Processing method and device for planning short message retransmission strategy based on network awareness state

Abstract

The embodiment of the invention relates to a processing method and device for planning a short message retransmission strategy based on a network awareness state. The method comprises the following steps: carding a state type set S, defining a cost index C, and configuring a strategy parameter set and a strategy parameter value domain DP corresponding to the strategy parameter set; constructing an index prediction model and a strategy prediction model; constructing a first data set; training an index prediction model based on the first data set; optimizing the optimal strategy configuration corresponding to each state sample by taking the minimum cost index as an optimization target, and constructing a second data set based on all the state samples and the corresponding optimal strategies; training a strategy prediction model based on the second data set; and the strategy prediction model and the index prediction model dynamically plan the short message retransmission strategy of the server according to the real-time network state of the short message sending server and predict the cost index of the planning strategy. According to the invention, the configuration flexibility of the short message retransmission strategy can be improved, and the cost control can be optimized.

Description

A method and apparatus for processing SMS retransmission strategies based on network-aware state planning Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a processing method and apparatus for planning SMS resending strategies based on network-aware state. Background Technology

[0002] SMS retransmission strategies are a crucial mechanism for SMS sending servers to ensure reliable message delivery. Currently, most mainstream SMS retransmission strategies employ statically configured exponential backoff algorithms. Their core parameters (base backoff time b, multiplier m, maximum backoff time M, and maximum number of retries R) are typically set once by operations personnel based on experience and remain unchanged over long-term operation. Practical application reveals a significant technical flaw in this static configuration approach: it cannot dynamically plan SMS retransmission strategies based on real-time network conditions for cost control purposes. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a method, apparatus, electronic device, and computer-readable storage medium for planning SMS retransmission strategies based on network-aware state. This invention organizes the set of state types S that the SMS sending server can perceive, defines the server's cost index C within a preset statistical period, sets the strategy parameter set corresponding to the exponential backoff strategy parameters, and configures the corresponding strategy parameter value range D for the strategy parameter set. P Two deep learning models were constructed: a cost indicator prediction model to predict cost indicators based on network state and policy characteristics, and a policy prediction model to predict optimal policy parameters based on network state characteristics. Within a preset sample collection period, the SMS retransmission policy parameters of the server were periodically configured according to a preset statistical cycle (each time starting from the policy parameter value range D). PA random set of policy parameters is sampled as the current configuration. Within each statistical period, state samples are collected for the initial network state of the current period, policy configuration samples are collected for the SMS retransmission policy parameters of the current period, and cost indicator samples are statistically analyzed for the current period. A first dataset is constructed based on all the obtained samples. An indicator prediction model is trained based on the first dataset. After the indicator prediction model is completed, the optimal policy configuration corresponding to each state sample is optimized based on a preset optimal parameter optimization algorithm and the indicator prediction model, with the goal of minimizing the cost indicator. A second dataset is constructed based on all state samples and their corresponding optimal policy configurations. A policy prediction model is trained based on the second dataset. After the policy prediction model is trained, the policy prediction model and the indicator prediction model dynamically plan the SMS retransmission policy of the server based on the real-time network state of the SMS sending server and predict the cost indicator of the planned policy. This invention enables dynamic planning of SMS retransmission policies based on real-time network state for cost control purposes and can predict the cost indicator corresponding to each planning result. This invention can improve the configuration flexibility of SMS retransmission policies and optimize cost control.

[0004] To achieve the above objectives, a first aspect of the present invention provides a processing method for planning SMS retransmission strategies based on network-aware state. The method includes: sorting through a set of network state types that the SMS sending server can perceive to obtain a state type set S; defining an index C for the SMS sending cost of the SMS sending server within a preset statistical period; setting a set of strategy parameters for the SMS retransmission strategy based on exponential backoff strategy parameters; and setting corresponding numerical constraint ranges, denoted as corresponding numerical ranges D, for the base backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R of the strategy parameter set. b D m D M D R ; and consists of four categories of numerical ranges D b D m D M D R The range D of the four-dimensional strategy parameters P Each specific type in the state type set S is denoted as the corresponding type s. i 1 ≤ index i ≤ N S N S The total number of preset status types; the cost index C is the economic cost caused by the total number of resent SMS messages within the preset statistical period, C=c SMS ×N SMS c SMS For the economic cost of a single text message, N SMSThe total number of resent SMS messages; the policy parameter set includes the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R, where b and M > 0, and m and R ≥ 1; a cost indicator prediction model is constructed to predict cost indicators based on network state and policy characteristics; and a policy prediction model is constructed to predict optimal policy parameters based on network state characteristics; the cost indicator prediction model is used to predict cost indicators based on the state sequence X and policy parameter sequence P input to the model and output the corresponding predicted cost indicator C. pre The state sequence X includes N S State x i The state x i With the type s i One-to-one correspondence; the policy parameter sequence P includes 4 policy parameters p j 1 ≤ index j ≤ 4; 4 policy parameters p j The base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R correspond one-to-one; the policy prediction model is used to determine the optimal policy parameters based on the state sequence X input to the model and output the corresponding predicted policy parameter sequence P. pre The prediction strategy parameter sequence P pre Includes 4 prediction parameters ; 4 prediction parameters The basic backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R correspond one-to-one; within a preset sample collection period, the SMS retransmission strategy parameters of the SMS sending server are periodically configured according to the preset statistical period, and each configuration is performed from the strategy parameter value range D. P A set of policy parameters is randomly sampled as the current configuration. Within each statistical period, state samples are collected for the initial network state of the current period, policy configuration samples are collected for the SMS retransmission policy parameters of the current period, and cost indicator samples of the current period are statistically analyzed. A first dataset is constructed based on all the obtained samples. The sample collection time is longer than a preset statistical period. The indicator prediction model is trained based on the first dataset. After the indicator prediction model is completed, the optimal policy configuration corresponding to each state sample is optimized based on a preset optimal parameter optimization algorithm and the indicator prediction model, with the goal of minimizing the cost indicator. A second dataset is constructed based on all state samples and their corresponding optimal policy configurations. The policy prediction model is trained based on the second dataset. After the policy prediction model is trained, the policy prediction model and the indicator prediction model dynamically plan the SMS retransmission policy of the server according to the real-time network state of the SMS sending server and predict the cost indicator of the planned policy.

[0005] Preferably, the N of the state type set S S The types s i The dataset is divided into at least three major subsets: a subset of types for SMS gateway connections, a subset of types for SMS recipients, and a subset of types for network quality. The subset of types for SMS gateway connections includes at least the gateway's connection status and the number of reconnections within the gateway's most recent duration L1. The connection status includes at least normal, disconnected, timed out, and authentication failure. L1 is a preset time length. The subset of types for SMS recipients includes at least the delivery success rate of all recipients within the most recent duration L1 and the offline percentage of all recipients within the most recent duration L1. The subset of types for network quality includes at least the server-to-gateway bandwidth, the average RTT of the server-to-gateway within the most recent duration L1, the average packet loss rate of the server-to-gateway within the most recent duration L1, and the gateway rate limiting status. The gateway rate limiting status includes rate limiting and no rate limiting. The first dataset includes multiple first data records. Each first data record includes the state sequence X, the policy parameter sequence P, and the cost tag C. tag The second dataset includes multiple second data records; the second data record includes the state sequence X and the policy parameter label P. tag The strategy parameter label P tag Includes 4 tag parameters ; 4 of the aforementioned tag parameters Each corresponds one-to-one with the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R.

[0006] Preferably, the index prediction model is composed of a first embedding coding layer, a first feature coding network, and a first linear layer connected sequentially; the first embedding coding layer is used to process each state x of the state sequence X input to the model. i The data type is identified, if the current state x i If the data type is categorical, then for the current state x i One-hot vector encoding is performed to obtain the corresponding embedding code. If the current state x i If the data type is numerical, then the current state x is normalized using the Z-score method. i Normalization encoding is performed to obtain the corresponding embedding encoding. ; and normalize each policy parameter p of the policy parameter sequence P input to the model using the Z-score normalization method. j Normalization encoding is performed to obtain the corresponding embedding encoding. ; and from the obtained N S+4 embedding codes The corresponding embedding vector H0 is sequentially concatenated and sent to the first feature encoding network; where 1 ≤ index k ≤ N S +4; The first feature encoding network is composed of LA first linear activation layers connected sequentially, where LA is the preset total number of first network layers; the first feature encoding network is used to perform feature encoding on the embedding vector H0 through the built-in LA first linear activation layers to obtain the corresponding feature vector H. LA Send to the first linear layer; wherein, the calculation method of the first linear activation layer of the u-th layer is as follows: ; 1 ≤ layer index u ≤ LA; H u-1 H u These are the input and output vectors of the first linear activation layer in the u-th layer, respectively; when u=1, H u-1 =H0, when u=LA, H LA =H u W u b u These are the weight matrix and bias vector corresponding to the first linear activation layer of the u-th layer, respectively; ReLU() is the ReLU activation function; the first linear layer is used to apply the feature vector H... LA The corresponding predicted cost index C is obtained by performing linear regression calculation on the cost index. pre Wherein, the predicted cost index C pre The calculation method is as follows: W C b C These are the weight matrix and bias vector corresponding to the first linear layer, respectively.

[0007] Preferably, the policy prediction model is composed of a second embedding coding layer, a second feature coding network, a second linear layer, a Softplus activation layer, and a rounding output layer connected sequentially; the second embedding coding layer is used to process each state x of the state sequence X input to the model. i The data type is identified, if the current state x i If the data type is categorical, then for the current state x i One-hot vector encoding is performed to obtain the corresponding embedding code. If the current state x i If the data type is numerical, then the current state x is normalized using the Z-score method. i Normalization encoding is performed to obtain the corresponding embedding encoding. ; and from the obtained N S Embedded encoding The embedding vector Z0 is sequentially concatenated and sent to the second feature encoding network. The second feature encoding network is composed of LB second linear activation layers connected sequentially, where LB is the preset total number of second network layers. The second feature encoding network is used to perform feature encoding on the embedding vector Z0 through the built-in LB second linear activation layers to obtain the corresponding feature vector Z. LB Send to the second linear layer; wherein, the calculation method for the second linear activation layer of the g-th layer is as follows: ; 1 ≤ layer index g ≤ LB; Z g-1 Z g These are the input and output vectors of the second linear activation layer in the g-th layer, respectively; when g=1, Z g-1 =Z0, g=LB, Z LB =Z g W g b g These are the weight matrix and bias vector corresponding to the second linear activation layer of the g-th layer, respectively; ReLU() is the ReLU activation function; the second linear layer is used to process the feature vector Z. LB A mapped feature vector Y of length 4 is obtained by mapping the feature vectors in the policy feature space and sent to the Softplus activation layer; wherein, the mapped feature vector Y is calculated as follows: W proj b proj These are the weight matrix and bias vector corresponding to the second linear layer, respectively; the mapped feature vector Y includes four feature scalars y. j The Softplus activation layer is used to apply the Softplus activation function to each of the feature scalars y in the mapped feature vector Y. j Activation calculations are performed to obtain the corresponding activation parameters. And from the four activation parameters obtained Form the corresponding activation vector P * Send to the rounding output layer; wherein, the four activation parameters The counting method is as follows: , ; 4 activation parameters The base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R in the policy parameter set correspond one-to-one; the rounding output layer is used to process the activation vector P. * Each of the activation parameters The result of the rounding is used as the corresponding prediction parameter. ; and by all the predicted parameters obtained The corresponding prediction strategy parameter sequence P is formed. pre .

[0008] Preferably, the step of collecting state samples of the initial network state of the current period, collecting policy configuration samples of SMS retransmission policy parameters of the current period, and statistically analyzing cost index samples of the current period within each statistical period, and constructing a first dataset based on all the obtained samples, specifically includes: at the beginning of each statistical period during the sample collection period, based on the state type set S, analyzing the N of the SMS sending server. S The type s described in the class i The real-time status information is collected to obtain the corresponding N. S The states x i And from the obtained N S The states x i A corresponding state sequence X is formed; and data is collected on the SMS retransmission policy parameters randomly configured by the SMS sending server for the current period to obtain a corresponding policy parameter sequence P; at the end of each statistical period during the sample collection period, the total number N of retransmitted SMS messages generated by the SMS sending server in the current period is calculated. SMS Statistical analysis was performed, and the economic cost of a single text message was calculated based on the preset value c. SMS And the total number N of resent text messages currently being counted. SMS Calculate the corresponding cost tag C tag =c SMS ×N SMS The state sequence X, the strategy parameter sequence P, and the cost label C are all defined for each statistical period. tag A corresponding first data record is formed; and all the obtained first data records form the corresponding first dataset.

[0009] Preferably, training the indicator prediction model based on the first dataset specifically includes: step 61, randomly dividing the first dataset into two subsets, designated as the first training set and the first evaluation set, based on a preset first partitioning ratio; wherein both the first training set and the first evaluation set consist of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first partitioning ratio; and the cost label C of each first data record in the first training set... tag Record as the corresponding 1 ≤ index q ≤ N tr1 N tr1 The total number of records in the first training set; the cost label C of each of the first data records in the first evaluation set. tag Record as the corresponding 1 ≤ index f ≤ N av1 Nav1 The total number of records in the first evaluation set; Step 62, input the state sequence X and the policy parameter sequence P of each of the first data records in the first training set into the indicator prediction model for prediction, and output the prediction cost indicator C from this prediction. pre Record as the corresponding ; and based on various predicted cost indicators and its corresponding cost label Form a corresponding first prediction-label pair; Step 63, obtain N tr1 Substituting the first prediction-label pair into the preset first model loss function L M1 The first loss value is obtained by calculation; wherein, the first model loss function L M1 for: Step 64: Identify whether the first loss value meets the preset first loss value range; if it does, proceed to step 65; if not, based on the preset first model optimizer, move towards making the first model loss function L... M1 The direction that reaches the minimum value modulates the model parameters of the indicator prediction model in one round, and returns to step 62 at the end of this round of adjustment; wherein, the first model optimizer includes the Adam optimizer and the SGD optimizer; step 65, the state sequence X and the policy parameter sequence P of each of the first data records of the first evaluation set are input into the indicator prediction model for prediction, and the prediction cost index C output by this prediction is used for prediction. pre Record as the corresponding ; and based on various predicted cost indicators and its corresponding cost label Form a corresponding second prediction-label pair; and obtain N av1 The second prediction-label pair is substituted into the preset first model evaluation function F. M1 The calculation yields the corresponding first evaluation value; wherein, the first model evaluation function F M1 for: Step 66: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 61; if it does, stop training and confirm that the indicator prediction model training has ended.

[0010] Preferably, the step of optimizing the optimal strategy configuration for each state sample based on the preset optimal parameter optimization algorithm and the indicator prediction model with the goal of minimizing the cost index, and constructing a second dataset based on all state samples and their corresponding optimal strategy configurations, specifically includes: Step 71, denoting each state sequence X in the first dataset as the corresponding X. rWhere 1 ≤ index r ≤ N1, and N1 is the total number of records in the first dataset; Step 72, according to each state sequence X r and the range of the strategy parameter D P Construct the optimization objective function: Where fM1() is the model function of the indicator prediction model; For the state sequence X r The corresponding optimal policy parameter sequence; the optimization objective function represents the state sequence X given the state sequence X. r In the case of the policy parameter range D P The optimal strategy parameter sequence that minimizes the prediction cost index of the index prediction model. Perform the solution; Step 73, based on the optimal parameter optimization algorithm and the optimization objective function, solve each of the state sequences X. r The corresponding optimal strategy parameter sequence The solution is performed; the optimal parameter optimization algorithm includes a global grid search algorithm and a Bayesian optimization algorithm; step 74, each of the state sequences X is solved. r and the corresponding optimal strategy parameter sequence As a corresponding set of state sequence X and policy parameter label P tag And form a corresponding second data record; and form the corresponding second dataset from all the obtained second data records.

[0011] Preferably, training the policy prediction model based on the second dataset specifically includes: step 81, randomly dividing the second dataset into two subsets based on a preset second partitioning ratio, denoted as the second training set and the second evaluation set; wherein, both the second training set and the second evaluation set consist of multiple second data records; the ratio of the total number of records in the second training set to the total number of records in the second evaluation set satisfies the second partitioning ratio; the policy parameter label P of each second data record in the second training set... tag Record as the corresponding Each strategy parameter label The four tag parameters Record as the corresponding 1 ≤ index o ≤ N tr2 N tr2 The total number of records in the second training set; the policy parameter label P of each of the second data records in the second evaluation set. tag Record as the corresponding Each strategy parameter label The four tag parameters Record as the corresponding 1 ≤ index t ≤ N av2 N av2 The total number of records in the second evaluation set; Step 82, input the state sequence X of each of the second data records in the second training set into the policy prediction model for prediction, and output the prediction policy parameter sequence P. pre Record as the corresponding and the current prediction strategy parameter sequence The four prediction parameters Record as the corresponding ; and composed of the parameter sequences of each prediction strategy and their corresponding strategy parameter labels Form a corresponding third prediction-label pair; Step 83, obtain N tr2 The third prediction-label pair is substituted into the preset second model loss function L. M2 The corresponding second loss value is obtained through calculation; where the second model loss function L M2 for: Step 84: Identify whether the second loss value meets the preset second loss value range; if it does, proceed to step 85; if not, based on the preset second model optimizer, move towards making the second model loss function L... M2 The direction that reaches the minimum value modulates the model parameters of the policy prediction model in one round, and returns to step 82 at the end of this round of adjustment; wherein, the second model optimizer includes the Adam optimizer and the SGD optimizer; step 85, the state sequence X of each of the second data records of the second evaluation set is input into the policy prediction model for prediction, and the predicted policy parameter sequence P output by this prediction is used for prediction. pre Record as the corresponding and the current prediction strategy parameter sequence The four prediction parameters Record as the corresponding ; and composed of the parameter sequences of each prediction strategy and their corresponding strategy parameter labels Form a corresponding fourth prediction-label pair; and obtain N av2 The fourth prediction-label pair is substituted into the preset second model evaluation function F. M2 The calculation yields the corresponding second evaluation value; wherein, the second model evaluation function F M2 for: Step 86: Identify whether the second evaluation value meets the preset range of the second evaluation value; if not, return to step 81; if it does, stop training and confirm that the training of the policy prediction model has ended.

[0012] Preferably, the step of dynamically planning the SMS retransmission strategy of the server based on the real-time network status of the SMS sending server and predicting the cost index of the planned strategy by the strategy prediction model and the index prediction model specifically includes: using the preset statistical period as the first duration; and every first duration, calculating the N values ​​of the SMS sending server at the current time. S The type s described in the class i The real-time status information is collected in one round, and N is obtained from this round of collection. S The states x i The current state sequence X is formed; the current state sequence X is input into the policy prediction model for prediction, and the predicted policy parameter sequence output by this prediction is used as the current policy parameter sequence P; the current state sequence X and the current policy parameter sequence P are input into the indicator prediction model for prediction, and the predicted cost indicator output by this prediction is used as the current cost indicator C. now ; and set the SMS retransmission strategy parameters of the SMS sending server based on the current strategy parameter sequence P; and based on the current time, the current state sequence X, the current strategy parameter sequence P, and the current cost index C. now Create a corresponding strategy-cost history record and save it.

[0013] A second aspect of this invention provides an apparatus for implementing the processing method of planning SMS retransmission strategy based on network-aware state as described in the first aspect. The apparatus includes: a preprocessing module, a model building module, a first dataset building module, an indicator prediction model training module, a second dataset building module, a strategy prediction model training module, and a model application module. The preprocessing module is used to sort through the set of network state types that the SMS sending server can perceive to obtain a state type set S; define the SMS sending cost of the SMS sending server within a preset statistical period to obtain a corresponding cost indicator C; set the strategy parameter set of the SMS retransmission strategy based on exponential backoff strategy parameters; and set corresponding numerical constraint ranges, denoted as corresponding numerical ranges D, for the basic backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R of the strategy parameter set. b D m D M D R ; and consists of four categories of numerical ranges D b D m D M D R The range D of the four-dimensional strategy parameters P Each specific type in the state type set S is denoted as the corresponding type s. i 1 ≤ index i ≤ N S NS The total number of preset status types; the cost index C is the economic cost caused by the total number of resent SMS messages within the preset statistical period, C=c SMS ×N SMS c SMS For the economic cost of a single text message, N SMS The total number of resent SMS messages; the policy parameter set includes the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R, where b and M > 0, and m and R ≥ 1; the model building module is used to build an indicator prediction model for predicting cost indicators based on network state and policy characteristics; and to build a policy prediction model for predicting optimal policy parameters based on network state characteristics; the indicator prediction model is used to predict cost indicators based on the state sequence X and policy parameter sequence P input to the model and output the corresponding predicted cost indicator C. pre The state sequence X includes N S State x i The state x i With the type s i One-to-one correspondence; the policy parameter sequence P includes 4 policy parameters p j 1 ≤ index j ≤ 4; 4 policy parameters p j The base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R correspond one-to-one; the policy prediction model is used to determine the optimal policy parameters based on the state sequence X input to the model and output the corresponding predicted policy parameter sequence P. pre The prediction strategy parameter sequence P pre Includes 4 prediction parameters ; 4 prediction parameters The first dataset construction module is used to periodically configure the SMS retransmission strategy parameters of the SMS sending server according to the preset statistical period within a preset sample collection duration, and to select from the strategy parameter value range D each time during configuration. This configuration corresponds one-to-one with the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R. PA set of policy parameters is randomly sampled as the current configuration; and within each statistical period, state samples are collected for the initial network state of the current period, policy configuration samples are collected for the SMS retransmission policy parameters of the current period, and cost index samples of the current period are statistically analyzed. A first dataset is constructed based on all the obtained samples; the sample collection time is longer than a preset statistical period; the indicator prediction model training module trains the indicator prediction model based on the first dataset; the second dataset construction module is used to optimize the optimal policy configuration corresponding to each state sample based on a preset optimal parameter optimization algorithm and the indicator prediction model after the indicator prediction model is completed, with the goal of minimizing the cost index, and constructs a second dataset based on all state samples and their corresponding optimal policy configurations; the policy prediction model training module trains the policy prediction model based on the second dataset; the model application module is used to dynamically plan the SMS retransmission policy of the server based on the real-time network state of the SMS sending server and predict the cost index of the planned policy after the policy prediction model is completed.

[0014] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver; the processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the steps of the method described in the first aspect; the transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the instructions described in the first aspect.

[0016] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for planning SMS retransmission strategies based on network-aware state. As described above, this invention organizes the set of state types S that the SMS sending server can perceive, defines the server's cost index C within a preset statistical period, sets the strategy parameter set corresponding to the exponential backoff strategy parameters, and configures the corresponding strategy parameter value range D for the strategy parameter set. P Two deep learning models were constructed: a cost indicator prediction model to predict cost indicators based on network state and policy characteristics, and a policy prediction model to predict optimal policy parameters based on network state characteristics. Within a preset sample collection period, the SMS retransmission policy parameters of the server were periodically configured according to a preset statistical cycle (each time starting from the policy parameter value range D). PA random set of policy parameters is sampled as the current configuration. Within each statistical period, state samples are collected for the initial network state of the current period, policy configuration samples are collected for the SMS retransmission policy parameters of the current period, and cost indicator samples of the current period are statistically analyzed. A first dataset is constructed based on all the obtained samples. An indicator prediction model is trained based on the first dataset. After the indicator prediction model is completed, the optimal policy configuration corresponding to each state sample is optimized based on a preset optimal parameter optimization algorithm and the indicator prediction model, with the goal of minimizing the cost indicator. A second dataset is constructed based on all state samples and their corresponding optimal policy configurations. A policy prediction model is trained based on the second dataset. After the policy prediction model is trained, the policy prediction model and the indicator prediction model dynamically plan the SMS retransmission policy of the server based on the real-time network state of the SMS sending server and predict the cost indicator of the planned policy. This embodiment of the invention can dynamically plan the SMS retransmission policy based on the real-time network state for the purpose of cost control, and can predict the cost indicator corresponding to each planning result; through this embodiment of the invention, the configuration flexibility of the SMS retransmission policy can be improved, and cost control can be optimized. Attached Figure Description

[0017] Figure 1 is a schematic diagram of a processing method for planning SMS retransmission strategy based on network-aware state provided in Embodiment 1 of the present invention; Figure 2 is a block diagram of the indicator prediction model provided in Embodiment 1 of the present invention; Figure 3 is a block diagram of the strategy prediction model provided in Embodiment 1 of the present invention; Figure 4 is a block diagram of a processing device for planning SMS retransmission strategy based on network-aware state provided in Embodiment 2 of the present invention; Figure 5 is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] This invention provides a method for planning SMS retransmission strategies based on network-aware state, as shown in Figure 1. The method mainly includes the following steps: Step 1, sorting out the set of network state types that the SMS sending server can perceive to obtain a state type set S; defining the SMS sending cost of the SMS sending server within a preset statistical period to obtain the corresponding cost index C; setting the strategy parameter set for the SMS retransmission strategy based on the exponential backoff strategy parameters; and setting corresponding numerical constraint ranges for the basic backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R of the strategy parameter set, denoted as the corresponding numerical range D. b D m D M D R ; and consists of four categories of numerical ranges D b D m D M D R The range D of the four-dimensional strategy parameters P .

[0020] Here, the preset statistical period in this embodiment of the invention is a pre-set time length parameter.

[0021] Each specific type in the state type set S of this embodiment of the invention is denoted as the corresponding type s. i 1 ≤ index i ≤ N S N S This represents the preset total number of state types. It should be noted that N is the number of state type sets S. S Types s i The types are divided into at least three major subsets: a subset of types related to SMS gateway connections, a subset of types related to SMS recipients, and a subset of types related to network quality. The subset of types related to SMS gateway connections includes at least the following specific types: gateway connection status and the number of reconnections within the gateway's most recent duration L1. Connection status includes at least normal, disconnected, timed out, and authentication request failed; L1 is a preset time length. The subset of types related to SMS recipients includes at least the following specific types: delivery success rate of all recipients within the most recent duration L1 and the offline percentage of all recipients within the most recent duration L1. The subset of types related to network quality includes at least the following specific types: server-to-gateway bandwidth, average round-trip time (RTT) between server and gateway within the most recent duration L1, average packet loss rate between server and gateway within the most recent duration L1, and gateway rate limiting status; gateway rate limiting status includes rate limiting and no rate limiting.

[0022] In this embodiment of the invention, the cost index C is the economic cost caused by the total number of resent SMS messages within a preset statistical period, C=c SMS ×N SMS c SMS For the economic cost of a single text message, N SMS This represents the total number of resent text messages.

[0023] The strategy parameter set in this embodiment of the invention includes a base backoff time b, a multiplier factor m, a maximum backoff time M, and a maximum number of retries R, wherein b, m, M, and R are all integers, and b and M > 0, and m and R ≥ 1.

[0024] Step 2: Construct an indicator prediction model for predicting cost indicators based on network state and policy characteristics; and construct a policy prediction model for predicting optimal policy parameters based on network state characteristics.

[0025] Here, the indicator prediction model of this invention is used to predict cost indicators based on the state sequence X and the strategy parameter sequence P input to the model, and output the corresponding predicted cost indicator C. pre .

[0026] The state sequence X includes N S State x i State x i With type s i One-to-one correspondence.

[0027] The policy parameter sequence P includes four policy parameters p. j 1 ≤ index j ≤ 4; 4 strategy parameters p j It corresponds one-to-one with the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R.

[0028] Figure 2 shows the module structure diagram of the index prediction model provided in Embodiment 1 of the present invention. The index prediction model is composed of a first embedding coding layer, a first feature coding network, and a first linear layer connected sequentially. Wherein: 1) First embedding coding layer: The first embedding coding layer is used to process each state x of the state sequence X input to the model. i The data type is identified, if the current state x i If the data type is categorical, then for the current state x i One-hot vector encoding is performed to obtain the corresponding embedding code. If the current state x i If the data type is numerical, then the current state x is normalized using the Z-score method. i Normalization encoding is performed to obtain the corresponding embedding encoding. ; and normalize each policy parameter p of the policy parameter sequence P input to the model using the Z-score normalization method. jNormalization encoding is performed to obtain the corresponding embedding encoding. ; and from the obtained N S +4 embedding codes The corresponding embedding vector H0 is sequentially concatenated and sent to the first feature encoding network.

[0029] Where 1 ≤ index k ≤ N S +4.

[0030] 2) First feature coding network: The first feature coding network is composed of LA first linear activation layers connected sequentially, where LA is the preset total number of first network layers.

[0031] The first feature encoding network is used to encode the embedded vector H0 using LA first linear activation layers to obtain the corresponding feature vector H. LA Send to the first linear layer.

[0032] Here, the calculation method of the first linear activation layer of the u-th layer of the first feature coding network in this embodiment of the invention is as follows: Where 1 ≤ layer index u ≤ LA; H u-1 H u These are the input and output vectors of the first linear activation layer of the u-th layer, respectively; when u=1, H u-1 =H0, when u=LA, H LA =H u W u b u , respectively, are the weight matrix and bias vector corresponding to the first linear activation layer of the u-th layer; ReLU() is the ReLU activation function.

[0033] 3) First linear layer: The first linear layer is used to determine the feature vector H. LA The corresponding predicted cost index C is obtained by performing linear regression calculation on the cost index. pre .

[0034] Here, the predicted cost index C in this embodiment of the invention pre The calculation method is as follows: Among them, W C b C These are the weight matrix and bias vector corresponding to the first linear layer, respectively.

[0035] The policy prediction model of this invention is used to determine the optimal policy parameters based on the state sequence X input to the model and output the corresponding predicted policy parameter sequence P. pre .

[0036] Prediction strategy parameter sequence P pre Includes 4 prediction parameters 4 prediction parameters It corresponds one-to-one with the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R.

[0037] Figure 3 shows the module structure diagram of the policy prediction model provided in Embodiment 1 of the present invention. The policy prediction model is composed of a second embedding coding layer, a second feature coding network, a second linear layer, a Softplus activation layer, and a rounding output layer connected in sequence. Among them: 1) Second embedding coding layer: The second embedding coding layer is used to process each state x of the state sequence X input to the model. i The data type is identified, if the current state x i If the data type is categorical, then for the current state x i One-hot vector encoding is performed to obtain the corresponding embedding code. If the current state x i If the data type is numerical, then the current state x is normalized using the Z-score method. i Normalization encoding is performed to obtain the corresponding embedding encoding. ; and from the obtained N S Embedded encoding The corresponding embedding vector Z0 is sequentially concatenated and sent to the second feature encoding network.

[0038] 2) Second feature encoding network: The second feature encoding network is composed of LB second linear activation layers connected sequentially, where LB is the preset total number of second network layers.

[0039] The second feature encoding network is used to encode the embedding vector Z0 using LB second linear activation layers to obtain the corresponding feature vector Z. LB Send to the second linear layer.

[0040] Here, the calculation method of the second linear activation layer of the g-th layer of the second feature coding network in this embodiment of the invention is as follows: Where 1 ≤ level index g ≤ LB; Z g-1 Z g These are the input and output vectors of the second linear activation layer in the g-th layer, respectively; when g=1, Z g-1 =Z0, g=LB, Z LB =Z g W g b g These are the weight matrix and bias vector corresponding to the second linear activation layer of the g-th layer, respectively.

[0041] 3) Second linear layer: The second linear layer is used to process the feature vector Z LB The feature vectors in the policy feature space are mapped to obtain a mapped feature vector Y with a length of 4, which is then sent to the Softplus activation layer.

[0042] Here, the method for calculating the mapping feature vector Y in this embodiment of the invention is as follows: Among them, W proj b proj These are the weight matrix and bias vector corresponding to the second linear layer, respectively.

[0043] It should be noted that the mapping feature vector Y in this embodiment of the invention includes four feature scalars y. j .

[0044] 4) Softplus activation layer: The Softplus activation layer is used to apply the Softplus activation function to each feature scalar y of the mapped feature vector Y. j Activation calculations are performed to obtain the corresponding activation parameters. And from the four activation parameters obtained Form the corresponding activation vector P * Send to the rounding output layer.

[0045] Here, the four activation parameters of this invention embodiment The counting method is as follows: , .

[0046] 4 activation parameters Each parameter corresponds one-to-one with the base backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R in the strategy parameter set.

[0047] 5) Rounding Output Layer: The rounding output layer is used to round the activation vector P. * Each activation parameter Perform rounding and use the rounded result as the corresponding prediction parameter. ; and from all the predicted parameters obtained The corresponding prediction strategy parameter sequence P is formed. pre .

[0048] Step 3: Within the preset sample collection period, periodically configure the SMS retransmission strategy parameters of the SMS sending server according to the preset statistical period, and each time configure, select the strategy parameter value range D. P A set of policy parameters is randomly sampled as the current configuration; and in each statistical period, the initial network state of the current period is sampled, the policy configuration of the SMS retransmission policy parameters of the current period is sampled, and the cost index samples of the current period are statistically analyzed. Based on all the obtained samples, the first dataset is constructed.

[0049] Specifically, this includes: Step 31, within a preset sample collection period, periodically configuring the SMS retransmission strategy parameters of the SMS sending server according to a preset statistical period, and retransmitting the parameters from the strategy parameter value range D each time during configuration.P A set of policy parameters is randomly sampled as the current configuration.

[0050] Here, the sample collection duration in this embodiment of the invention is a preset time length parameter, where the sample collection duration is greater than the preset statistical period.

[0051] Step 32: In each statistical period, collect state samples of the initial network state of the current period, collect policy configuration samples of SMS retransmission policy parameters of the current period, and statistically analyze cost index samples of the current period, and construct the first dataset based on all the obtained samples.

[0052] Specifically, this includes: Step 321, at the beginning of each statistical period during the sample collection period, based on the state type set S, the N of the SMS sending server... S Class type s i The real-time status information is collected to obtain the corresponding N. S State x i And from the obtained N S State x i A corresponding state sequence X is formed; and data is collected on the SMS retransmission policy parameters randomly configured by the SMS sending server for the current period to obtain a corresponding policy parameter sequence P.

[0053] Step 322: At the end of each statistical period during the sample collection period, calculate the total number N of retransmitted SMS messages generated by the SMS sending server in the current period. SMS Statistical analysis was conducted, based on a preset economic cost per SMS message (c). SMS And the current total number of resent text messages N SMS Calculate the corresponding cost label C tag =c SMS ×N SMS .

[0054] Step 323 involves a set of state sequences X, strategy parameter sequences P, and cost labels C corresponding to each statistical period. tag A corresponding first data record is formed; and all the obtained first data records form the corresponding first dataset.

[0055] Here, the first dataset in this embodiment of the invention includes multiple first data records; the first data records include a state sequence X, a policy parameter sequence P, and a cost label C. tag .

[0056] Step 4: Train the metric prediction model based on the first dataset.

[0057] Specifically, it includes: Step 41, randomly dividing the first dataset into two sub-datasets based on a preset first segmentation ratio, denoted as the corresponding first training set and first evaluation set.

[0058] Here, the first segmentation ratio in this embodiment of the invention is a preset ratio parameter, such as 8:2. Both the first training set and the first evaluation set consist of multiple first data records. The ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio.

[0059] It should be noted that the cost label C of each first data record in the first training set... tag Record as the corresponding 1 ≤ index q ≤ N tr1 N tr1 This represents the total number of records in the first training set.

[0060] It should also be noted that the cost label C of each first data record in the first evaluation set tag Record as the corresponding 1 ≤ index f ≤ N av1 N av1 The total number of records in the first evaluation set.

[0061] Step 42: Input the state sequence X and policy parameter sequence P of each first data record in the first training set into the index prediction model for prediction, and output the prediction cost index C. pre Record as the corresponding ; and based on various predicted cost indicators and its corresponding cost label Form a corresponding first prediction-label pair.

[0062] Step 43, obtain N tr1 Each first prediction-label pair is substituted into the preset first model loss function L. M1 The corresponding first loss value is obtained through calculation.

[0063] Here, the first model loss function L in this embodiment of the invention M1 for: .

[0064] Step 44: Identify whether the first loss value meets the preset first loss value range; if it does, proceed to step 45; if not, based on the preset first model optimizer, move towards making the first model loss function L... M1 The direction that reaches the minimum value modulates the model parameters of the indicator prediction model in one round, and returns to step 42 when the adjustment is completed.

[0065] Here, the first loss value range in this embodiment of the invention is a pre-set numerical range. The first model optimizer includes the Adam optimizer and the SGD optimizer.

[0066] Step 45: Input the state sequence X and policy parameter sequence P of each first data record in the first evaluation set into the index prediction model for prediction, and output the prediction cost index C. pre Record as the corresponding ; and based on various predicted cost indicators and its corresponding cost label Form a corresponding second prediction-label pair; and obtain N av1 Each second prediction-label pair is substituted into the preset first model evaluation function F. M1 The corresponding first evaluation value is obtained through calculation.

[0067] Here, the first model evaluation function F in this embodiment of the invention M1 for: .

[0068] Step 46: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 41; if it does, stop training and confirm that the indicator prediction model training has ended.

[0069] Here, the first evaluation value range in this embodiment of the invention is a pre-set numerical range.

[0070] Step 5: After the indicator prediction model is completed, the optimal strategy configuration corresponding to each state sample is optimized based on the preset optimal parameter optimization algorithm and indicator prediction model, with the goal of minimizing the cost indicator. Then, a second dataset is constructed based on all state samples and their corresponding optimal strategy configurations.

[0071] Specifically, this includes: Step 51, recording each state sequence X of the first dataset as the corresponding X. r .

[0072] Where 1 ≤ index r ≤ N1, and N1 is the total number of records in the first dataset.

[0073] Step 52, based on each state sequence X r and the range of strategy parameters D P Construct an optimization objective function.

[0074] Here, the optimization objective function of this embodiment of the invention is: Where fM1() is the model function of the indicator prediction model; For the state sequence X r The corresponding optimal strategy parameter sequence.

[0075] The above optimization objective function represents: given a state sequence X r In the case of the policy parameter range D P The optimal strategy parameter sequence that minimizes the prediction cost index of the indicator prediction model. Solve the problem.

[0076] Step 53: Based on the optimal parameter optimization algorithm and the optimization objective function, optimize each state sequence X. r The corresponding optimal policy parameter sequence Solve the problem.

[0077] Here, the optimal parameter optimization algorithm in this embodiment of the invention includes a global grid search algorithm and a Bayesian optimization algorithm.

[0078] Step 54, convert each state sequence X r and its corresponding optimal policy parameter sequence As a corresponding set of state sequence X and policy parameter label P tag And form a corresponding second data record; and form a corresponding second dataset from all the obtained second data records.

[0079] Here, the second dataset in this embodiment of the invention includes multiple second data records. Each second data record includes a state sequence X and a policy parameter label P. tag Strategy parameter label P tag Includes 4 tag parameters ; 4 tag parameters It corresponds one-to-one with the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R.

[0080] Step 6: Train the policy prediction model based on the second dataset.

[0081] Specifically, it includes: Step 61, randomly dividing the second dataset into two sub-datasets based on a preset second segmentation ratio, denoted as the corresponding second training set and second evaluation set.

[0082] Here, the second segmentation ratio in this embodiment of the invention is a pre-set ratio parameter, such as 8:2. Both the second training set and the second evaluation set consist of multiple second data records. The ratio of the total number of records in the second training set to the total number of records in the second evaluation set satisfies the second segmentation ratio.

[0083] It should be noted that the policy parameter label P of each second data record in the second training set... tag Record as the corresponding Each strategy parameter label The four tag parameters Record as the corresponding 1 ≤ index o ≤ N tr2 N tr2 This represents the total number of records in the second training set.

[0084] It should also be noted that the policy parameter label P of each second data record in the second evaluation set... tag Record as the corresponding Each strategy parameter label The four tag parameters Record as the corresponding 1 ≤ index t ≤ N av2 N av2 This represents the total number of records in the second evaluation set.

[0085] Step 62: Input the state sequence X of each second data record in the second training set into the policy prediction model for prediction, and output the prediction policy parameter sequence P. pre Record as the corresponding and the current prediction strategy parameter sequence The four prediction parameters Record as the corresponding ; and composed of the parameter sequences of each prediction strategy and their corresponding strategy parameter labels This forms a corresponding third prediction-label pair.

[0086] Step 63, obtain N tr2 Each third prediction-label pair is substituted into the preset second model loss function L. M2 The corresponding second loss value is obtained through calculation.

[0087] Here, the second model loss function L in this embodiment of the invention M2 for: .

[0088] Step 64: Identify whether the second loss value meets the preset range of the second loss value; if it does, proceed to step 65; if not, based on the preset second model optimizer, move towards making the second model loss function L... M2 The direction that reaches the minimum value modulates the model parameters of the policy prediction model in one round, and returns to step 62 when the adjustment is completed.

[0089] Here, the second loss value range in this embodiment of the invention is a pre-set numerical range. The second model optimizer includes the Adam optimizer and the SGD optimizer.

[0090] Step 65: Input the state sequence X of each second data record in the second evaluation set into the policy prediction model for prediction, and output the prediction policy parameter sequence P. pre Record as the corresponding and the current prediction strategy parameter sequence The four prediction parameters Record as the corresponding ; and composed of the parameter sequences of each prediction strategy and their corresponding strategy parameter labels Form a corresponding fourth prediction-label pair; and obtain N av2 The fourth prediction-label pair is substituted into the preset second model evaluation function F. M2 The corresponding second evaluation value is obtained through calculation.

[0091] Here, the second model evaluation function F in this embodiment of the invention M2 for: .

[0092] Step 66: Identify whether the second evaluation value meets the preset range of the second evaluation value; if not, return to step 61; if it does, stop training and confirm that the training of the policy prediction model has ended.

[0093] Here, the second evaluation value range in this embodiment of the invention is a pre-set numerical range.

[0094] Step 7: After the strategy prediction model is trained, the strategy prediction model and the indicator prediction model dynamically plan the SMS retransmission strategy of the server based on the real-time network status of the SMS sending server and predict the cost indicators of the planned strategy.

[0095] Specifically, this includes: using a preset statistical period as the first duration; and every first duration, retrieving N data from the SMS sending server at the current time. S Class type s i The real-time status information is collected in one round, and N is obtained from this round of collection. S State x i The current state sequence X is constructed; the current state sequence X is input into the policy prediction model for prediction, and the predicted policy parameter sequence output by this prediction is used as the current policy parameter sequence P; the current state sequence X and the current policy parameter sequence P are input into the index prediction model for prediction, and the predicted cost index output by this prediction is used as the current cost index C. now ; and set the SMS retransmission policy parameters of the SMS sending server based on the current policy parameter sequence P; and based on the current time, current state sequence X, current policy parameter sequence P, and current cost index C. now Create a corresponding strategy-cost history record and save it.

[0096] Figure 4 is a block diagram of a processing device for planning SMS retransmission strategies based on network-aware state, provided in Embodiment 2 of the present invention. This device can be a terminal device or server implementing the aforementioned method embodiments, or it can be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiments. For example, the device can be a device or chip system of the aforementioned terminal device or server. As shown in Figure 4, the device includes: a preprocessing module 201, a model building module 202, a first dataset building module 203, an indicator prediction model training module 204, a second dataset building module 205, a strategy prediction model training module 206, and a model application module 207.

[0097] The preprocessing module 201 is used to sort out the set of network state types that the SMS sending server can perceive to obtain a state type set S; define the SMS sending cost of the SMS sending server within a preset statistical period to obtain the corresponding cost index C; set the strategy parameter set of the SMS retransmission strategy based on the exponential backoff strategy parameters; and set the corresponding numerical constraint ranges for the basic backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R of the strategy parameter set, denoted as the corresponding numerical range D. b D m D M D R ; and consists of four categories of numerical ranges D b D m D M D R The range D of the four-dimensional strategy parameters P Each specific type in the state type set S is denoted as the corresponding type s. i 1 ≤ index i ≤ N S N S The total number of preset status types; the cost index C is the economic cost caused by the total number of resent SMS messages within the preset statistical period, C=c SMS ×N SMS c SMS For the economic cost of a single text message, N SMS The total number of resent SMS messages; the strategy parameter set includes the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R, where b and M > 0, and m and R ≥ 1.

[0098] Model building module 202 is used to build an indicator prediction model for predicting cost indicators based on network state and policy characteristics; and to build a policy prediction model for predicting optimal policy parameters based on network state characteristics; the indicator prediction model is used to predict cost indicators based on the state sequence X and policy parameter sequence P input to the model and output the corresponding predicted cost indicator C. pre The state sequence X includes N S State xi State x i With type s i One-to-one correspondence; the policy parameter sequence P includes 4 policy parameters p j 1 ≤ index j ≤ 4; 4 strategy parameters p j The base backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R correspond one-to-one; the policy prediction model is used to determine the optimal policy parameters based on the state sequence X input to the model and output the corresponding predicted policy parameter sequence P. pre Prediction strategy parameter sequence P pre Includes 4 prediction parameters 4 prediction parameters It corresponds one-to-one with the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R.

[0099] The first dataset construction module 203 is used to periodically configure the SMS retransmission strategy parameters of the SMS sending server according to a preset statistical period within a preset sample collection duration, and to retrieve the strategy parameter value range D each time during configuration. P A set of policy parameters is randomly sampled as the current configuration; and in each statistical period, the initial network state of the current period is sampled, the policy configuration of the SMS retransmission policy parameters of the current period is sampled, and the cost index samples of the current period are statistically analyzed. The first dataset is constructed based on all the obtained samples; the sample collection time is greater than the preset statistical period.

[0100] The indicator prediction model training module 204 trains the indicator prediction model based on the first dataset.

[0101] The second dataset construction module 205 is used to optimize the optimal strategy configuration corresponding to each state sample based on the preset optimal parameter optimization algorithm and the indicator prediction model after the indicator prediction model is completed, with the goal of minimizing the cost indicator, and to construct the second dataset based on all state samples and their corresponding optimal strategy configurations.

[0102] The policy prediction model training module 206 trains the policy prediction model based on the second dataset.

[0103] The model application module 207 is used to dynamically plan the SMS retransmission strategy of the server based on the real-time network status of the SMS sending server after the strategy prediction model training is completed, and to predict the cost index of the planned strategy.

[0104] The processing device for planning SMS retransmission strategies based on network-aware state provided in this embodiment of the invention can execute the method steps in the above method embodiment. Its implementation principle and technical effect are similar, and will not be repeated here.

[0105] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, the preprocessing module can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and called and executed by a processing element of the device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0106] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-on-a-Chip (SOC).

[0107] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the foregoing method embodiments are generated. The computer described above can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The aforementioned computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the aforementioned computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, Bluetooth, microwave, etc.) means. The aforementioned computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The aforementioned available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0108] Figure 5 is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. This electronic device can be a terminal device or server implementing the methods of the aforementioned embodiments, or it can be a terminal device or server connected to the aforementioned terminal device or server implementing the methods of the aforementioned embodiments. As shown in Figure 5, the electronic device may include: a processor 301 (e.g., CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transmission and reception operations of the transceiver 303. The memory 302 may store various instructions for completing various processing functions and implementing the processing steps described in the methods of the aforementioned embodiments. Preferably, the electronic device involved in the embodiments of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The communication port 306 is used for communication between the electronic device and other peripherals.

[0109] The system bus 305 mentioned in Figure 5 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include Non-Volatile Memory, such as at least one disk storage device.

[0110] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), graphics processing units (GPUs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0111] It should be noted that the embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to perform the methods and processes provided in the above embodiments.

[0112] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for planning SMS retransmission strategies based on network-aware state. As described above, this invention organizes the set of state types S that the SMS sending server can perceive, defines the server's cost index C within a preset statistical period, sets the strategy parameter set corresponding to the exponential backoff strategy parameters, and configures the corresponding strategy parameter value range D for the strategy parameter set. P Two deep learning models were constructed: a cost indicator prediction model to predict cost indicators based on network state and policy characteristics, and a policy prediction model to predict optimal policy parameters based on network state characteristics. Within a preset sample collection period, the SMS retransmission policy parameters of the server were periodically configured according to a preset statistical cycle (each time starting from the policy parameter value range D). PA random set of policy parameters is sampled as the current configuration. Within each statistical period, state samples are collected for the initial network state of the current period, policy configuration samples are collected for the SMS retransmission policy parameters of the current period, and cost indicator samples of the current period are statistically analyzed. A first dataset is constructed based on all the obtained samples. An indicator prediction model is trained based on the first dataset. After the indicator prediction model is completed, the optimal policy configuration corresponding to each state sample is optimized based on a preset optimal parameter optimization algorithm and the indicator prediction model, with the goal of minimizing the cost indicator. A second dataset is constructed based on all state samples and their corresponding optimal policy configurations. A policy prediction model is trained based on the second dataset. After the policy prediction model is trained, the policy prediction model and the indicator prediction model dynamically plan the SMS retransmission policy of the server based on the real-time network state of the SMS sending server and predict the cost indicator of the planned policy. This embodiment of the invention can dynamically plan the SMS retransmission policy based on the real-time network state for the purpose of cost control, and can predict the cost indicator corresponding to each planning result; through this embodiment of the invention, the configuration flexibility of the SMS retransmission policy can be improved, and cost control can be optimized.

[0113] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0114] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A processing method for planning SMS retransmission strategies based on network-aware state, characterized in that, The method includes: sorting out the set of network state types that the SMS sending server can perceive to obtain a state type set S; defining the SMS sending cost of the SMS sending server within a preset statistical period to obtain a corresponding cost index C; setting the strategy parameter set of the SMS retransmission strategy based on the exponential backoff strategy parameters; and setting corresponding numerical constraint ranges for the basic backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R of the strategy parameter set, denoted as the corresponding numerical range D. b D m D M D R ; and consists of four categories of numerical ranges D b D m D M D R The range D of the four-dimensional strategy parameters P Each specific type in the state type set S is denoted as the corresponding type s. i 1 ≤ index i ≤ N S N S The total number of preset status types; the cost index C is the economic cost caused by the total number of resent SMS messages within the preset statistical period, C=c SMS ×N SMS c SMS For the economic cost of a single text message, N SMS The total number of resent SMS messages; the policy parameter set includes the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R, where b and M > 0, and m and R ≥ 1; a cost indicator prediction model is constructed to predict cost indicators based on network state and policy characteristics; and a policy prediction model is constructed to predict optimal policy parameters based on network state characteristics; the cost indicator prediction model is used to predict cost indicators based on the state sequence X and policy parameter sequence P input to the model and output the corresponding predicted cost indicator C. pre The state sequence X includes N S State x i The state x i With the type s i One-to-one correspondence; the policy parameter sequence P includes 4 policy parameters p j 1 ≤ index j ≤ 4; 4 policy parameters p j The base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R correspond one-to-one; the policy prediction model is used to determine the optimal policy parameters based on the state sequence X input to the model and output the corresponding predicted policy parameter sequence P. pre The prediction strategy parameter sequence P pre Includes 4 prediction parameters ; 4 prediction parameters The basic backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R correspond one-to-one; within a preset sample collection period, the SMS retransmission strategy parameters of the SMS sending server are periodically configured according to the preset statistical period, and each configuration is performed from the strategy parameter value range D. P A set of policy parameters is randomly sampled as the current configuration. Within each statistical period, state samples are collected for the initial network state of the current period, policy configuration samples are collected for the SMS retransmission policy parameters of the current period, and cost indicator samples of the current period are statistically analyzed. A first dataset is constructed based on all the obtained samples. The sample collection time is longer than a preset statistical period. The indicator prediction model is trained based on the first dataset. After the indicator prediction model is completed, the optimal policy configuration corresponding to each state sample is optimized based on a preset optimal parameter optimization algorithm and the indicator prediction model, with the goal of minimizing the cost indicator. A second dataset is constructed based on all state samples and their corresponding optimal policy configurations. The policy prediction model is trained based on the second dataset. After the policy prediction model is trained, the policy prediction model and the indicator prediction model dynamically plan the SMS retransmission policy of the server according to the real-time network state of the SMS sending server and predict the cost indicator of the planned policy.

2. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 1, characterized in that, The state type set S has N S The types s i The dataset is divided into at least three major subsets: a subset of types for SMS gateway connections, a subset of types for SMS recipients, and a subset of types for network quality. The subset of types for SMS gateway connections includes at least the gateway's connection status and the number of reconnections within the gateway's most recent duration L1. The connection status includes at least normal, disconnected, timed out, and authentication failure. L1 is a preset time length. The subset of types for SMS recipients includes at least the delivery success rate of all recipients within the most recent duration L1 and the offline percentage of all recipients within the most recent duration L1. The subset of types for network quality includes at least the server-to-gateway bandwidth, the average RTT of the server-to-gateway within the most recent duration L1, the average packet loss rate of the server-to-gateway within the most recent duration L1, and the gateway rate limiting status. The gateway rate limiting status includes rate limiting and no rate limiting. The first dataset includes multiple first data records. Each first data record includes the state sequence X, the policy parameter sequence P, and the cost tag C. tag ; The second dataset includes multiple second data records; the second data records include the state sequence X and the policy parameter label P. tag ; The strategy parameter label P tag Includes 4 tag parameters ; 4 of the aforementioned tag parameters Each of the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R corresponds to one another.

3. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 1, characterized in that, The index prediction model is composed of a first embedding coding layer, a first feature coding network, and a first linear layer connected sequentially; the first embedding coding layer is used to process each state x of the state sequence X input to the model. i The data type is identified, if the current state x i If the data type is categorical, then for the current state x i One-hot vector encoding is performed to obtain the corresponding embedding code. If the current state x i If the data type is numerical, then the current state x is normalized according to the Z-score method. i Normalization encoding is performed to obtain the corresponding embedding encoding. ; and normalize each policy parameter p of the policy parameter sequence P input to the model using the Z-score normalization method. j Normalization encoding is performed to obtain the corresponding embedding encoding. ; and from the obtained N S +4 embedding codes The corresponding embedding vector H0 is sequentially concatenated and sent to the first feature encoding network; where 1 ≤ index k ≤ N S +4; The first feature encoding network is composed of LA first linear activation layers connected sequentially, where LA is the preset total number of first network layers; the first feature encoding network is used to perform feature encoding on the embedding vector H0 through the built-in LA first linear activation layers to obtain the corresponding feature vector H. LA Send to the first linear layer; wherein, the calculation method of the first linear activation layer of the u-th layer is as follows: ; 1 ≤ layer index u ≤ LA; H u-1 H u These are the input and output vectors of the first linear activation layer in the u-th layer, respectively; when u=1, H u-1 =H0, when u=LA, H LA =H u W u b u These are the weight matrix and bias vector corresponding to the first linear activation layer of the u-th layer, respectively; ReLU() is the ReLU activation function; the first linear layer is used to apply the feature vector H... LA The corresponding predicted cost index C is obtained by performing linear regression calculation on the cost index. pre Wherein, the predicted cost index C pre The calculation method is as follows: W C b C These are the weight matrix and bias vector corresponding to the first linear layer, respectively.

4. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 1, characterized in that, The policy prediction model is composed of a second embedding coding layer, a second feature coding network, a second linear layer, a Softplus activation layer, and a rounding output layer connected sequentially; the second embedding coding layer is used to process each state x of the state sequence X input to the model. i The data type is identified, if the current state x i If the data type is categorical, then for the current state x i One-hot vector encoding is performed to obtain the corresponding embedding code. If the current state x i If the data type is numerical, then the current state x is normalized using the Z-score method. i Normalization encoding is performed to obtain the corresponding embedding encoding. ; and from the obtained N S Embedded encoding The embedding vector Z0 is sequentially concatenated and sent to the second feature encoding network. The second feature encoding network is composed of LB second linear activation layers connected sequentially, where LB is the preset total number of second network layers. The second feature encoding network is used to perform feature encoding on the embedding vector Z0 through the built-in LB second linear activation layers to obtain the corresponding feature vector Z. LB Send to the second linear layer; wherein, the calculation method for the second linear activation layer of the g-th layer is as follows: ; 1 ≤ layer index g ≤ LB; Z g-1 Z g These are the input and output vectors of the second linear activation layer in the g-th layer, respectively; when g=1, Z g-1 =Z0, g=LB, Z LB =Z g W g b g These are the weight matrix and bias vector corresponding to the second linear activation layer of the g-th layer, respectively; ReLU() is the ReLU activation function; the second linear layer is used to process the feature vector Z. LB A mapped feature vector Y of length 4 is obtained by mapping the feature vectors in the policy feature space and sent to the Softplus activation layer; wherein, the mapped feature vector Y is calculated as follows: W proj b proj These are the weight matrix and bias vector corresponding to the second linear layer, respectively; the mapped feature vector Y includes four feature scalars y. j The Softplus activation layer is used to apply the Softplus activation function to each of the feature scalars y in the mapped feature vector Y. j Activation calculations are performed to obtain the corresponding activation parameters. And from the four activation parameters obtained Form the corresponding activation vector P * Send to the rounding output layer; wherein, the four activation parameters The counting method is as follows: ， ; 4 activation parameters The base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R in the policy parameter set correspond one-to-one; the rounding output layer is used to process the activation vector P. * Each of the activation parameters The result of the rounding is used as the corresponding prediction parameter. ; and by all the predicted parameters obtained The corresponding prediction strategy parameter sequence P is formed. pre .

5. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 2, characterized in that, The process involves collecting state samples of the initial network state for the current period, collecting policy configuration samples of SMS retransmission policy parameters for the current period, and statistically analyzing cost indicator samples for the current period within each statistical period. A first dataset is then constructed based on all the obtained samples. Specifically, this includes: at the beginning of each statistical period during the sample collection phase, analyzing the N values ​​of the SMS sending server based on the state type set S. S The type s described in the class i The real-time status information is collected to obtain the corresponding N. S The states x i And from the obtained N S The states x i A corresponding state sequence X is formed; and data is collected on the SMS retransmission policy parameters randomly configured by the SMS sending server for the current period to obtain a corresponding policy parameter sequence P; at the end of each statistical period during the sample collection period, the total number N of retransmitted SMS messages generated by the SMS sending server in the current period is calculated. SMS Statistical analysis was performed, and the economic cost of a single text message was calculated based on the preset value c. SMS And the total number N of resent text messages currently being counted. SMS Calculate the corresponding cost tag C tag =c SMS ×N SMS The state sequence X, the strategy parameter sequence P, and the cost label C are all defined for each statistical period. tag A corresponding first data record is formed; and all the obtained first data records form the corresponding first dataset.

6. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 2, characterized in that, The step of training the indicator prediction model based on the first dataset specifically includes: Step 61, randomly dividing the first dataset into two subsets based on a preset first partitioning ratio, denoted as the first training set and the first evaluation set; wherein, both the first training set and the first evaluation set consist of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first partitioning ratio; the cost label C of each first data record in the first training set... tag Record as the corresponding 1 ≤ index q ≤ N tr1 N tr1 The total number of records in the first training set; the cost label C of each of the first data records in the first evaluation set. tag Record as the corresponding 1 ≤ index f ≤ N av1 N av1 The total number of records in the first evaluation set; Step 62, input the state sequence X and the policy parameter sequence P of each of the first data records in the first training set into the indicator prediction model for prediction, and output the prediction cost indicator C from this prediction. pre Record as the corresponding ; and based on various predicted cost indicators and its corresponding cost label Form a corresponding first prediction-label pair; Step 63, obtain N tr1 Substituting the first prediction-label pair into the preset first model loss function L M1 The first loss value is obtained by calculation; wherein, the first model loss function L M1 for: Step 64: Identify whether the first loss value meets the preset first loss value range; if it does, proceed to step 65; if not, based on the preset first model optimizer, move towards making the first model loss function L... M1 The direction that reaches the minimum value modulates the model parameters of the indicator prediction model in one round, and returns to step 62 at the end of this round of adjustment; wherein, the first model optimizer includes the Adam optimizer and the SGD optimizer; step 65, the state sequence X and the policy parameter sequence P of each of the first data records of the first evaluation set are input into the indicator prediction model for prediction, and the prediction cost index C output by this prediction is used for prediction. pre Record as the corresponding ; and based on various predicted cost indicators and its corresponding cost label Form a corresponding second prediction-label pair; and obtain N av1 The second prediction-label pair is substituted into the preset first model evaluation function F. M1 The calculation yields the corresponding first evaluation value; wherein, the first model evaluation function F M1 for: Step 66: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 61; if yes, stop training and confirm that the indicator prediction model training has ended.

7. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 2, characterized in that, The optimization objective is to minimize the cost index. Based on a preset optimal parameter optimization algorithm and the index prediction model, the optimal strategy configuration corresponding to each state sample is optimized. A second dataset is then constructed based on all state samples and their corresponding optimal strategy configurations. Specifically, this includes: Step 71, denoting each state sequence X in the first dataset as the corresponding X... r Where 1 ≤ index r ≤ N1, and N1 is the total number of records in the first dataset; Step 72, according to each state sequence X r and the range of the strategy parameter D P Construct the optimization objective function: Where fM1() is the model function of the indicator prediction model; For the state sequence X r The corresponding optimal policy parameter sequence; the optimization objective function represents the state sequence X given the state sequence X. r In the case of the policy parameter range D P The optimal strategy parameter sequence that minimizes the prediction cost index of the index prediction model. Perform the solution; Step 73, based on the optimal parameter optimization algorithm and the optimization objective function, solve each of the state sequences X. r The corresponding optimal strategy parameter sequence The solution is performed; the optimal parameter optimization algorithm includes a global grid search algorithm and a Bayesian optimization algorithm; step 74, each of the state sequences X is solved. r and the corresponding optimal strategy parameter sequence As a corresponding set of state sequence X and policy parameter label P tag And form a corresponding second data record; and form the corresponding second dataset from all the obtained second data records.

8. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 2, characterized in that, The step of training the policy prediction model based on the second dataset specifically includes: step 81, randomly dividing the second dataset into two subsets based on a preset second partitioning ratio, denoted as the second training set and the second evaluation set; wherein, both the second training set and the second evaluation set consist of multiple second data records; the ratio of the total number of records in the second training set to the total number of records in the second evaluation set satisfies the second partitioning ratio; and the policy parameter label P of each second data record in the second training set... tag Record as the corresponding Each strategy parameter label The four tag parameters Record as the corresponding 1 ≤ index o ≤ N tr2 N tr2 The total number of records in the second training set; the policy parameter label P of each of the second data records in the second evaluation set. tag Record as the corresponding Each strategy parameter label The four tag parameters Record as the corresponding 1 ≤ index t ≤ N av2 N av2 The total number of records in the second evaluation set; Step 82, input the state sequence X of each of the second data records in the second training set into the policy prediction model for prediction, and output the prediction policy parameter sequence P. pre Record as the corresponding and the current prediction strategy parameter sequence The four prediction parameters Record as the corresponding ; and composed of the parameter sequences of each prediction strategy and their corresponding strategy parameter labels Form a corresponding third prediction-label pair; Step 83, obtain N tr2 The third prediction-label pair is substituted into the preset second model loss function L. M2 The corresponding second loss value is obtained through calculation; where the second model loss function L M2 for: Step 84: Identify whether the second loss value meets the preset second loss value range; if it does, proceed to step 85; if not, based on the preset second model optimizer, move towards making the second model loss function L... M2 The direction that reaches the minimum value modulates the model parameters of the policy prediction model in one round, and returns to step 82 at the end of this round of adjustment; wherein, the second model optimizer includes the Adam optimizer and the SGD optimizer; step 85, the state sequence X of each of the second data records of the second evaluation set is input into the policy prediction model for prediction, and the predicted policy parameter sequence P output by this prediction is used for prediction. pre Record as the corresponding and the current prediction strategy parameter sequence The four prediction parameters Record as the corresponding ; and composed of the parameter sequences of each prediction strategy and their corresponding strategy parameter labels Form a corresponding fourth prediction-label pair; and obtain N av2 The fourth prediction-label pair is substituted into the preset second model evaluation function F. M2 The calculation yields the corresponding second evaluation value; wherein, the second model evaluation function F M2 for: Step 86: Identify whether the second evaluation value meets the preset range of the second evaluation value; if not, return to step 81; if yes, stop training and confirm that the training of the policy prediction model has ended.

9. The processing method for planning SMS retransmission strategies based on network-aware state as described in claim 1, characterized in that, The step of dynamically planning the SMS retransmission strategy of the server based on the real-time network status of the SMS sending server and predicting the cost index of the planned strategy by the strategy prediction model and the index prediction model specifically includes: taking the preset statistical period as the first duration; and every first duration, calculating the N values ​​of the SMS sending server at the current time. S The type s described in the class i The real-time status information is collected in one round, and N is obtained from this round of collection. S The states x i The current state sequence X is formed; the current state sequence X is input into the policy prediction model for prediction, and the predicted policy parameter sequence output by this prediction is used as the current policy parameter sequence P; the current state sequence X and the current policy parameter sequence P are input into the indicator prediction model for prediction, and the predicted cost indicator output by this prediction is used as the current cost indicator C. now ; and set the SMS retransmission strategy parameters of the SMS sending server based on the current strategy parameter sequence P; and based on the current time, the current state sequence X, the current strategy parameter sequence P, and the current cost index C. now Create a corresponding strategy-cost history record and save it.

10. An apparatus for executing the processing method for planning SMS retransmission strategy based on network-aware state as described in any one of claims 1-9, characterized in that, The device includes: a preprocessing module, a model building module, a first dataset building module, an indicator prediction model training module, a second dataset building module, a strategy prediction model training module, and a model application module. The preprocessing module is used to sort through the set of network state types perceptible to the SMS sending server to obtain a state type set S; define the SMS sending cost of the SMS sending server within a preset statistical period to obtain the corresponding cost indicator C; set the strategy parameter set for the SMS retransmission strategy based on the exponential backoff strategy parameters; and set corresponding numerical constraint ranges, denoted as the corresponding numerical ranges D, for the base backoff time b, multiplier factor m, maximum backoff time M, and maximum number of retries R of the strategy parameter set. b D m D M D R ; and consists of four categories of numerical ranges D b D m D M D R The range D of the four-dimensional strategy parameters P Each specific type in the state type set S is denoted as the corresponding type s. i 1 ≤ index i ≤ N S N S The total number of preset status types; the cost index C is the economic cost caused by the total number of resent SMS messages within the preset statistical period, C=c SMS ×N SMS c SMS For the economic cost of a single text message, N SMS The total number of resent SMS messages; the policy parameter set includes the base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R, where b and M > 0, and m and R ≥ 1; the model building module is used to build an indicator prediction model for predicting cost indicators based on network state and policy characteristics; and to build a policy prediction model for predicting optimal policy parameters based on network state characteristics; the indicator prediction model is used to predict cost indicators based on the state sequence X and policy parameter sequence P input to the model and output the corresponding predicted cost indicator C. pre The state sequence X includes N S State x i The state x i With the type s i One-to-one correspondence; the policy parameter sequence P includes 4 policy parameters p j 1 ≤ index j ≤ 4; 4 policy parameters p j The base backoff time b, the multiplier factor m, the maximum backoff time M, and the maximum number of retries R correspond one-to-one; the policy prediction model is used to determine the optimal policy parameters based on the state sequence X input to the model and output the corresponding predicted policy parameter sequence P. pre The prediction strategy parameter sequence P pre Includes 4 prediction parameters ; 4 prediction parameters The first dataset construction module is used to periodically configure the SMS retransmission strategy parameters of the SMS sending server according to the preset statistical period within a preset sample collection duration, and to select from the strategy parameter value range D each time during configuration. P A random sample of the policy parameter set is used as the current configuration; and within each statistical period, state samples are collected for the initial network state of the current period, policy configuration samples are collected for the SMS retransmission policy parameters of the current period, and cost indicator samples of the current period are statistically analyzed, and a first dataset is constructed based on all the obtained samples; the sample collection time is greater than the preset statistical period; the indicator prediction model training module trains the indicator prediction model based on the first dataset; the second dataset construction module is used to optimize the optimal policy configuration corresponding to each state sample based on the preset optimal parameter optimization algorithm and the indicator prediction model after the indicator prediction model is completed, with the goal of minimizing the cost indicator, and construct a second dataset based on all state samples and their corresponding optimal policy configurations; the policy prediction model training module trains the policy prediction model based on the second dataset; the model application module is used to dynamically plan the SMS retransmission policy of the server based on the real-time network state of the SMS sending server and predict the cost indicator of the planned policy after the policy prediction model is completed.

11. An electronic device, characterized in that, include: Memory, processor, and transceiver; The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method according to any one of claims 1-9; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1-9.

Images