents the CQI vector of the M-1 secondary subcarriers respectively relative to the kth user, q k Indicates the data size;

[0023] According to the CQI vector of each user, the traffic scheduler allocates different numbers of resource block groups to each user.

[0024] Furthermore, the action equation of Q-Learning is defined as follows:

[0025] Let the action of the kth user be defined as:

[0026] a k =[α k ]=[α k,1 ,α k,2 ,...,α k,m ,.,α k,M-1 ] T

[0027] where α k,m =1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

[0028] Furthermore, the state function formula is as follows:

[0029] s k =[τ k ,l k ,v k ];

[0030] Among them, τ in the state function k Indicates whether the action is applied. The calculation formula is as follows:

[0031]

[0032] in, is the predicted value of the next data arrival time of the kth user, T k,eis the arrival time of the previous data. If the difference between the two variables is less than the trigger interval T of the CCA algorithm, CCA , indicating that there is no need to activate a new SCC; T CCA Indicates the trigger interval of the CCA algorithm;

[0033] l in the state function k Indicates the number of SCCs that need to be activated. The calculation formula is as follows:

[0034]

[0035] in, and q max They represent the predicted value of the next data size of the kth user and the maximum capacity of a subcarrier, respectively, and M represents the total number of subcarriers;

[0036] v in the state function k Indicates whether SCC needs to be activated. The calculation formula is as follows:

[0037]

[0038] in, and They represent the average rate of the kth user in the last CCA interval and the user's quality of service requirement QoS respectively; The calculation of is as follows:

[0039]

[0040] where a k,m When it is 1, it means that the mth CC is activated for the kth user, otherwise it is deactivated. k,m represents the achievable rate of the m-th CC for the k-th user, which is calculated based on the defined φk and the resource block group allocated to the user.

[0041] Furthermore, the reward function formula is as follows: represents the number of SCCs activated by the kth user at the tth iteration, τ k Indicates whether the action is applied, then the reward function of the kth user is defined as follows:

[0042]

[0043] Among them, r k,1 The formula is as follows:

[0044]

[0045] It represents the reward value when an activation action is not applied. The reward value is 1 only when only PCC is activated.

[0046] in denote the number of SCCs activated by the kth user at the t-2th iteration, t-1th iteration, and tth iteration, respectively;

[0047] r k,2 The formula is as follows:

[0048]

[0049] Indicates that when the data size exceeds the PCC capacity, more SCCs are activated; conversely, the throughput and the number of activated SCCs are balanced; ω is a parameter used to control the rate and number of SCCs, which can be set according to the situation, q k is the data size of the kth user, q max is the maximum capacity of the PCC;

[0050] r k,3 The formula is as follows:

[0051]

[0052] represents the reward obtained by the user's average rate relative to the user's required rate, where represents the average rate of the kth user, represents the rate requirement of the kth user.

[0053] Furthermore, we select the action with the highest reward and calculate the total rate of user k as follows:

[0054]

[0055] Among them, V k,m (t) represents the achievable rate of the m-th subcarrier for the k-th user, where α k,m (t)=1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

[0056] The traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power authorized frequency band described in the present invention is based on the 230MHz power dedicated spectrum and includes:

[0057] Acquisition module: used to obtain data arrival time and data size of different users;

[0058] Prediction module: used to predict the data arrival time and data size of different users based on LSTM neural network;

[0059] Allocation module: used to allocate different resources to users with different needs based on the predicted data arrival time and size using the Q-Learning-based subcarrier activation and deactivation algorithm.

[0060] Furthermore, in the acquisition module, the data arrival time and data size of different users are obtained as follows:

[0061] The N historical data packet arrival time intervals of each user among the K users are constructed as a K×N random matrix I K×N , where K is the number of rows in the matrix, representing K users; N is the length of the time interval sequence of the user's data packets;

[0062] The N historical data packet sizes of each user among the K users are constructed as a K×N random matrix S K×N , where K is the number of rows in the matrix, representing K users; N is the length of the user's data packet size sequence;

[0063] The random matrix I K×N , S K×N , divide the training set and test set.

[0064] Furthermore, in the prediction module, the data arrival time and data size of different users are predicted based on the LSTM neural network as follows: neural network training parameters are set; the test set and the training set are respectively input into the LSTM neural network for training and prediction.

[0065] Furthermore, in the allocation module, using a Q-Learning-based subcarrier activation and deactivation algorithm, allocating different resources to users with different needs includes the following steps:

[0066] Define the total number of users as K and the total number of subcarriers as M; represent each user as an agent with its CCA instance: define the CCA observation vector of the kth user as C k ; Define the average rate of the kth user in the last CCA interval as Define the user's service quality requirements as

[0067] According to the predicted data arrival size and data arrival time; according to the defined total number of users K, total number of subcarriers M and agent of CCA instance, the action equation and state function of Q-Learning are set, and then the reward function of the corresponding action under the state is calculated according to the action equation and state function; the action with the highest reward is selected, and the total rate of user k is calculated.

[0068] Furthermore, in the allocation module, the total number of users is defined as K and the total number of subcarriers is defined as M as follows:

[0069] Assume that there are K users in total. The network uses carrier aggregation technology CCA to perform carrier aggregation on M subcarriers CC, where each subcarrier contains N resource block groups, and a resource block group consists of multiple resource blocks; the maximum number of CCs activated by the kth user is M k , including a primary subcarrier PCC and multiple secondary subcarriers SCC; PCC is always activated for users.

[0070] Furthermore, in the allocation module, each user is represented as an agent with its CCA instance as follows:

[0071] The CCA observation vector of the kth user is defined as C k =[φ k ,q k ], where φ k =[φ k,1 ,φ k,2 ,…,φ k,M-1 ] represents the CQI vector of the M-1 secondary subcarriers respectively relative to the kth user, q k Indicates the data size;

[0072] According to the CQI vector of each user, the traffic scheduler allocates different numbers of resource block groups to each user.

[0073] Furthermore, in the allocation module, the action equation of Q-Learning is defined as follows:

[0074] Let the action of the kth user be defined as:

[0075] a k =[α k ]=[α k,1 ,α k,2 ,...,α k,m ,...,α k,M-1 ] T

[0076] where α k,m =1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

[0077] Furthermore, in the allocation module, the state function formula is as follows:

[0078] s k =[τ k ,l k ,v k ];

[0079] Among them, τ in the state function k Indicates whether the action is applied. The calculation formula is as follows:

[0080]

[0081] in, is the predicted value of the next data arrival time of the kth user, T k,e is the arrival time of the previous data. If the difference between the two variables is less than the trigger interval T of the CCA algorithm, CCA , indicating that there is no need to activate a new SCC; T CCA Indicates the trigger interval of the CCA algorithm;

[0082] l in the state function k Indicates the number of SCCs that need to be activated. The calculation formula is as follows:

[0083]

[0084] in, and q max They represent the predicted value of the next data size of the kth user and the maximum capacity of a subcarrier, respectively, and M represents the total number of subcarriers;

[0085] v in the state function k Indicates whether SCC needs to be activated. The calculation formula is as follows:

[0086]

[0087] in, and They represent the average rate of the kth user in the last CCA interval and the user's quality of service requirement QoS respectively; The calculation of is as follows:

[0088]

[0089] where a k,m When it is 1, it means that the mth CC is activated for the kth user, otherwise it is deactivated. k,m represents the achievable rate of the m-th CC for the k-th user, which is calculated based on the defined φk and the resource block group allocated to the user.

[0090] Furthermore, in the allocation module, the reward function formula is as follows: represents the number of SCCs activated by the kth user at the tth iteration, τ k Indicates whether the action is applied, then the reward function of the kth user is defined as follows:

[0091]

[0092] Among them, r k,1 The formula is as follows:

[0093]

[0094] It represents the reward value when an activation action is not applied. The reward value is 1 only when only PCC is activated.

[0095] in denote the number of SCCs activated by the kth user at the t-2th iteration, t-1th iteration, and tth iteration, respectively;

[0096] r k,2 The formula is as follows:

[0097]

[0098] Indicates that when the data size exceeds the PCC capacity, more SCCs are activated; conversely, the throughput and the number of activated SCCs are balanced; ω is a parameter used to control the rate and number of SCCs, which can be set according to the situation, q k is the data size of the kth user, q max is the maximum capacity of the PCC;

[0099] r k,3 The formula is as follows:

[0100]

[0101] represents the reward obtained by the user's average rate relative to the user's required rate, where represents the average rate of the kth user, represents the rate requirement of the kth user.

[0102] Furthermore, in the allocation module, the action with the highest reward is selected, and the total rate formula for user k is calculated as follows:

[0103]

[0104] Among them, V k,m (t) represents the achievable rate of the m-th subcarrier for the k-th user, where α k,m (t)=1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

[0105] An electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, it implements any one of the traffic adaptive carrier aggregation and resource scheduling methods for the 230 MHz power authorization frequency band.

[0106] A storage medium described in the present invention stores a computer program, and when the computer program is executed by a processor, it implements any one of the traffic adaptive carrier aggregation and resource scheduling methods for the 230MHz power authorization frequency band.

[0107] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: user traffic can be predicted and subcarriers can be activated in advance to meet user rate requirements. When activating subcarriers to transmit user data, it can not only meet the user's service rate requirements, but also transmit data with the lowest energy consumption. Subcarriers can be dynamically activated or deactivated, which improves resource utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0109] Figure 2 It is a flow chart of the LSTM prediction algorithm on which the present invention is based;

[0110] Figure 3 is a schematic diagram of the resource block structure of the present invention;

[0111] Figure 4 is a subcarrier activation scheduling timing diagram for a single user of the present invention;

[0112] Figure 5 It is a flow chart of the subcarrier activation and deactivation algorithm based on Q-Learning of the present invention;

[0113] Figure 6 is the average number of subcarriers activated by each user in different modes of the present invention;

[0114] Figure 7 are the rate requirements and throughputs of different users under different modes of the present invention;

[0115] Figure 8 It is the transmission data delay of different users under different modes of the present invention. DETAILED DESCRIPTION

[0116] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0117] like Figure 1 As shown, the embodiment of the present invention provides a traffic adaptive carrier aggregation and resource scheduling method for a 230MHz power authorized frequency band, based on a 230MHz power dedicated spectrum, comprising the following steps: obtaining data arrival time and data size of different users; specifically as follows: constructing the arrival time intervals of N historical data packets of K users and each user into a K×N type random matrix I K×N, where K is the number of rows in the matrix, representing K users; N is the length of the time interval sequence of the user's data packets arriving at the destination; K users and each user's N historical data packet sizes are constructed into a K×N random matrix S K×N , where K is the number of rows in the matrix, representing K users; N is the length of the user's data packet size sequence; the data arrival time and data size are divided into training set and test set.

[0118] Based on LSTM neural network, the data arrival time and data size of different users are predicted; Figure 2 As shown, the details are as follows: Set the neural network training parameters: set the input data dimension to K, representing K users; set the number of training times to c; set the learning rate η; set the activation function to the sigmoid function.

[0119] The test set and training set are respectively input into the LSTM neural network for training and prediction. When training the network of user data packet arrival time interval, the random matrix I K×N Divided into training set I K×Tr , test set I K×Te , Tr, Te are all less than N. Then the training set I K×Tr 1 to K rows and 1 to Tr-1 columns are used as training input, alternating one time step, that is, I K×Tr The 1 to K rows and 2 to Tr columns of are used as training outputs.

[0120] When training the user data packet size, the random matrix S K×N Divide into training set S K×Tr , test set S K×Te , Tr, Te are all less than N. Then the training set S K×Tr 1 to K rows and 1 to Tr-1 columns are used as training input, alternating one time step, i.e. S K×Tr The 1 to K rows and 2 to Tr columns of are used as training outputs.

[0121] According to the trained LSTM network model, the test set data can be used to make traffic predictions. The prediction steps are as follows: When predicting the time interval between arrival of user data packets, the test set matrix I K×Te The 1 to K rows and 1 to J columns, J < Te, are used as prediction inputs to predict the time interval prediction matrix P (P is a matrix with K rows and 1 column) for the next data packet arrival of K users. If it is assumed that the initial time interval of the first data packet arrival of K users in the test set is 0, then I k,l represents the kth row and lth column element of the kth user (k∈[1,K]). is the arrival time of the next data packet of K users.

[0122] When predicting the size of user data packets, the test set matrix S K×Te The 1 to K rows and 1 to J columns of Te are used as the prediction input, and the next data packet size prediction matrix Q of the K users can be predicted (Q is a matrix of K rows and 1 column), and the next data packet size matrix Q of the K users is obtained.

[0123] According to the predicted data arrival time and size, the subcarrier activation and deactivation algorithm based on Q-Learning is used to allocate different resources to users with different needs. The following steps are included: define the total number of users as K and the total number of subcarriers as M; represent each user as an agent with its CCA instance; define the CCA observation vector of the kth user as C k ; Define the average rate of the kth user in the last CCA interval as Define the user's service quality requirements as

[0124] According to the predicted data arrival size and data arrival time; according to the defined total number of users K, total number of subcarriers M and agent of CCA instance, the action equation and state function of Q-Learning are set, and then the reward function of the corresponding action under the state is calculated according to the action equation and state function; the action with the highest reward is selected, and the total rate of user k is calculated.

[0125] Define the total number of users as K and the total number of subcarriers as M as follows: Assume that there are K users in total, and the network uses carrier aggregation technology CCA to perform carrier aggregation on M subcarriers CC, where each subcarrier contains N resource block groups, and a resource block group consists of multiple resource blocks; where the maximum number of CCs activated by the kth user is M k , including a primary subcarrier PCC and multiple secondary subcarriers SCC; PCC is always activated for users.

[0126] Each user is represented as an agent with its CCA instance, such as Figure 3 As shown, the details are as follows: The CCA observation vector of the kth user is defined as C k =[φ k ,q k ], where φ k =[φ k,1 ,φ k,2 ,…,φ k,M-1 ] represents the CQI vector of the M-1 secondary subcarriers respectively relative to the kth user, q k Indicates the data size; the above parameters are all passed from the scheduler and upper layer modules to the user. According to the CQI vector of each user, the traffic scheduler allocates different numbers of resource block groups to each user.

[0127] like Figure 5 As shown, the action equation defining Q-Learning is as follows:

[0128] Let the action of the kth user be defined as:

[0129] a k =[α k ]=[α k,1 ,α k,2 ,...,α k,m ,...,α k,M-1 ] T

[0130] where α k,m =1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

[0131] The timing and interaction between components of a user are as follows Figure 4 As shown. Where T represents the time interval between data arrivals, T CCA Indicates the trigger interval of the CCA algorithm. The state function formula is as follows:

[0132] s k =[τ k ,l k ,v k ];

[0133] Among them, τ in the state function k Indicates whether the action is applied. The calculation formula is as follows:

[0134]

[0135] in, is the predicted value of the next data arrival time of the kth user, T k,e is the arrival time of the previous data. If the difference between the two variables is less than the trigger interval T of the CCA algorithm, CCA , indicating that there is no need to activate a new SCC; T CCA Indicates the trigger interval of the CCA algorithm;

[0136] l in the state function k Indicates the number of SCCs that need to be activated. The calculation formula is as follows:

[0137]

[0138] in, and q max They represent the predicted value of the next data size of the kth user and the maximum capacity of a subcarrier, respectively, and M represents the total number of subcarriers;

[0139] v in the state function k Indicates whether SCC needs to be activated. The calculation formula is as follows:

[0140]

[0141] in, and They represent the average rate of the kth user in the last CCA interval and the user's quality of service requirement QoS respectively; The calculation of is as follows:

[0142]

[0143] where a k,m When it is 1, it means that the mth CC is activated for the kth user, otherwise it is deactivated. k,m represents the achievable rate of the m-th CC for the k-th user, which is calculated based on the defined φk and the resource block group allocated to the user.

[0144] The reward function formula is as follows: represents the number of SCCs activated by the kth user at the tth iteration, τ k Indicates whether the action is applied, then the reward function of the kth user is defined as follows:

[0145]

[0146] Among them, r k,1 The formula is as follows:

[0147]

[0148] It represents the reward value when an activation action is not applied. The reward value is 1 only when only PCC is activated.

[0149] in denote the number of SCCs activated by the kth user at the t-2th iteration, t-1th iteration, and tth iteration, respectively;

[0150] r k,2 The formula is as follows:

[0151]

[0152] Indicates that when the data size exceeds the PCC capacity, more SCCs are activated; conversely, the throughput and the number of activated SCCs are balanced; ω is a parameter used to control the rate and number of SCCs, which can be set according to the situation, q k is the data size of the kth user, q max is the maximum capacity of the PCC;

[0153] rk,3 The formula is as follows:

[0154]

[0155] represents the reward obtained by the user's average rate relative to the user's required rate, where represents the average rate of the kth user, represents the rate requirement of the kth user.

[0156] Select the action with the highest reward and calculate the total rate formula for user k as follows:

[0157]

[0158] Among them, V k,m (t) represents the achievable rate of the m-th subcarrier for the k-th user, where α k,m (t)=1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

[0159] In order to verify the reliability of the proposed method, the following simulation settings are performed: three users are defined in the simulation, with rates of 4.086kbit / s, 700kbit / s, and 1200kbit / s respectively. In the simulation, the arrival time and arrival size of the data predicted by LSTM are used as input to train the combination of the number of CCs to be activated and the user throughput when each data arrives. A parameter ω is set as an indicator to measure the throughput and the number of activated CCs. The larger ω is, the more the system focuses on throughput. While meeting the user rate requirements, use as few CCs as possible to transmit data to save energy. In addition, the feasibility of the system is evaluated by comparing it with the method of activating only one CC and activating all CCs.

[0160] Figure 6 is the average number of CCs activated by each user under different CC activation methods. It can be seen that activating all CCs requires the largest number of CCs, which means that the most energy is consumed. Figure 7 Figure 1 is a schematic diagram of the rate requirements and throughput of different users in different modes. It can be seen that the present invention has different throughputs according to different user rate requirements, but all of them are based on the premise of meeting the user rate requirements. Although activating all CCs will have the highest throughput, Figure 6 As shown in Figure 2, this method also consumes the most energy. Although activating a CC consumes less energy, sometimes the throughput may not meet the user's rate requirements, such as in user 2 and user 3. Figure 8The delay of different users to transmit all data under different modes is shown. It can be seen that as the user rate requirement increases, the data transmission delay of the method proposed in this paper will decrease. Similarly, although activating all CCs will have the lowest delay, it will also consume the most energy.

[0161] The embodiment of the present invention further provides a traffic adaptive carrier aggregation and resource scheduling system for a 230 MHz power licensed frequency band, based on a 230 MHz power dedicated spectrum, including:

[0162] Acquisition module: used to obtain the data arrival time and data size of different users; the specific details of obtaining the data arrival time and data size of different users are as follows: construct the arrival time intervals of K users and N historical data packets of each user into a K×N random matrix I K×N , where K is the number of rows in the matrix, representing K users; N is the length of the time interval sequence of the user's data packets arriving at the destination; K users and each user's N historical data packet sizes are constructed into a K×N random matrix S K×N , where K is the number of rows in the matrix, representing K users; N is the length of the user's data packet size sequence; the data arrival time and data size are divided into training set and test set.

[0163] Prediction module: used to predict the data arrival time and data size of different users based on the LSTM neural network; the specific prediction of the data arrival time and data size of different users based on the LSTM neural network is as follows: set the neural network training parameters; input the test set and training set into the LSTM neural network for training and prediction respectively.

[0164] Allocation module: used to allocate different resources to users with different needs according to the predicted data arrival time and size using the Q-Learning-based subcarrier activation and deactivation algorithm. Using the Q-Learning-based subcarrier activation and deactivation algorithm to allocate different resources to users with different needs includes the following steps: According to the predicted data arrival time and size, using the Q-Learning-based subcarrier activation and deactivation algorithm to allocate different resources to users with different needs. It includes the following steps: define the total number of users as K and the total number of subcarriers as M; represent each user as an agent with its CCA instance: define the CCA observation vector of the kth user as C k ; Define the average rate of the kth user in the last CCA interval as Define the user's service quality requirements as

[0165] According to the predicted data arrival size and data arrival time; according to the defined total number of users K, total number of subcarriers M and agent of CCA instance, the action equation and state function of Q-Learning are set, and then the reward function of the corresponding action under the state is calculated according to the action equation and state function; the action with the highest reward is selected, and the total rate of user k is calculated.

[0166] The total number of users is defined as K, and the total number of subcarriers is defined as M. The specific details are as follows: Assume that there are K users in total, and the network uses carrier aggregation technology CCA to perform carrier aggregation on M subcarriers CC, where each subcarrier contains N resource block groups, and a resource block group consists of multiple resource blocks; the maximum number of CCs activated by the kth user is M k , including a primary subcarrier PCC and multiple secondary subcarriers SCC; PCC is always activated for users.

[0167] Each user is represented as an agent with its CCA instance as follows: The CCA observation vector of the kth user is defined as C k =[φ k ,q k ], where φ k =[φ k,1 ,φ k,2 ,…,φ k,M-1 ] represents the CQI vector of the M-1 secondary subcarriers respectively relative to the kth user, q k Indicates the data size; according to the CQI vector of each user, the traffic scheduler allocates a different number of resource block groups to each user.

[0168] In the allocation module, the action equations for Q-Learning are defined as follows:

[0169] Let the action of the kth user be defined as:

[0170] a k =[α k ]=[α k,1 ,α k,2 ,...,α k,m ,...,α k,M-1 ] T

[0171] where α k,m =1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

[0172] The state function formula is as follows:

[0173] s k =[τ k ,l k ,v k];

[0174] Among them, τ in the state function k Indicates whether the action is applied. The calculation formula is as follows:

[0175]

[0176] in, is the predicted value of the next data arrival time of the kth user, T k,e is the arrival time of the previous data. If the difference between the two variables is less than the trigger interval T of the CCA algorithm, CCA , indicating that there is no need to activate a new SCC; T CCA Indicates the trigger interval of the CCA algorithm;

[0177] l in the state function k Indicates the number of SCCs that need to be activated. The calculation formula is as follows:

[0178]

[0179] in, and q max They represent the predicted value of the next data size of the kth user and the maximum capacity of a subcarrier, respectively, and M represents the total number of subcarriers;

[0180] v in the state function k Indicates whether SCC needs to be activated. The calculation formula is as follows:

[0181]

[0182] in, and They represent the average rate of the kth user in the last CCA interval and the user's quality of service requirement QoS respectively; The calculation of is as follows:

[0183]

[0184] where a k,m When it is 1, it means that the mth CC is activated for the kth user, otherwise it is deactivated. k,m represents the achievable rate of the m-th CC for the k-th user, which is calculated based on the defined φk and the resource block group allocated to the user.

[0185] The reward function formula is as follows: represents the number of SCCs activated by the kth user at the tth iteration, τ k Indicates whether the action is applied, then the reward function of the kth user is defined as follows:

[0186]

[0187] Among them, r k,1 The formula is as follows:

[0188]

[0189] It represents the reward value when an activation action is not applied. The reward value is 1 only when only PCC is activated.

[0190] in denote the number of SCCs activated by the kth user at the t-2th iteration, t-1th iteration, and tth iteration, respectively;

[0191] r k,2 The formula is as follows:

[0192]

[0193] Indicates that when the data size exceeds the PCC capacity, more SCCs are activated; conversely, the throughput and the number of activated SCCs are balanced; ω is a parameter used to control the rate and number of SCCs, which can be set according to the situation, q k is the data size of the kth user, q max is the maximum capacity of the PCC;

[0194] r k,3 The formula is as follows:

[0195]

[0196] represents the reward obtained by the user's average rate relative to the user's required rate, where represents the average rate of the kth user, represents the rate requirement of the kth user.

[0197] Select the action with the highest reward and calculate the total rate formula for user k as follows:

[0198]

[0199] Among them, V k,m (t) represents the achievable rate of the m-th subcarrier for the k-th user, where αk,m(t)=1 indicates that the m-th CC is activated for the k-th user, otherwise it is deactivated.

[0200] An embodiment of the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the computer program implements any one of the traffic adaptive carrier aggregation and resource scheduling methods for the 230 MHz power authorization frequency band.

[0201] An embodiment of the present invention also provides a storage medium storing a computer program, and when the computer program is executed by a processor, it implements any one of the traffic adaptive carrier aggregation and resource scheduling methods for the 230 MHz power authorization frequency band.

Claims

1. A traffic adaptive carrier aggregation and resource scheduling method for the 230MHz power licensed frequency band, characterized in that: Based on the 230MHz power-only spectrum, the process includes the following steps: Obtain data arrival time and data size of different users; Predict the data arrival time and data size of different users based on LSTM neural network; According to the predicted data arrival time and size, a Q-Learning-based subcarrier activation and deactivation algorithm is used to allocate different resources to users with different needs.

2. According to claim 1, a method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band is characterized in that: The data arrival time and data size of different users are obtained as follows: The N historical data packet arrival time intervals of each user among the K users are constructed as a K×N random matrix I K×N , where K is the number of rows in the matrix, representing K users; N is the length of the time interval sequence of the user's data packets; The N historical data packet sizes of each user among the K users are constructed as a K×N random matrix S K×N , where K is the number of rows in the matrix, representing K users; N is the length of the user's data packet size sequence; The random matrix I K×N , S K×N , divide the training set and the test set.

3. According to claim 1, a method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band is characterized in that: The data arrival time and data size of different users are predicted based on the LSTM neural network as follows: Set neural network training parameters; The test set and training set are input into the LSTM neural network for training and prediction respectively.

4. According to claim 1, a method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band is characterized in that: Using the Q-Learning-based subcarrier activation and deactivation algorithm, different resources are allocated to users with different needs, including the following steps: Define the total number of users as K and the total number of subcarriers as M; represent each user as an agent with its CCA instance: define the CCA observation vector of the kth user as C k ; Define the average rate of the kth user in the last CCA interval as Define the user's service quality requirements as According to the predicted data arrival size and data arrival time; according to the defined total number of users K, total number of subcarriers M and agent of CCA instance, the action equation and state function of Q-Learning are set, and then the reward function of the corresponding action under the state is calculated according to the action equation and state function; the action with the highest reward is selected, and the total rate of user k is calculated.

5. The method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band according to claim 4, characterized in that: The total number of users is defined as K and the total number of subcarriers is defined as M as follows: Assume that there are K users in total. The network uses carrier aggregation technology CCA to perform carrier aggregation on M subcarriers CC, where each subcarrier contains N resource block groups, and a resource block group consists of multiple resource blocks; the maximum number of CCs activated by the kth user is M k , including a primary subcarrier PCC and multiple secondary subcarriers SCC; PCC is always activated for users.

6. The method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band according to claim 4, characterized in that: Each user is represented as an agent with its CCA instance as follows: The CCA observation vector of the kth user is defined as C k =[φ k ,q k ], where φ k =[φ k,1 ,φ k,2 ,…,φ k,M-1 ] represents the CQI vector of the M-1 secondary subcarriers respectively relative to the kth user, q k Indicates the data size; According to the CQI vector of each user, the traffic scheduler allocates different numbers of resource block groups to each user.

7. A method for traffic adaptive carrier aggregation and resource scheduling for a 230 MHz power authorization frequency band according to claim 4, characterized in that: The action equations that define Q-Learning are as follows: Let the action of the kth user be defined as: a k ＝[α k ]＝[α k,1 ,α k,2 ,...,α k,m ,...,α k,M-1 ] T where α k,m =1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

8. The method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band according to claim 4, characterized in that: The state function formula is as follows: s k =[τ k ,l k ,v k ]; Among them, τ in the state function k Indicates whether the action is applied. The calculation formula is as follows: in, is the predicted value of the next data arrival time of the kth user, T k,e is the arrival time of the previous data. If the difference between the two variables is less than the trigger interval T of the CCA algorithm, CCA , indicating that there is no need to activate a new SCC; T CCA Indicates the trigger interval of the CCA algorithm; l in the state function k Indicates the number of SCCs that need to be activated. The calculation formula is as follows: in, and q max They represent the predicted value of the next data size of the kth user and the maximum capacity of a subcarrier, respectively, and M represents the total number of subcarriers; v in the state function k Indicates whether SCC needs to be activated. The calculation formula is as follows: in, and They represent the average rate of the kth user in the last CCA interval and the user's quality of service requirement QoS respectively; The calculation of is as follows: where a k,m When it is 1, it means that the mth CC is activated for the kth user, otherwise it is deactivated. k,m represents the achievable rate of the m-th CC for the k-th user, which is calculated based on the defined φk and the resource block group allocated to the user.

9. The method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band according to claim 4, characterized in that: The reward function formula is as follows: represents the number of SCCs activated by the kth user at the tth iteration, τ k Indicates whether the action is applied, then the reward function of the kth user is defined as follows: Among them, r k,1 The formula is as follows: It represents the reward value when an activation action is not applied. The reward value is 1 only when only PCC is activated. in denote the number of SCCs activated by the kth user at the t-2th iteration, t-1th iteration, and tth iteration, respectively; r k,2 The formula is as follows: Indicates that when the data size exceeds the PCC capacity, more SCCs are activated; conversely, the throughput and the number of activated SCCs are balanced; ω is a parameter used to control the rate and number of SCCs, which can be set according to the situation, q k is the data size of the kth user, q max is the maximum capacity of the PCC; r k,3 The formula is as follows: represents the reward obtained by the user's average rate relative to the user's required rate, where represents the average rate of the kth user, represents the rate requirement of the kth user.

10. The method for traffic adaptive carrier aggregation and resource scheduling for the 230MHz power authorization frequency band according to claim 4, characterized in that: Select the action with the highest reward and calculate the total rate formula for user k as follows: Among them, V k,m (t) represents the achievable rate of the m-th subcarrier for the k-th user, where α k,m (t)=1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

11. A traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band, characterized in that: Based on 230MHz power-only spectrum, including: Acquisition module: used to obtain data arrival time and data size of different users; Prediction module: used to predict the data arrival time and data size of different users based on LSTM neural network; Allocation module: used to allocate different resources to users with different needs based on the predicted data arrival time and size using the Q-Learning-based subcarrier activation and deactivation algorithm.

12. A traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 11, characterized in that: In the acquisition module, the data arrival time and data size of different users are obtained as follows: The N historical data packet arrival time intervals of each user among the K users are constructed as a K×N random matrix I K×N , where K is the number of rows in the matrix, representing K users; N is the length of the time interval sequence of the user's data packets; The N historical data packet sizes of each user among the K users are constructed as a K×N random matrix S K×N , where K is the number of rows in the matrix, representing K users; N is the length of the user's data packet size sequence; The random matrix I K×N , S K×N , divide the training set and the test set.

13. A traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 11, characterized in that: In the prediction module, the data arrival time and data size of different users are predicted based on the LSTM neural network as follows: the neural network training parameters are set; the test set and the training set are input into the LSTM neural network for training and prediction respectively.

14. The traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 11, characterized in that: In the allocation module, the subcarrier activation and deactivation algorithm based on Q-Learning is used to allocate different resources to users with different needs. The details are as follows: Define the total number of users as K and the total number of subcarriers as M; represent each user as an agent with its CCA instance: define the CCA observation vector of the kth user as C k ; Define the average rate of the kth user in the last CCA interval as Define the user's service quality requirements as The action equation and state function of Q-Learning are set according to the predicted data arrival size, data arrival time, and the definition of the agent of the CCA instance. Then, the reward function of the corresponding action under the state is calculated based on the action equation and state function. The action with the highest reward is selected, and the total rate of user k is calculated.

15. The traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 4, characterized in that: In the allocation module, the total number of users is defined as K and the total number of subcarriers is defined as M as follows: Assume that there are K users in total. The network uses carrier aggregation technology CCA to perform carrier aggregation on M subcarriers CC, where each subcarrier contains N resource block groups, and a resource block group consists of multiple resource blocks; the maximum number of CCs activated by the kth user is M k , including a primary subcarrier PCC and multiple secondary subcarriers SCC; PCC is always activated for users.

16. The traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 4, characterized in that: In the allocation module, each user is represented as an agent with its CCA instance as follows: The CCA observation vector of the kth user is defined as C k =[φ k ,q k ], where φ k =[φ k,1 ,φ k,2 ,…,φ k,M-1 ] represents the CQI vector of the M-1 secondary subcarriers respectively relative to the kth user, q k Indicates the data size; According to the CQI vector of each user, the traffic scheduler allocates different numbers of resource block groups to each user.

17. A traffic adaptive carrier aggregation and resource scheduling system for 230MHz power licensed frequency band according to claim 14, characterized in that: In the allocation module, the action equations for Q-Learning are defined as follows: Let the action of the kth user be defined as: a k ＝[α k ]＝[α k,1 ,α k,2 ,...,α k,m ,...,α k,M-1 ] T where α k,m =1 indicates that the mth CC is activated for the kth user, otherwise it is deactivated.

18. The traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 14, characterized in that: In the allocation module, the state function formula is as follows: s k =[τ k ,l k ,v k ]; Among them, τ in the state function k Indicates whether the action is applied. The calculation formula is as follows: in, is the predicted value of the next data arrival time of the kth user, T k,e is the arrival time of the previous data. If the difference between the two variables is less than the trigger interval T of the CCA algorithm, CCA , indicating that there is no need to activate a new SCC; T CCA Indicates the trigger interval of the CCA algorithm; l in the state function k Indicates the number of SCCs that need to be activated. The calculation formula is as follows: in, and q max They represent the predicted value of the next data size of the kth user and the maximum capacity of a subcarrier, respectively, and M represents the total number of subcarriers; v in the state function k Indicates whether SCC needs to be activated. The calculation formula is as follows: in, and They represent the average rate of the kth user in the last CCA interval and the user's quality of service requirement QoS respectively; The calculation of is as follows: where a k,m When it is 1, it means that the mth CC is activated for the kth user, otherwise it is deactivated. k,m represents the achievable rate of the m-th CC for the k-th user, which is calculated based on the defined φk and the resource block group allocated to the user.

19. The traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 14, characterized in that: In the allocation module, the reward function formula is as follows: represents the number of SCCs activated by the kth user at the tth iteration, τ k Indicates whether the action is applied, then the reward function of the kth user is defined as follows: Among them, r k,1 The formula is as follows: It represents the reward value when an activation action is not applied. The reward value is 1 only when only PCC is activated. in denote the number of SCCs activated by the kth user at the t-2th iteration, t-1th iteration, and tth iteration, respectively; r k,2 The formula is as follows: Indicates that when the data size exceeds the PCC capacity, more SCCs are activated; conversely, the throughput and the number of activated SCCs are balanced; ω is a parameter used to control the rate and number of SCCs, which can be set according to the situation, q k is the data size of the kth user, q max is the maximum capacity of the PCC; r k,3 The formula is as follows: represents the reward obtained by the user's average rate relative to the user's required rate, where represents the average rate of the kth user, represents the rate requirement of the kth user.

20. The traffic adaptive carrier aggregation and resource scheduling system for the 230MHz power licensed frequency band according to claim 14, characterized in that: In the allocation module, the action with the highest reward is selected, and the total rate formula for user k is calculated as follows: Among them, V k,m (t) represents the achievable rate of the m-th subcarrier for the k-th user, where αk,m(t)=1 indicates that the m-th CC is activated for the k-th user, otherwise it is deactivated.

21. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into the processor, it implements the traffic adaptive carrier aggregation and resource scheduling method for the 230 MHz power authorization frequency band according to any one of claims 1 to 10.

22. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a traffic adaptive carrier aggregation and resource scheduling method for a 230 MHz power authorization frequency band is implemented according to any one of claims 1 to 10.

Images