An intelligent short message dispatching method and device based on multi-dimension dynamic optimization

Abstract

The application provides a kind of intelligent short message scheduling method and device based on multidimensional dynamic optimization, the performance data of multiple short message channels are acquired, and the channel health score is calculated based on weight dynamic adjustment model;Determine the scheduling strategy according to the priority identifier of the message to be sent, and perform channel screening and optimal matching;Execute message sending and monitor the sending state;The terminal state detection is carried out to the failed message through the operator base station signaling, and the decision tree model is applied to determine the retry strategy;The P2 marketing message of retry failure is compressed by Huffman coding, and batch sending is executed in idle time window.The application also constructs a comprehensive performance evaluation index and reward function model, and applies reinforcement learning algorithm for parameter optimization.The application realizes multidimensional dynamic channel scoring, intelligent retry decision based on terminal state perception, cost-time efficient batch processing and closed-loop self-optimization system, significantly improves the short message delivery rate, and reduces the invalid retry rate and sending cost.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, and in particular to an intelligent SMS scheduling method and apparatus based on multi-dimensional dynamic optimization. Background Technology

[0002] SMS service, as an important means of communication, is widely used in scenarios such as financial verification, notification reminders, and marketing promotions. To ensure efficient and reliable message delivery, the SMS dispatch system plays a crucial bridging role between the sending terminal and the operator's network, responsible for core functions such as message routing, channel selection, and retry mechanisms.

[0003] Currently, common SMS dispatching technologies in the industry mainly include polling and simple retry mechanisms. Polling allocates SMS messages to different channels in a fixed order using a pre-set channel list; while simple retry mechanisms retry failed messages at fixed time intervals (e.g., 5 minutes). Some systems also use static priority queues, setting different sending priorities based on message type (e.g., verification codes, notifications, marketing messages), or optimizing only a single dimension such as success rate or cost.

[0004] Existing technologies employ SMS dispatch systems based on preset rules. These systems schedule SMS messages by manually configured channel selection rules and fixed retry strategies. The system uses success rate as the primary criterion for channel selection, performs simple timed retries on failed messages, and adjusts system parameters through manual monitoring.

[0005] However, this traditional scheduling technology has obvious drawbacks: First, it lacks dynamic adaptability and cannot respond in real time to sudden changes in channel status (such as operator downtime or network congestion); second, it wastes resources severely, with indiscriminate retries for unreachable terminals (such as those in a powered-off state) accounting for more than 25%; third, it lacks multi-objective coordination, making it difficult to simultaneously meet the timeliness requirements of high-priority messages and the cost control of low-priority messages; and finally, it has high operation and maintenance costs, requiring continuous manual monitoring of channel quality and manual adjustment of strategies, with response delays typically exceeding 30 minutes.

[0006] Therefore, there is an urgent need for an intelligent SMS scheduling method that can dynamically adapt to changes in channel status, intelligently identify terminal status, balance timeliness and cost, and achieve self-optimization, so as to improve SMS delivery rate and reduce invalid retry rate and sending cost. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent SMS scheduling method and device based on multi-dimensional dynamic optimization, so as to solve the problems of insufficient dynamic adaptability, serious resource waste, lack of multi-objective coordination and high operation and maintenance costs of existing SMS scheduling systems.

[0008] To achieve the above objectives, the technical solution provided by the present invention is as follows:

[0009] A smart SMS scheduling method based on multi-dimensional dynamic optimization includes the following steps:

[0010] The system acquires performance data of multiple SMS channels within a preset time period. The performance data includes success rate, latency, cost, and quota utilization rate. The performance data is cleaned and normalized. Based on a weighted dynamic adjustment model, the health score of the multiple SMS channels is calculated to obtain the channel health evaluation result.

[0011] Receive messages to be sent and their priority identifiers, the priority identifiers including P0 level verification code messages, P1 level notification messages and P2 level marketing messages, determine scheduling policy requirements based on the priority identifiers, perform channel filtering and optimal matching based on the channel health evaluation results and the scheduling policy requirements, and obtain the matching results between messages and channels;

[0012] Based on the matching result between the message and the channel, the message is sent, the sending status and channel response are monitored in real time, the sending result data is received and parsed, and the sending result information is obtained.

[0013] The system obtains the failure message data from the sending result information, performs terminal status detection through operator base station signaling, applies a decision tree model to determine the retry strategy, obtains the retry decision result, and filters out the P2 level marketing messages that failed to retry from the retry decision result.

[0014] The P2-level marketing messages that failed to retry are compressed using Huffman coding. They are then grouped by region based on the receiver's base station information. During off-peak hours, the compressed P2-level marketing messages are sent in batches. The sending results and corresponding cost-saving data are then obtained.

[0015] Preferably, the method further includes: constructing a comprehensive performance evaluation index using the sending result information and the cost saving data, constructing a reward function model based on the comprehensive performance evaluation index, applying a reinforcement learning algorithm to optimize the parameters, generating optimized weight coefficients and a scheduling strategy, and updating the SMS scheduling operation based on the scheduling strategy.

[0016] Preferably, the performance data is cleaned and normalized, and the health score of the multiple SMS channels is calculated based on a weighted dynamic adjustment model to obtain the channel health evaluation result, including:

[0017] The performance data is subjected to outlier detection and cleaning to obtain a cleaned dataset;

[0018] The success rate, latency, cost, and quota utilization rate indicators in the cleaned dataset are normalized to unify the multidimensional indicator data into the 0-1 range, resulting in a normalized dataset.

[0019] Select a preset weight template based on the characteristics of the current time period, determine the success rate weight, delay weight, cost weight and quota weight, and obtain the weight coefficient configuration.

[0020] Based on the normalized dataset and the weight coefficient configuration, the health scores of the multiple SMS channels are calculated using a weight dynamic adjustment model to obtain the channel health evaluation results.

[0021] Preferably, based on the channel health evaluation results and the scheduling policy requirements, channel filtering and optimal matching are performed to obtain the message-channel matching results, including:

[0022] Receive and parse the message content to be sent, identify the priority identifier corresponding to the message content to be sent, and obtain the priority classification result;

[0023] Based on the priority classification results, the corresponding scheduling strategy requirements are determined to obtain the strategy parameter configuration. The scheduling strategy requirements include: P0 level messages require a health score ≥ 90 and a delay ≤ 3 seconds; P1 level messages require a health score ≥ 70 and a delay ≤ 15 seconds; and P2 level messages require a health score ≥ 50 and prioritize cost factors.

[0024] Combining the channel health evaluation results and the strategy parameter configuration, multi-level screening is performed, including screening based on health score thresholds, screening based on hard conditions of specific dimensions, and screening based on availability of quota management, to obtain a candidate set of channels that meet the conditions.

[0025] An optimal matching algorithm is applied to the candidate channel set. The utility value of each channel is calculated using a utility function. The channel with the highest utility value is selected to obtain the matching result between the message and the channel.

[0026] Preferably, message sending is performed based on the matching result between the message and the channel, and the sending status and channel response are monitored in real time to obtain sending result information, including:

[0027] Prepare message sending parameters based on the matching result between the message and the channel, perform content format conversion and security signature processing on the message sending parameters, construct request parameters that conform to the channel API specification, and obtain the sending request configuration;

[0028] Based on the aforementioned sending request configuration, an asynchronous non-blocking HTTP client initiates a message sending request to the selected channel, recording the request initiation timestamp and tracking identifier to obtain the sending execution status;

[0029] Real-time monitoring of the sending execution status, tracking the complete cycle of network connection establishment, request sending, and response reception, recording key time nodes and status changes, and obtaining status tracking data;

[0030] Receive and parse the transmission result data returned by the channel, map the proprietary status codes of each channel to the system's unified status definition, perform success determination, temporary failure determination, and permanent failure determination, and obtain standardized transmission results;

[0031] The channel performance metrics are updated based on the standardized transmission results, the channel health score is recalculated and fed back to the channel scoring system, and the transmission result information is obtained.

[0032] Preferably, the failure message data in the transmission result information is obtained, terminal status is detected through operator base station signaling, a decision tree model is applied to determine the retry strategy, and a retry decision result is obtained, including:

[0033] The failure message data in the sending result information is obtained, and the error codes returned by the channel are mapped and analyzed to obtain the failure reason classification result. The failure reason classification result includes: channel failure, terminal problem, content problem or quota problem.

[0034] Based on the failure cause classification results, the terminal's real-time status is queried through operator base station signaling or manufacturer API to obtain terminal status data, which includes: online reachable, inaccessible when powered off, weak and unstable signal, or roaming status.

[0035] The failure reason classification results and the terminal status data are input into the decision tree model. Based on the failure type, terminal status, message priority and historical retry count, multi-level condition judgment is performed to obtain the retry strategy type.

[0036] Execute the corresponding retry operation according to the retry strategy type, including immediate retry, channel switching, delayed retry, timed retry, content adjustment or termination retry, and obtain the retry execution result;

[0037] Record the retry execution results, update the retry effect statistics and decision model parameters, calculate the success rate index of different retry strategies, and obtain the retry decision results.

[0038] Preferably, the P2-level marketing messages that failed to retry are subjected to Huffman coding compression, geographically grouped according to the receiver's base station information, and the compressed P2-level marketing messages are sent in batches during off-peak hours. The sending result information and corresponding cost-saving data are then obtained, including:

[0039] Collect P2-level marketing messages from the messages to be sent and the P2-level marketing messages that failed to be retried, identify the content of marketing messages by keyword matching, and perform time window aggregation to obtain a set of P2-level messages;

[0040] The P2 level message set is compressed by applying the Huffman coding algorithm to construct a character frequency dictionary and an encoding mapping table, and the original message is converted into a compressed binary stream to obtain compressed message data.

[0041] Based on the recipient's mobile phone number, the location and operator information are parsed, and the data is grouped into three levels: first-level grouping by operator type, second-level grouping by provincial administrative division, and third-level grouping by city or region, resulting in regional grouping results. Each region corresponds to a set of compressed message data to be sent.

[0042] Set the off-peak window to the period of 00:00-06:00, formulate a batch sending plan for the compressed message data corresponding to the regional grouping results, and use the operator discount channel to perform load balancing distribution to obtain the off-peak sending plan;

[0043] During a specified off-peak window, the off-peak sending plan is executed for each region in the region grouping results. Compression rate, sending cost, and delivery rate data are recorded, and the cost savings compared to the standard sending method are calculated to obtain the cost savings data.

[0044] Preferably, the method further includes: employing a decentralized multi-agent reinforcement learning architecture and applying reinforcement learning algorithms to optimize parameters, generating optimized weight coefficients and scheduling strategies, including:

[0045] A multi-agent framework is established, in which multiple distributed agents are deployed, including a channel evaluation agent, a routing decision agent, a retry policy agent, and a resource scheduling agent, with each agent distributed across different physical nodes;

[0046] A minimal communication mechanism is established among agents in the multi-agent framework, exchanging only local reward information and achieving secure data exchange through encrypted communication channels, thus obtaining a distributed cooperative network.

[0047] In the distributed collaborative network, a participant-critic network structure is constructed for each agent. The participant network is used to learn the optimal policy mapping and output action selection, while the critic network is used to evaluate the value of actions and provide policy gradient guidance.

[0048] In the participant-critic network structure, a hybrid reward structure is designed, which includes individual rewards and team rewards. Policy diversity is maintained by policy entropy regularization, and a collaborative behavior incentive mechanism is established.

[0049] Distributed model updates are achieved through consensus algorithms, which support dynamic adjustment of the number of agents according to the system scale, enable cross-scenario knowledge transfer, and obtain global optimization parameters.

[0050] Based on the global optimization parameters, the channel evaluation agent outputs the optimized channel score weight coefficients, the routing decision agent outputs the optimized channel selection strategy, the retry strategy agent outputs the optimized retry decision parameters, and the resource scheduling agent outputs the optimized resource allocation strategy, thus obtaining the optimized weight coefficients and scheduling strategy.

[0051] Preferably, an optimal matching algorithm is applied to the candidate channel set, the utility value of each channel is calculated using a utility function, and the channel with the highest utility value is selected to obtain the matching result between the message and the channel, including:

[0052] Based on the state information of the channel candidate set, a dual deep Q network architecture is constructed, including a main network responsible for action selection and a target network for calculating the target Q value. Multi-dimensional information such as channel health score, message characteristics and system load are fused as state representation.

[0053] The greedy strategy optimization operation is implemented on the dual deep Q network architecture. The action with the highest current Q value is selected, and the probability of random exploration is retained with ε. As the learning progresses, the exploration probability is gradually reduced according to a preset ratio, and the greedy action selection strategy is output.

[0054] Based on the greedy action selection strategy, time series analysis is performed to identify traffic characteristics in different time periods, and routing decisions are made considering the network topology relationship between channels. The bandwidth, latency and reliability of each link are monitored in real time to generate a link activation scheme based on traffic and topology.

[0055] A hierarchical scheduling architecture is established based on the link activation scheme based on traffic and topology, including a master scheduler responsible for global resource allocation and a sub-scheduler focused on local optimization, to achieve efficient state sharing and decision coordination, resulting in a multi-agent DDQN allocator.

[0056] The multi-agent DDQN allocator is equipped with neural network quantization and GPU acceleration technology to implement a caching mechanism for scene decision results, and batch processing of similar requests to obtain the matching results of the message and the channel.

[0057] Preferably, time series analysis is performed based on the greedy action selection strategy to identify traffic characteristics in different time periods, routing decisions are made considering the network topology relationships between channels, and the bandwidth, latency, and reliability of each link are monitored in real time to generate a link activation scheme based on traffic and topology, including:

[0058] Historical traffic data for each time period is collected, and periodic traffic patterns and burst traffic characteristics are identified through an LSTM time series model to predict the traffic demand distribution for future time periods, thus obtaining traffic prediction results.

[0059] Based on the traffic prediction results, a network topology map between channels is constructed, the geographical location, network latency and bandwidth capacity of each channel are recorded, the path cost and reachability matrix between channels are calculated, and a topology relationship model is obtained.

[0060] By combining the aforementioned topology model to monitor the current load rate, average latency, and packet loss rate of each link in real time, potential network bottlenecks and fault risks can be identified, and a link status assessment can be obtained.

[0061] Based on the link status assessment, a predictive activation strategy is obtained by activating backup links and expanding channel capacity in advance during peak traffic periods, and releasing redundant resources and reducing operating costs during off-peak traffic periods.

[0062] By combining the predictive activation strategy with real-time traffic scheduling, the weight allocation and load balancing parameters of each link are dynamically adjusted to obtain the link activation scheme based on traffic and topology.

[0063] Preferably, before executing the off-peak sending plan to each region in the regional grouping results during a specified off-peak window, recording compression rate, sending cost, and delivery rate data, and calculating the cost savings compared to the standard sending method, the method further includes:

[0064] Based on the idle time sending plan, a dual-sum duel depth Q-Learning architecture is constructed. Two independent Q networks are used to reduce overestimation bias. The Q value is decomposed into a state value function V(s) and an advantage function A(s,a), and the Q value is calculated to obtain the D3QL network structure.

[0065] The D3QL network structure is used to implement a multi-level service deployment strategy, which divides the processing logic into a core layer, an edge layer, and a terminal layer. The core layer is used to handle key business logic and data storage, the edge layer is used to handle localized message processing and caching, and the terminal layer is used to implement preprocessing and status detection, thus determining the layered service architecture.

[0066] Based on the intelligent prediction of service demand distribution in the layered service architecture, service instances are pre-deployed in the optimal location, and the distribution of service instances is dynamically adjusted according to load changes to obtain an intelligent service placement strategy.

[0067] Based on the intelligent service placement strategy, differentiated service quality requirements are defined for different service types, a dedicated resource pool is reserved for the first priority service, and QoS parameters are dynamically adjusted according to load and service importance.

[0068] We adopt a microservice architecture and containerized deployment approach, use Kubernetes to manage the dynamic scaling of service instances, minimize state dependencies between services, and test the system's adaptability through fault injection.

[0069] Preferably, the method further includes: obtaining an intelligent service placement strategy using a greedy multiple access scheme, including:

[0070] Based on the intelligent service placement strategy, key messages are transmitted in parallel using multiple independent channels. Forward error correction coding technology is applied to enhance transmission reliability and generate a multi-channel parallel transmission mechanism.

[0071] The success rate, latency, and stability of each transmission path are evaluated in real time using the multi-channel parallel transmission mechanism. The traffic allocation ratio is dynamically adjusted according to the path quality, and the best-performing path is given priority, resulting in an adaptive path selection strategy.

[0072] Based on the adaptive path selection strategy, which comprehensively considers channel capacity, network latency and transmission cost, a weighted round-robin algorithm is applied to achieve intelligent load distribution, resulting in a load balancing algorithm.

[0073] A channel failure detection and switching mechanism is established based on the load balancing algorithm, and the system automatically switches to the backup channel when an abnormality is detected in the main channel.

[0074] The greedy multiple access scheme is obtained by combining the channel fault detection and switching mechanism with the multi-channel parallel transmission mechanism, the adaptive path selection strategy, and the load balancing algorithm.

[0075] The present invention also provides an intelligent SMS scheduling device based on multi-dimensional dynamic optimization, including a memory and a processor, wherein the processor is used to execute the method described in any of the above-mentioned embodiments.

[0076] The beneficial effects of this invention are:

[0077] An innovative dynamic channel scoring mechanism integrates four dimensions: success rate, latency, cost, and quota. It calculates channel health scores in real time using an adjustable weight formula, enabling the system to quickly respond to changes in channel quality and improving the accuracy and adaptability of channel selection.

[0078] Intelligent retry decision tree based on terminal status awareness: By acquiring the status of user equipment in real time through operator base station signaling, the optimal retry strategy is automatically selected for different failure reasons, which significantly reduces the invalid retry rate from 25-40% to less than 5%, and improves the efficiency of system resource utilization.

[0079] Cost-Time Balanced Batch Processing Framework: Combining Huffman coding compression, regional grouping, and off-peak channel utilization, this framework reduces the cost of low-priority messages while ensuring the timeliness of high-priority messages. The cost per message is reduced from 0.048 yuan to 0.026 yuan, achieving a cost saving of approximately 45%.

[0080] Closed-loop reinforcement learning self-optimizing system: Through a reward function-driven reinforcement learning algorithm, the system achieves automatic tuning of scheduling parameters and continuous evolution of the system, enabling it to adapt to environmental changes and continuously improve performance.

[0081] A complete multi-dimensional hierarchical scheduling architecture: from channel scoring, message classification, intelligent routing to failure retries, a complete scheduling closed loop is formed, which solves the limitations of single-dimensional optimization in traditional solutions, improves the overall success rate by 4.5-9.5%, and reduces latency by 60-80% during peak hours. Attached Figure Description

[0082] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this disclosure and, together with the specification, serve to explain the technical solutions of this disclosure. It should be understood that the following drawings only show some embodiments of this disclosure and should not be considered as limiting the scope. Those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0083] Figure 1 A flowchart illustrating the intelligent SMS scheduling method based on multi-dimensional dynamic optimization provided in this embodiment of the invention;

[0084] Figure 2 A flowchart of the dynamic channel scoring system provided in an embodiment of the present invention;

[0085] Figure 3 A flowchart for intelligent channel selection based on message priority and channel health score provided in an embodiment of the present invention;

[0086] Figure 4 This is a flowchart for sending execution messages and monitoring their status, provided in an embodiment of the present invention. Detailed Implementation

[0087] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0088] Please see Figure 1 , Figure 1This is a flowchart illustrating an intelligent SMS scheduling method based on multi-dimensional dynamic optimization provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0089] Step S101: Obtain performance data of multiple SMS channels within a preset time period. The performance data includes success rate, latency, cost, and quota utilization rate. Clean and normalize the performance data. Calculate the health score of the multiple SMS channels based on a weighted dynamic adjustment model to obtain the channel health evaluation result.

[0090] In this embodiment, the SMS dispatch system (hereinafter referred to as the system) first obtains real-time performance data from each SMS channel, including indicators in four dimensions: success rate, latency, cost, and quota utilization. This data is collected from the operator's system via an API interface or recorded by the system's own monitoring module. The collected raw data is cleaned to remove outliers and invalid data, and then normalized to unify indicators of different dimensions into the 0-1 range for easier subsequent calculations.

[0091] The system selects the appropriate weight template based on the characteristics of the current time period (such as peak hours, weekdays, or nighttime) to determine the weight coefficients of each dimension indicator. For example, during peak hours, success rate and latency may be given more importance, while at night, cost factors may be more emphasized. Based on normalized data and weight coefficients, the system calculates the health score (0-100 points) for each channel using the formula: Health Score = W1 × Success Rate + W2 × (1 / Latency) + W3 × (1 / Cost) + W4 × Quota Utilization Rate, thus forming the channel health evaluation result.

[0092] Step S102: Receive the message to be sent and its priority identifier. The priority identifier includes P0 level verification code message, P1 level notification message and P2 level marketing message. Determine the scheduling strategy requirements according to the priority identifier. Perform channel screening and optimal matching based on the channel health evaluation results and the scheduling strategy requirements to obtain the matching result between the message and the channel.

[0093] In this embodiment, the system receives SMS sending requests from business applications and parses the message content and priority identifier. Priorities are divided into three levels: P0 (high-priority messages such as verification codes), P1 (medium-priority notification messages), and P2 (low-priority marketing messages). The system determines corresponding scheduling strategy requirements based on message priority. For example, P0-level messages require a health score ≥ 90 and a delay ≤ 3 seconds; P1-level messages require a health score ≥ 70 and a delay ≤ 15 seconds; and P2-level messages require a health score ≥ 50 and prioritize cost factors.

[0094] Based on the channel health assessment results and scheduling strategy requirements, the system performs multi-level screening, including screening based on health score thresholds, screening based on hard conditions of specific dimensions, and availability screening based on quota management, to obtain a candidate set of channels that meet the conditions. Then, the optimal matching algorithm is applied to the candidate set of channels, and the utility value of each channel is calculated through a utility function. The channel with the highest utility value is selected to obtain the matching result between the message and the channel.

[0095] Step S103: Based on the matching result of the message and the channel, execute message sending, monitor the sending status and channel response in real time, receive and parse the sending result data, and obtain the sending result information.

[0096] In this embodiment, the system prepares message sending parameters based on the matching results, performs content format conversion and secure signature processing on the parameters, and constructs request parameters that conform to the channel API specification. A message sending request is initiated to the selected channel via an asynchronous non-blocking HTTP client, recording the request initiation timestamp and tracking identifier.

[0097] The system monitors the transmission execution status in real time, tracking the complete cycle of network connection establishment, request transmission, and response reception, recording key time nodes and status changes. It receives and parses the transmission result data returned by the channels, mapping the proprietary status codes of each channel to a unified system status definition, and performs success, temporary failure, and permanent failure determinations to obtain standardized transmission results. Based on the transmission results, it updates the channel performance indicators, recalculates the channel health score, and feeds it back to the channel scoring system.

[0098] Step S104: Obtain the failure message data in the sending result information, perform terminal status detection through operator base station signaling, apply a decision tree model to determine the retry strategy, obtain the retry decision result, and filter out the P2 level marketing messages that failed to retry from the retry decision result.

[0099] In this embodiment, the system obtains the failure message data in the sending result, performs mapping analysis on the error codes returned by the channel, and obtains the failure reason classification result, including types such as channel failure, terminal problem, content problem or quota problem.

[0100] Based on the failure cause classification results, the system queries the terminal's real-time status through operator base station signaling (such as AT+CPING command) or manufacturer API to obtain terminal status data, including information such as online availability, inaccessible when powered off, weak and unstable signal, or roaming status.

[0101] The failure reason classification results and terminal status data are input into the decision tree model. Based on failure type, terminal status, message priority, and historical retries, multi-level conditional judgments are performed to determine the retry strategy type. The corresponding retry operation is executed according to the strategy type, including immediate retry, channel switching, delayed retry, timed retry, content adjustment, or termination retry, yielding the retry execution result. The system records the retry execution results, updates the retry effect statistics and decision model parameters, and filters out P2-level marketing messages that failed retryes for subsequent batch processing.

[0102] Step S105: Perform Huffman coding compression on the P2 level marketing messages that failed to retry, group them by region according to the receiver base station information, perform batch sending of the compressed P2 level marketing messages in the idle window, and obtain the sending result information and corresponding cost saving data after the operation is completed.

[0103] In this embodiment, the system collects P2-level marketing messages from messages to be sent and retry failure messages, identifies the content of marketing messages by keyword matching, and performs time window aggregation to obtain a set of P2-level messages.

[0104] The Huffman coding algorithm is applied to the P2 level message set for content compression. A character frequency dictionary and encoding mapping table are constructed to convert the original message into a compressed binary stream, resulting in compressed message data. The compression rate usually reaches more than 35%.

[0105] Based on the recipient's mobile phone number, the location and carrier information are analyzed, and the data is grouped into three levels: first-level grouping by carrier type, second-level grouping by provincial administrative division, and third-level grouping by city or region, resulting in the geographical grouping results.

[0106] The system sets the off-peak window to the period from 00:00 to 06:00, and formulates a batch sending plan for compressed message data corresponding to the regional grouping results. It utilizes the discounted channels provided by the operator during this period for load balancing. Batch sending operations are performed during the designated off-peak window, recording compression rate, sending cost, and delivery rate data. The cost savings compared to the standard sending method are calculated, yielding cost-saving data.

[0107] Step S106: Construct a comprehensive performance evaluation index using the sending result information and the cost saving data, construct a reward function model based on the comprehensive performance evaluation index, apply a reinforcement learning algorithm to optimize the parameters, generate optimized weight coefficients and scheduling strategies, and update the SMS scheduling operation based on the scheduling strategies.

[0108] In this embodiment, the system utilizes the transmission result information and cost-saving data to construct a comprehensive performance evaluation index, reflecting the overall operating effect of the system. Based on the evaluation index, a reward function model is constructed in the form: R = α × success rate - β × latency - γ × cost, where α, β, and γ are weighting coefficients, satisfying α + β + γ = 1.

[0109] The system employs reinforcement learning algorithms (such as DQN, DDPG, or PPO) for parameter optimization. By continuously trying different parameter configurations and evaluating their effects, it gradually finds the optimal weight coefficients and scheduling strategy. The generated optimization results include channel scoring weight coefficients, channel selection strategy parameters, retry decision parameters, and resource allocation strategies. The system applies these optimized parameters to the corresponding modules to update SMS scheduling operations, forming a closed-loop optimization mechanism.

[0110] Please see Figure 2 , Figure 2 A flowchart of a dynamic channel scoring system provided in an embodiment of the present invention. Figure 2 As shown, step S101 specifically includes the following sub-steps:

[0111] Step S201: Perform outlier detection and cleaning on the performance data to obtain a cleaned dataset.

[0112] In this embodiment, the system performs outlier detection and cleaning on the raw performance data collected from the operator's API interface. Outlier detection uses statistical methods (such as the 3σ rule) or machine learning algorithms (such as isolated forest) to identify data points that significantly deviate from the normal range. Cleaning includes removing outliers, imputing missing values ​​(using the mean, median, or interpolation), and smoothing (using methods such as moving averages to reduce data fluctuations). The resulting cleaned dataset contains valid performance metrics for each channel.

[0113] Step S202: Normalize the success rate, latency, cost, and quota utilization rate indicators in the cleaned dataset, unifying the multi-dimensional indicator data to the 0-1 range to obtain a normalized dataset.

[0114] In this embodiment, the system normalizes the four dimensions of the cleaned dataset, converting indicators with different scales and ranges into a uniform 0-1 interval for easier subsequent weighted calculations. The normalization process uses a maximum-minimum normalization method: Xnorm = (X - Xmin) / (Xmax - Xmin), where X is the original value, Xmin and Xmax are the minimum and maximum values ​​of the indicator, respectively, and Xnorm is the normalized value. For indicators like latency and cost, which are "smaller is better," a 1-Xnorm transformation is used to maintain consistency in the "larger is better" principle for all indicators.

[0115] Step S203: Select a preset weight template based on the characteristics of the current time period, determine the success rate weight, delay weight, cost weight and quota weight, and obtain the weight coefficient configuration.

[0116] In this embodiment, the system selects the appropriate weight template based on the characteristics of the current time period. The system presets weight templates for various time period characteristics, such as:

[0117] Peak period template (e.g., weekdays 9:00-18:00): Success rate weight = 0.4, Delay weight = 0.4, Cost weight = 0.1, Quota weight = 0.1;

[0118] For routine periods (such as other times of the weekday): success rate weight = 0.35, delay weight = 0.3, cost weight = 0.2, quota weight = 0.15;

[0119] Nighttime template (e.g., 00:00-06:00): Success rate weight = 0.3, Delay weight = 0.2, Cost weight = 0.4, Quota weight = 0.1;

[0120] The system determines the current time period by judging the time, selects the corresponding weight template, and obtains the specific weight coefficient configuration. This dynamic weight adjustment mechanism enables the system to flexibly adjust the importance of various dimensional indicators according to the business characteristics of different time periods.

[0121] Step S204: Based on the normalized dataset and the weight coefficient configuration, calculate the health score of the multiple SMS channels based on the weight dynamic adjustment model to obtain the channel health evaluation result.

[0122] In this embodiment, the system calculates the health score for each channel using a dynamic weight adjustment model based on a normalized dataset and weight coefficient configuration. The calculation formula is as follows:

[0123] Health score = W1×S + W2×(1 / D) + W3×(1 / C) + W4×Q

[0124] Where S is the normalized success rate, D is the normalized latency, C is the normalized cost, Q is the normalized quota utilization rate, and W1, W2, W3, and W4 are the weighting coefficients of each indicator.

[0125] Multiply the calculation result by 100 to obtain the channel health score within the range of 0-100. The system organizes the health score of each channel and its constituent elements (scores and weights of each dimension) into a structured channel health evaluation result, which serves as an important basis for subsequent channel selection.

[0126] Please see Figure 3, Figure 3 This is a flowchart illustrating intelligent channel selection based on message priority and channel health score, provided as an embodiment of the present invention. Figure 3 As shown, step S102 specifically includes the following sub-steps:

[0127] Step S301: Receive and parse the message content to be sent, identify the priority identifier corresponding to the message content to be sent, and obtain the priority classification result.

[0128] In this embodiment, the system receives SMS requests sent by the business system through a RESTful API or message queue (such as Kafka / RabbitMQ), decodes the message content and verifies its format to ensure that it meets system requirements (such as length limits, character encoding, etc.).

[0129] The system identifies message priority indicators, supporting two methods:

[0130] Explicit marking: The business system directly specifies the priority parameter in the API request;

[0131] Intelligent inference: The system automatically determines message priority through keyword matching, content analysis, and historical pattern recognition;

[0132] For example, messages containing keywords such as "verification code" and "security confirmation" will be automatically classified as P0 level; messages containing words such as "notification" and "reminder" will be identified as P1 level; and messages containing words such as "event" and "discount" will be identified as P2 level. The system outputs priority classification results that include standardized message content and clear priority indicators.

[0133] Step S302: Determine the corresponding scheduling strategy requirements based on the priority classification results to obtain the strategy parameter configuration. The scheduling strategy requirements include: P0 level messages require a health score ≥ 90 and a delay ≤ 3 seconds; P1 level messages require a health score ≥ 70 and a delay ≤ 15 seconds; and P2 level messages require a health score ≥ 50 and prioritize cost factors.

[0134] In this embodiment, the system converts priority identifiers into specific scheduling parameters based on priority classification results and through a preset policy mapping table in the configuration database.

[0135] For P0 level messages (such as verification codes), the system applies the following policy parameters:

[0136] Timeliness requirement: Delivery within 3 seconds;

[0137] Channel screening criteria: Health score ≥ 90;

[0138] Resource allocation: Allows parallel transmission across multiple channels;

[0139] Retry strategy: Trigger fast retry immediately upon failure;

[0140] For P1 level messages (such as notifications), the applied policy parameters are:

[0141] Timeliness requirement: Delivery within 15 seconds;

[0142] Channel screening criteria: Health score ≥ 70;

[0143] Resource allocation: Standard channel single-path transmission;

[0144] Retry policy: Retry within 30 seconds of failure;

[0145] For P2 level messages (such as marketing messages), the application's strategy parameters are:

[0146] Time requirement: Delivery within ≤30 minutes;

[0147] Channel selection criteria: Health score ≥ 50, cost factors are given priority;

[0148] Resource allocation: Allows batch processing and delayed sending;

[0149] Retry policy: Allow long window for retrying (1-4 hours);

[0150] The system will also dynamically adjust these strategy parameters based on the current load and time period characteristics to achieve optimal resource allocation. The output strategy parameter configuration includes specific timeliness indicators, channel filtering conditions, resource allocation rules, and retry strategies.

[0151] Step S303: Combine the channel health evaluation results and the strategy parameter configuration to perform multi-level screening, and perform screening based on health score threshold, screening based on hard conditions of specific dimensions, and availability screening based on quota management to obtain a set of candidate channels that meet the conditions.

[0152] In this embodiment, the system obtains the health score data of all currently available channels from the channel pool and performs multi-level filtering in conjunction with the configured strategy parameters:

[0153] Level 1: Threshold filtering based on health score:

[0154] For P0 messages, filter channels with a health score ≥ 90;

[0155] For P1 messages, filter channels with a health score ≥ 70;

[0156] For P2 messages, filter channels with a health score ≥ 50;

[0157] Level 2: Filtering based on hard criteria of specific dimensions:

[0158] For P0 messages, further filter channels with a delay of ≤3 seconds and a success rate of ≥95%;

[0159] For P1 messages, further filter channels with a delay of ≤15 seconds and a success rate of ≥90%;

[0160] For P2 messages, further filter for channels with lower costs (e.g., unit price ≤ 0.05 yuan / message).

[0161] Level 3: Availability filtering based on quota management:

[0162] Exclude channels that have reached their daily limit;

[0163] Exclude channels that have reached their peak traffic.

[0164] Exclude channels that are under maintenance;

[0165] For special scenarios, the system will also apply additional filtering logic:

[0166] High-priority messages (P0): Ensure there are at least two alternative channels;

[0167] Sensitive content messages: Filter for compliant channels that support the corresponding content types;

[0168] International SMS: Filter channels with routing capabilities for the corresponding country / region;

[0169] The screening results form a candidate set of channels that meet the criteria, including the health score and key performance indicators for each channel.

[0170] Step S304: Apply the optimal matching algorithm to the candidate channel set, calculate the utility value of each channel through the utility function, select the channel with the largest utility value, and obtain the matching result of the message and the channel.

[0171] In this embodiment, the system applies different matching algorithms based on message priority:

[0172] P0 level messages: The "highest reliability first" algorithm is applied, mainly considering success rate and latency metrics;

[0173] P1 level messages: Apply a "balanced" algorithm that balances success rate, latency, and cost.

[0174] P2 level messages: Apply a "cost-first" algorithm to minimize costs while meeting basic delivery requirements;

[0175] The mathematical model of the matching algorithm is based on the utility function:

[0176] U = w1 × (success rate) + w2 × (1 / delay) + w3 × (1 / cost)

[0177] Where w1, w2, and w3 are weighting coefficients, adjusted according to message priority:

[0178] P0 level: w1=0.5, w2=0.4, w3=0.1;

[0179] Level P1: w1=0.4, w2=0.3, w3=0.3;

[0180] Level P2: w1=0.3, w2=0.1, w3=0.6;

[0181] The system calculates the utility value of each channel in the candidate channel set and selects the channel with the highest utility value as the optimal matching result. For special scenarios, the system also applies additional processing mechanisms:

[0182] For extremely high priority messages (such as bank verification codes), enable parallel sending mode;

[0183] For targeted regional messages, priority is given to matching the local channel of the target region;

[0184] For messages from important clients, a VIP channel reservation strategy should be applied;

[0185] The system also implements load balancing and flow control, monitoring the concurrent load of each channel in real time to avoid overloading a single channel and ensure a balanced distribution of channel utilization. The final output shows the matching results between messages and channels, including message ID, selected channel information, backup channel information, and sending parameter configurations.

[0186] Please see Figure 4 , Figure 4 This is a flowchart illustrating the execution message sending and status monitoring provided in an embodiment of the present invention. Figure 4 As shown, step S103 specifically includes the following sub-steps:

[0187] Step S401: Prepare message sending parameters based on the matching result between the message and the channel, perform content format conversion and security signature processing on the message sending parameters, construct request parameters that conform to the channel API specification, and obtain the sending request configuration.

[0188] In this embodiment, the system prepares message sending parameters based on the matching results, including basic information such as the recipient's number, message content, and sender's signature. These parameters undergo content format conversion, such as Unicode encoding, URL encoding, or Base64 encoding, to ensure compliance with the channel API requirements.

[0189] The system also performs secure signature processing, applying encryption algorithms (such as MD5, HMAC-SHA256) to generate secure signatures for API calls, desensitizing or encrypting sensitive information (such as mobile phone numbers), implementing anti-replay attack mechanisms, and adding timestamps and unique request identifiers.

[0190] Based on the API specifications of the selected channel, the system constructs complete request parameters, including specific parameters required by the channel (such as signature, template ID, business code, etc.). The system also performs pre-checks and verifications, executing parameter integrity and validity checks before sending, and verifying whether the recipient's number format and message content conform to the channel requirements. Finally, a standardized sending request configuration is obtained, preparing for the actual sending operation.

[0191] Step S402: Based on the sending request configuration, initiate a message sending request to the selected channel through an asynchronous non-blocking HTTP client, record the request initiation timestamp and tracking identifier to obtain the sending execution status.

[0192] In this embodiment, the system uses an asynchronous, non-blocking HTTP client (based on the Netty / Reactor pattern) to initiate message sending requests to the selected channel. HTTPS encrypted communication ensures data transmission security, and TCP connection pool management is implemented to optimize connection reuse efficiency.

[0193] The system employs a token bucket algorithm to achieve precise sending frequency control, avoiding triggering channel rate limiting mechanisms. The number of concurrent sending threads is configured based on channel capacity to maximize throughput. A priority queue is implemented to ensure that high-priority messages are processed first.

[0194] When sending a request, the system records the request initiation timestamp for subsequent delay calculation; assigns a globally unique request tracking ID to achieve end-to-end tracing; and records request status changes in real time, including stages such as queuing, sending, and response receiving.

[0195] For special scenarios, the system implements differentiated processing strategies:

[0196] For high-priority P0 messages, a multi-channel parallel sending mechanism is enabled.

[0197] For new access channels or configuration changes, implement a canary release strategy;

[0198] When the system is under high load, it supports targeted degradation to prioritize the sending of core business messages.

[0199] The system also implements a circuit breaker mode, which quickly switches to a backup channel when a channel anomaly is detected; application timeout control, which actively interrupts and retryes slow-responding requests; and maintains request context information to support correct matching of asynchronous responses.

[0200] Step S403: Monitor the sending execution status in real time, track the complete cycle of network connection establishment, request sending, and response reception, record key time nodes and status changes, and obtain status tracking data.

[0201] In this embodiment, the system implements a multi-level status monitoring mechanism:

[0202] HTTP layer monitoring: Track the complete cycle of network connection establishment, request sending, and response receiving;

[0203] Business layer monitoring: Parse the channel response content and identify success / failure status and error codes;

[0204] End-to-end monitoring: Obtain the final delivery status of the message through channel callbacks or status query APIs;

[0205] The system performs real-time data collection and analysis, records key time nodes (request initiation time, response reception time, status update time), calculates key performance indicators (API response time, processing latency, end-to-end latency), and classifies and statistically analyzes error responses (network errors, authentication errors, business errors).

[0206] The system implements anomaly detection and early warning. It uses a sliding window algorithm to detect channel performance anomalies (such as a sudden drop in success rate or a sudden increase in latency), sets multi-level threshold alarms, and triggers a system alarm when the indicators exceed the normal range. It also applies anomaly pattern recognition to automatically detect recurring error patterns.

[0207] Monitoring data is written to a distributed time-series database (such as InfluxDB) in real time, providing a real-time status query interface, enabling business systems to obtain message sending progress, archiving historical data, and supporting long-term trend analysis. The system also builds a real-time monitoring dashboard to intuitively display the health status and performance indicators of each channel, provides a function for quickly locating abnormal channels, and supports custom monitoring views.

[0208] Step S404: Receive and parse the transmission result data returned by the channel, map the proprietary status codes of each channel to the unified status definition of the system, perform success determination, temporary failure determination and permanent failure determination, and obtain standardized transmission results.

[0209] In this embodiment, the system obtains the sending result data through multiple methods:

[0210] Synchronous response parsing: Processes direct HTTP response data from API calls;

[0211] Asynchronous callback processing: Receive status callback notifications pushed by the channel (Webhook);

[0212] Active query: For channels that do not support callbacks, periodically call the query API to obtain the latest status;

[0213] The system employs a channel adapter pattern, implementing dedicated parsers for the return formats of different channels. It maps the proprietary status codes of each channel to a unified system status definition, and categorizes returned error messages into types such as network errors, authentication errors, and content errors.

[0214] The system implements the logic for determining the result status:

[0215] Success determination: Confirm whether the message was successfully sent or delivered according to the channel specifications;

[0216] Temporary failure determination: Identify temporary errors (such as channel congestion) that can be resolved by retrying;

[0217] Permanent failure determination: Identifies permanent errors that cannot be resolved by retries (such as a number not existing);

[0218] The system also performs advanced semantic parsing to extract structured failure reason information from error descriptions, identify specific error patterns (such as "blacklist", "content violation" etc.), and use NLP technology to perform semantic analysis and classification on unstructured error descriptions.

[0219] The system associates the sent result data with the original request information to form a complete request-response record, calculates end-to-end performance metrics, adds environmental context information (such as system load and network conditions at the time of sending), and obtains standardized sending results.

[0220] Step S405: Update the channel performance indicators according to the standardized transmission results, recalculate the channel health score and feed it back to the channel scoring system to obtain the transmission result information.

[0221] In this embodiment, the system updates the channel performance metrics in real time based on the standardized transmission results:

[0222] Success rate calculation: Incorporate new sending results into the sliding window (e.g., within 5 minutes) for success rate calculation;

[0223] Latency statistics: metrics such as average latency of the update channel and P95 latency (95% of message latency is below this value);

[0224] Error distribution: Update the percentage statistics of various types of errors and identify the main failure modes;

[0225] Throughput capacity: The actual processing rate and load level of the update channel;

[0226] The system recalculates the channel health score using the formula in step S101 based on the updated performance metrics, and feeds back the latest health score to the channel scoring system for channel selection in subsequent messages.

[0227] The system writes complete sending records to persistent storage (such as a distributed database), performs real-time backups of high-value data to ensure data security, and implements a tiered data storage strategy, where hot data is stored in high-speed storage and cold data is migrated to low-cost storage.

[0228] The system pushes and sends results according to the callback methods preset by the business system (HTTP callback, message queue, etc.), provides differentiated result data for different result statuses (success / failure), and triggers real-time alarm notifications for failure results of important business.

[0229] The system also performs real-time aggregation and analysis of the sending result data, generates operational reports, identifies abnormal patterns and performance bottlenecks, and extracts business value information, such as user active time periods and regional distribution, to obtain complete sending result information.

[0230] Step S104 specifically includes the following sub-steps:

[0231] Step S501: Obtain the failure message data in the sending result information, perform mapping analysis on the error codes returned by the channel, and obtain the failure reason classification result. The failure reason classification result includes: channel failure, terminal problem, content problem or quota problem.

[0232] In this embodiment, the system filters out message records with a status of "failed" from the sending result information, collects complete failure context information, including message content summary, recipient number characteristics, sending time, channel used, etc., and extracts the original error code and error description information returned by the channel.

[0233] The system performs error code mapping analysis, mapping the error codes returned by the channels to standardized failure types. NLP techniques are applied to unstructured error descriptions to extract key information and failure causes. Comparison with historical failure cases identifies similar failure patterns.

[0234] The system categorizes the analysis results into system-defined standard failure types:

[0235] Temporary channel failures: such as network fluctuations or server overload;

[0236] Permanent channel failures: such as interface changes or authentication failures;

[0237] Temporary terminal issues: such as weak signal or brief offline status;

[0238] Permanent issues with the device: such as a non-existent number or a device that has been switched off for an extended period;

[0239] Content-related issues: such as sensitive words, exceeding length limits;

[0240] Quota or traffic restriction issues: such as daily quota being exhausted or frequency exceeding the limit;

[0241] The system calculates a failure severity score based on failure type and frequency. For failures of the same type occurring in batches, its severity rating is increased, assigning a higher severity level to failures of critical business messages. The system also detects correlation patterns between failure messages, analyzes failure time distribution, assesses whether failures are related to message content characteristics, and obtains structured failure cause classification results.

[0242] Step S502: Based on the failure cause classification results, query the real-time status of the terminal through operator base station signaling or manufacturer API to obtain terminal status data. The terminal status data includes: online reachable, power off and inaccessible, weak and unstable signal, or roaming status.

[0243] In this embodiment, based on the failure cause classification results, the system performs terminal status detection in multiple ways for failures that may be related to the terminal status:

[0244] Base station signaling detection: Query terminal status using base station signaling commands such as AT+CPING provided by the operator;

[0245] Manufacturer API Integration: For mobile phone manufacturers that support device status queries, call their open APIs;

[0246] Push service status: Obtain the terminal's online status using the system-integrated push service;

[0247] Carrier Network API: Connects to terminal status query interfaces provided by carriers such as Carrier 1, Carrier 3, and Carrier 2;

[0248] The system constructs a standardized status query request, which includes the target mobile phone number and query type. Based on the number's location and operator characteristics, it selects the corresponding query channel, sends the query request, sets an appropriate timeout parameter (usually 1-3 seconds), receives and parses the query results, and extracts the status information.

[0249] The system classifies terminal status into the following categories:

[0250] Online / Accessible: The terminal is in normal working order and can receive SMS messages;

[0251] Power off / Unable to access: The terminal is powered off or has not been connected to the network for an extended period of time;

[0252] Weak / unstable signal: The terminal is located at the edge of signal coverage or in an area with unstable network;

[0253] Roaming status: The terminal is in international or domestic roaming status;

[0254] Unknown state: Unable to obtain definite state information;

[0255] The system performs a status credibility assessment. When multiple detection methods are available, it integrates status information from multiple sources, assesses the current credibility based on the acquisition time of the status data, and checks whether the status information is consistent with the failure mode.

[0256] The system implements a tiered caching mechanism to reduce redundant queries. Different cache expiration times are set based on the status type (e.g., 30 minutes for power-off status, 5 minutes for weak signal status). For high-priority messages, real-time queries are enforced, ignoring the cache. Ultimately, reliable terminal status data is obtained.

[0257] Step S503: Input the failure reason classification result and the terminal status data into the decision tree model, and perform multi-level condition judgment based on failure type, terminal status, message priority and historical retry count to obtain the retry strategy type.

[0258] In this embodiment, the system takes the failure cause classification results and terminal status data as input, and applies an intelligent decision tree model to determine the optimal retry strategy for each failure message.

[0259] The core decision factors of the decision tree model include: failure type, failure reason, terminal state, message priority, and historical retries. The decision tree structure employs multi-layered conditional decision nodes, with each node branching based on specific conditions. The implementation combines pre-defined rules and machine learning into a hybrid decision system, supporting online updates to the decision rules.

[0260] The main logic of the decision-making process includes:

[0261] First-level judgment: Initial routing based on failure type (temporary / permanent);

[0262] Second-level judgment: refine the processing strategy based on the terminal status (online / offline / weak signal);

[0263] Three-layer judgment: Optimize the strategy by considering message priority and historical retry scenarios;

[0264] Special case handling: Apply specific handling rules to specific error codes (such as sensitive content);

[0265] The system defines the following retry policy types:

[0266] Retry immediately: Retry immediately using the same channel (applicable to temporary network fluctuations);

[0267] Channel switching: Retry using an alternative channel (applicable to specific channel failures);

[0268] Delayed retry: Retry after setting a specific delay time (applicable when the receiver is temporarily unreachable);

[0269] Scheduled retry: Schedule a retry at a specific time (suitable for users who turn off their devices at night).

[0270] Content adjustment: Modify the message content and try again (applicable to content violation issues);

[0271] Terminate retry: Abandon further retry attempts (applicable to permanent failures or reaching the maximum number of retries);

[0272] The system also implements advanced policy optimization features, including adaptive delay computation, parallel retry decision-making, cost constraint handling, and incremental retries. Based on the decision tree's judgment, the system outputs the specific retry policy type.

[0273] Step S504: Execute the corresponding retry operation according to the retry strategy type, including immediate retry, channel switching, delayed retry, timed retry, content adjustment, or termination retry, and obtain the retry execution result.

[0274] In this embodiment, the system executes specific retry operations according to the retry policy type. The system constructs a retry task queue, sorts it by priority and planned execution time, and implements a distributed timed scheduling system to ensure that retry tasks are executed on schedule.

[0275] For different retry strategies, the system implements differentiated processing logic:

[0276] Retry immediately: Retain the original sending parameters and submit a new sending request directly;

[0277] Channel switching: Select a new sending channel based on the channel health score and current load, and update the sending parameters;

[0278] Delayed retry: Set a delay trigger to initiate a retry after a specified delay (e.g., 5 minutes, 30 minutes);

[0279] Scheduled retries: Calculate the optimal retry time window (such as during peak user activity periods) and schedule retries at designated times;

[0280] Content adjustment: Apply content optimization rules (such as sensitive word replacement, length adjustment) to generate new message content;

[0281] Parallel retry: Submit sending requests to multiple channels simultaneously, using a "first success, return" strategy;

[0282] The system implements retry resource management and control, including retry quota control to prevent a single message from consuming resources through unlimited retries; a priority-based resource allocation strategy to ensure that high-priority messages receive sufficient retry opportunities; and an intelligent traffic shaping algorithm to prevent retry storms from causing system overload.

[0283] The system tracks the retry history and current status of each message in real time, monitors the trend of retry success rate changes, identifies abnormal patterns, and provides a manual intervention interface, allowing operations and maintenance personnel to adjust the automatically generated retry strategy.

[0284] The system also handles various boundary conditions, including a maximum retry limit, a retry timeout mechanism, and emergency interrupt handling. After executing a retry operation, the system records detailed retry execution results.

[0285] Step S505: Record the retry execution results, update the retry effect statistics and decision model parameters, calculate the success rate index of different retry strategies, and obtain the retry decision results.

[0286] In this embodiment, the system tracks the execution status and result of each retry operation, records key performance indicators such as retry latency, channel response time, and final status, establishes the correlation between the original failure and the retry result, and forms a complete failure-retry chain.

[0287] The system calculates the success rate metrics for different retry strategies:

[0288] Immediate retry success rate = Number of successful immediate retries / Total number of immediate retries;

[0289] Success rate of delayed retries = Number of successful delayed retries / Total number of delayed retries;

[0290] Channel handover success rate = Number of successful channel handovers / Total number of channel handovers;

[0291] The system analyzes the relationship between retry time and success rate, identifies the optimal retry time window, and evaluates the accuracy of terminal status detection and its correlation with retry success.

[0292] The system updated several system data items:

[0293] Update channel performance metrics: incorporate retry data into the calculation of channel success rate, latency, and other metrics;

[0294] Update decision model parameters: Adjust the weights and branch conditions of the decision tree based on the retry results;

[0295] Update the terminal status database: record the user terminal's activity mode and reachability characteristics;

[0296] Update the failure mode library: accumulate the mapping relationship between failure causes and optimal recovery strategies;

[0297] The system generates a retry performance analysis report, including a comparison of the success rates of various strategies, calculation of resource efficiency indicators: the ratio of retry resource input to success benefits, identification of system bottlenecks and optimization opportunities, and provision of time and geographical distribution analysis of retry patterns.

[0298] The system establishes a retry knowledge base, accumulates experience data on efficient retry strategies, implements a strategy effectiveness scoring mechanism, provides reward signals for reinforcement learning, generates retry strategy optimization suggestions, and guides the adjustment of system parameters. Finally, it outputs complete retry decision results and filters out P2-level marketing messages that failed retryes for subsequent batch processing.

[0299] Step S105 specifically includes the following sub-steps:

[0300] Step S601: Collect P2 level marketing messages from the messages to be sent and the P2 level marketing messages that failed to retry, identify the content of marketing messages by keyword matching, and perform time window aggregation to obtain a set of P2 level messages.

[0301] In this embodiment, the system collects P2-level marketing messages from two sources: new P2-level messages received directly from the business system, and P2-level marketing messages filtered from retry failure messages. The system applies multiple criteria to identify low-priority messages:

[0302] Priority marking: Directly filter messages marked as P2 level;

[0303] Content feature identification: Identify marketing messages (such as "discounts", "events", "promotions") through keyword matching;

[0304] Sender characteristics: based on the historical classification of the sending business system (such as e-commerce marketing platform);

[0305] Timeliness requirement: Filter non-urgent messages with an SLA requirement of more than 30 minutes;

[0306] The system implements intelligent classification algorithms, including rule-based classifiers, machine learning models (such as BERT and FastText), and hybrid classification strategies to improve classification accuracy.

[0307] The system is designed with a batch collection mechanism, including time window aggregation (setting a dynamic time window of 5-30 minutes to collect messages), capacity threshold aggregation (triggering batch processing when a specific number, such as 1000 messages, is reached), and intelligent dynamic window (automatically adjusting the collection window based on message traffic and time sensitivity).

[0308] The system performs grouping and indexing, grouping by content similarity and recipient characteristics to build efficient indexes that support fast querying and retrieval. The system also implements priority balancing and resource scheduling, monitors system load in real time, and dynamically adjusts collection strategies, appropriately reducing the collection window during peak periods and expanding it during off-peak periods. The final result is a structured P2-level message set.

[0309] Step S602: Apply the Huffman coding algorithm to the P2 level message set for content compression processing, construct a character frequency dictionary and encoding mapping table, convert the original message into a compressed binary stream, and obtain compressed message data.

[0310] In this embodiment, the system first performs content analysis and preprocessing on the P2 level message set, including text feature extraction, template recognition, variable extraction, and content normalization, in preparation for compression processing.

[0311] The system implements the Huffman coding algorithm, including the following steps:

[0312] Construct a frequency dictionary: Count the frequency of occurrence of characters or phrases in a message set;

[0313] Generating Huffman Trees: Constructing the Optimal Prefix Encoding Tree Based on the Frequency Dictionary;

[0314] Encoding mapping table generation: Maps common characters or phrases to short codes;

[0315] Compressed execution: Converts the original message into a Huffman-coded binary stream;

[0316] The system also implements adaptive encoding, dynamically adjusting the encoding strategy according to the characteristics of different message types to improve compression efficiency.

[0317] The system employs a variety of advanced compression technologies:

[0318] Template-based compression: For similar template messages, only the template ID and variable parameters are transmitted;

[0319] Incremental compression: For batch messages with high similarity, a transmission method of base message + difference data is adopted;

[0320] Dictionary compression: Create compressed dictionaries for commonly used marketing terms to achieve more efficient phrase-level compression;

[0321] Multi-stage compression: Combining Huffman coding and algorithms such as LZ77 to achieve higher compression ratios;

[0322] The system evaluates and controls the compression effect, monitors the actual compression rate achieved (average ≥35%), ensures that the time cost of compression processing does not affect the overall processing efficiency, verifies the integrity and decodeability of the compressed content, and automatically skips the compression step when the compression effect is poor (compression rate <10%).

[0323] The system also addresses channel compatibility issues, adjusting the compressed output based on the data formats supported by the target channel, embedding necessary decoding instructions into the compressed data, designing an efficient batch compressed message transmission protocol, and automatically switching to standard transmission mode for channels that do not support compressed transmission. The result is highly compressed message data.

[0324] Step S603: Based on the recipient's mobile phone number, parse the location and operator information, group the data by operator type (first level), by provincial administrative division (second level), and by city or region (third level) to obtain the regional grouping results. Each region corresponds to a set of compressed message data to be sent.

[0325] In this embodiment, the system uses various technologies to parse the location and carrier information of the recipient's mobile phone number:

[0326] Number segment database query: Maintains a nationwide database of mobile phone number segment locations, supporting high-speed queries;

[0327] Carrier identification: Identifying carrier affiliation based on number segment rules;

[0328] Real-time API query: For numbers that cannot be resolved locally, call the operator's real-time query interface;

[0329] Caching mechanism: Establish multi-level caching to accelerate the location resolution of high-frequency numbers;

[0330] The system implements a multi-dimensional grouping strategy:

[0331] First-level grouping: Grouped by carrier type;

[0332] Secondary grouping: Grouped by provincial-level administrative regions;

[0333] Three-level grouping: further subdivided by city or region;

[0334] Virtual operator processing: Establishing special mapping rules for virtual operator numbers;

[0335] The system applies intelligent routing optimization algorithms to perform channel-region affinity analysis and identify the delivery performance of each channel in different regions; it constructs a cost matrix to establish a tariff difference matrix between operators and regions; it calculates the optimal grouping scheme based on performance and cost data; and it considers load balancing to avoid overloading of a single channel while ensuring cost optimization.

[0336] The system establishes an efficient grouped data structure and management mechanism, creates multi-level indexes for grouped messages, maintains metadata such as group information, quantity statistics, and priority, sets the validity period and processing priority of groups, and supports dynamic adjustment of groups.

[0337] The system also handles special scenarios, including cross-border message processing, handling of messages from unknown locations, small-scale packet merging, and high-priority message routing. Ultimately, it yields complete regional grouping results, with each region corresponding to a set of compressed message data to be sent.

[0338] Step S604: Set the idle window to the period from 00:00 to 06:00, formulate a batch sending plan for the compressed message data corresponding to the regional grouping results, and use the operator discount channel to perform load balancing distribution to obtain the idle sending plan.

[0339] In this embodiment, the system first defines and optimizes the idle window:

[0340] Standard off-peak window: set to 00:00-06:00 based on industry experience;

[0341] Dynamic window adjustment: By analyzing historical traffic patterns, accurately identify the true off-peak periods;

[0342] Tiered time slot division: The off-peak window is subdivided into multiple time slots (such as 00:00-02:00, 02:00-04:00, 04:00-06:00).

[0343] Differential pricing perception: Identifying the optimal discount periods for different operators and channels;

[0344] The system design includes a sending plan algorithm, which includes a load balancing algorithm (distributing messages evenly within the idle window to avoid instantaneous traffic peaks), priority sorting (maintaining the relative priority between messages even in batch processing), timeliness constraint handling (ensuring that messages with deadline requirements are sent before the deadline), and dynamic scheduling optimization (automatically adjusting the sending plan based on the real-time system load).

[0345] The system implements intelligent time point selection, constructs receiver activity patterns based on historical data, and selects time points with higher activity probabilities; predicts the possible delivery rate at different time points and selects the best delivery rate period; balances the discount rate and delivery rate to calculate the optimal sending time point; for large batches of messages, a small-scale trial sending mechanism is adopted to verify the effect before main sending.

[0346] The system performs capacity planning and control, evaluates the processing capacity of each discount channel at different times, sets a smooth sending rate curve, reserves some channel capacity for possible emergency messages, and intelligently decides to postpone or switch to regular sending when batch messages exceed the idle window capacity.

[0347] The system generates a visual delivery plan that supports manual review. It transforms the delivery plan into precise scheduled tasks, monitors the plan's execution and actual delivery results, and allows for manual intervention and adjustments during plan execution. The result is an optimized off-peak delivery plan.

[0348] Step S605: Execute the off-peak sending plan for each region in the region grouping results during the specified off-peak window, record the compression rate, sending cost and delivery rate data, calculate the cost savings compared to the standard sending method, and obtain the cost savings data.

[0349] In this embodiment, the system implements a batch sending execution mechanism, including accurately triggering batch sending tasks based on a preset plan, performing a final channel availability and parameter validity check before execution, decomposing large batch tasks into multiple small batches for progressive sending, dynamically adjusting the sending rate based on channel feedback, and automatically detecting and recovering from various anomalies during the sending process.

[0350] The system employs batch API optimization technology to merge multiple messages into a single API request, reducing the number of interface calls; maintains persistent connections with the channel server to reduce connection establishment overhead; optimizes request packet size and format based on channel characteristics; and processes different batches in parallel using multi-threading to improve overall throughput.

[0351] The system collects and analyzes cost data, records the theoretical sending cost of messages in standard channels and standard time periods, records the actual sending cost in discounted channels and off-peak windows, quantifies the API call cost savings brought about by content compression, quantifies the cross-network cost savings brought about by regional grouping optimization, and summarizes the comprehensive cost savings brought about by various optimizations.

[0352] The system implements delivery quality monitoring and assurance, closely tracks the delivery rate of batches, triggers an alarm when the delivery rate is lower than the expected threshold, automatically switches batches with abnormal delivery rates to backup channels, and collects and analyzes detailed delivery status reports.

[0353] The system evaluates the optimization effect and provides feedback, analyzes the cost-quality balance, compares the cost-effectiveness and delivery performance during different off-peak periods, generates optimization suggestions for future batch processing based on actual results, tracks long-term cost-saving trends, and verifies the continued effectiveness of the optimization strategy. Ultimately, detailed cost-saving data is obtained, including a reduction in the cost per message from the standard 0.048 yuan / message to 0.026 yuan / message, achieving approximately 45% cost savings.

[0354] Step S106 specifically includes the following sub-steps:

[0355] Step S701: Collect system operation data and send result statistics to build a complete performance dataset.

[0356] In this embodiment, the system implements a multi-dimensional data acquisition architecture to collect multiple types of data:

[0357] Sending result data includes core metrics such as success rate, delivery delay, and status distribution.

[0358] Channel performance data: Real-time health status, load level, and failure mode of each channel;

[0359] Resource consumption data: resource metrics such as API call volume, bandwidth usage, and processing time;

[0360] Cost-related data: unit cost, actual expenditure, and cost savings for each channel;

[0361] User experience data: message delivery timeliness, stability, and user feedback;

[0362] The system adopts real-time data stream processing technology, collects real-time events based on the Kafka / Flink stream processing architecture, uses professional time-series databases (such as InfluxDB) to store performance metrics, realizes end-to-end data acquisition based on OpenTelemetry, implements intelligent sampling for high-frequency data, and balances data accuracy and storage costs.

[0363] The system employs a variety of statistical analysis methods, including sliding window analysis over different time spans (5 minutes, 1 hour, 24 hours), anomaly pattern recognition based on statistics and machine learning, correlation and causal relationship analysis among various indicators, and trend prediction based on LSTM time series models.

[0364] The system performs data aggregation and standardization, realizing hierarchical aggregation from single messages to batches, channels, and business lines. It normalizes indicators of different dimensions, assigns weights to different indicators according to business importance, and generates comprehensive indicators that reflect the overall health of the system.

[0365] The system also employs data quality assurance measures, including checking the integrity of data collection, ensuring consistency and comparability of data across systems, guaranteeing the real-time nature of critical data, and identifying and handling quality issues such as outliers and duplicate data. Ultimately, a complete performance dataset is constructed to provide a foundation for subsequent analysis.

[0366] Step S702: Construct a reward function model (R=α×success rate-β×delay-γ×cost) to quantitatively evaluate the system behavior.

[0367] In this embodiment, the system designs a mathematical model for the reward function:

[0368] R = α × success rate - β × delay - γ × cost

[0369] Among them, the success rate is the percentage of messages that are finally delivered (0-100%), the delay is the average delivery time (seconds), the cost is the average cost of sending a single message (yuan), and α, β, and γ are the weight coefficients of each factor, satisfying α + β + γ = 1.

[0370] The system implements a dynamic weight adjustment mechanism, pre-setting weight templates according to different business scenarios:

[0371] CAPTCHA scenario: α=0.3, β=0.6, γ=0.1 (high importance placed on timeliness);

[0372] Notification scenario: α=0.5, β=0.3, γ=0.2 (balancing success rate and timeliness);

[0373] Marketing scenario: α=0.4, β=0.1, γ=0.5 (focus on success rate and cost control);

[0374] The system supports adaptive weights, which automatically fine-tune the weight coefficients based on actual operating results, using different weight configurations during peak and off-peak periods.

[0375] The system performs reward normalization and balancing, transforming indicators of different dimensions into the [0,1] interval, applying nonlinear functions to adjust the sensitivity of certain indicators (such as taking the logarithm of the delay), setting upper and lower thresholds for indicators, controlling the impact of extreme values, and applying techniques such as moving averages to reduce the interference of data fluctuations on reward calculation.

[0376] The system implements a multi-objective balancing technique to find the Pareto optimal solution among multiple objectives, sets hard constraints on key indicators (such as a minimum success rate of 95%), prioritizes the core objective, optimizes secondary objectives, and dynamically adjusts the relative importance of each objective based on the current system state.

[0377] The system also addresses reward delay and credit allocation issues, applying a time decay factor to rewards for historical behavior to rationally distribute the final reward to the decision-making stages of contribution. It also handles reward expectation calculations in some observable scenarios, balancing immediate rewards and long-term cumulative benefits. Ultimately, this results in a scientific and flexible reward function model capable of accurately evaluating the comprehensive effects of different scheduling strategies.

[0378] Step S703: Apply reinforcement learning algorithms to optimize parameters and find the optimal parameter combination by balancing exploration and utilization.

[0379] In this embodiment, the system designs a reinforcement learning framework, defining a state space (a representation of the system's current state, including channel health status, message queue length, time period characteristics, etc.), an action space (actions that the system can take, such as adjusting channel scoring weights, modifying retry policy parameters, etc.), an agent architecture (agent design based on deep Q-networks or policy gradient algorithms), and an environment model (constructing a system environment model to predict the consequences and rewards of actions).

[0380] The system implements the core algorithm, selects a suitable RL algorithm (such as DQN, DDPG, PPO, etc.) according to the characteristics of the problem, maintains an experience pool, stores the <state, action, reward, next state> quadruple, approximates the Q function or value function through a neural network, optimizes the parameterization strategy using gradient ascent, and implements an ε-greedy or entropy-based exploration strategy to balance exploration and exploitation.

[0381] The system controls the learning process, dynamically adjusts the learning rate, and balances convergence speed and stability; it samples mini-batches from the experience pool for batch training; it updates the target network using soft updates or periodic updates; it designs reasonable early stopping conditions to avoid overfitting; and it gradually transitions from simple scenarios to complex scenarios in its learning process.

[0382] The system performs multi-parameter joint optimization, including channel scoring weights (optimizing parameters w1~W4 to make channel selection more accurate), retry strategy parameters (learning the best retry interval and retry channel selection strategy), batch processing threshold (optimizing the triggering conditions and execution parameters of batch processing), and resource allocation ratio (learning the optimal resource allocation scheme for messages of different priorities).

[0383] The system implements safety constraints and interpretability design, limiting the range of parameter variations to avoid extremely unreasonable configurations; limiting the magnitude of single adjustments to ensure system stability; supporting a hybrid learning mode of expert knowledge guidance and manual intervention; and providing an explanation mechanism for decision-making basis and reasons for parameter changes. Through continuous learning and self-adjustment, the system gradually optimizes the configuration of key parameters and improves overall performance.

[0384] Step S704: Generate new weight coefficients and scheduling strategies to provide optimized decision-making basis for system operation.

[0385] In this embodiment, the system extracts optimized parameters from the learning model, including channel score weight coefficients (w1~W4), various decision thresholds (such as health score screening thresholds and retry trigger thresholds), and key parameters of the scheduling strategy (such as priority queue configuration and resource allocation ratio), and converts the model output into the system configuration format to ensure compatibility.

[0386] The system performs parameter verification and adjustment, checks whether the generated parameters are within a reasonable range, ensures the logical consistency between related parameters, imposes reasonable constraints on parameters that exceed the safe range, and smooths out parameters with excessively large changes to avoid system oscillations.

[0387] The system enables scenario-based strategy customization, generating differentiated strategies for different time periods (peak / off-peak / low-peak), customizing dedicated strategies for different business types (finance / e-commerce / government, etc.), adjusting matching strategies based on channel characteristics (such as stability / cost characteristics), and customizing optimized regional strategies for user groups in different regions.

[0388] The system performs strategy packaging and version management, packaging relevant parameters into complete strategy configuration sets, assigning a unique version number to each strategy set, supporting version tracking, recording detailed parameter changes and optimization purposes, and supporting quick rollback to previous versions when necessary.

[0389] The system prepares deployment and switchover plans, identifies the subset of parameters that need to be updated, and supports incremental deployment; designs a progressive deployment scheme to control the scope of impact; predicts the performance changes and business impacts brought about by new parameters; and formulates emergency handling procedures for deployment anomalies. Finally, it generates an optimized scheduling strategy configuration that can be directly applied to the production environment.

[0390] Step S705: Deploy the optimized parameters and monitor the effect to form a closed-loop optimization mechanism and realize the system's self-evolution.

[0391] In this embodiment, the system implements parameter deployment, pushing parameter configurations to a distributed configuration center (such as Apollo / Nacos), supporting the system to dynamically load new parameters without restarting. A batch deployment strategy is implemented according to service instances or regions to ensure that relevant parameters are updated simultaneously as a whole, avoiding inconsistent states.

[0392] The system performs canary release and control, initially switching 10% of traffic to the new parameter configuration, while running both the old and new parameter configurations simultaneously, collecting comparative data, and comparing the performance differences of the old and new configurations on key indicators in real time. Based on the initial effect evaluation, the application ratio of the new configuration is gradually expanded.

[0393] The system performs comprehensive performance monitoring, tracks changes in core performance indicators such as success rate, latency, and throughput, observes changes in system resource (CPU, memory, network) utilization, calculates changes in the processing cost per unit message, closely monitors abnormal indicators such as error rate and timeout rate, and evaluates changes in end-user experience through sampling or feedback channels.

[0394] The system performs data analysis and effect evaluation, calculates the percentage improvement of key indicators, verifies whether the performance improvement is statistically significant, evaluates the impact of parameter optimization on abnormal scenarios and edge cases, and quantifies the ratio of investment (such as computing resources and development costs) to performance improvement.

[0395] The system establishes a closed-loop feedback and continuous optimization mechanism, using monitoring results as input data for the next round of learning. It identifies potential problems or limitations introduced by new parameters, adjusts the focus of the next round of optimization based on actual results, and sets appropriate learning cycles according to system stability and the rate of environmental change. Through this closed-loop mechanism, the system can continuously evolve, adapt to environmental changes, and constantly improve its performance.

[0396] This invention also provides a decentralized multi-agent reinforcement learning architecture, and applies reinforcement learning algorithms for parameter optimization to generate optimized weight coefficients and scheduling strategies, including:

[0397] Step S801: Establish a multi-agent framework, in which multiple distributed agents are deployed, including a channel evaluation agent, a routing decision agent, a retry policy agent, and a resource scheduling agent, with each agent distributed across different physical nodes.

[0398] In this embodiment, the system replaces the single central learning system in the original scheme, establishing a fully decentralized multi-agent reinforcement learning (MARL) architecture. In this architecture, the system deploys multiple distributed agents, each responsible for different aspects of decision optimization:

[0399] Channel evaluation agent: responsible for optimizing the channel health score calculation strategy and adjusting the weight coefficients of various dimensional indicators;

[0400] Routing decision agent: responsible for message routing and channel selection strategies, and optimizing matching algorithms;

[0401] Retry policy agent: responsible for retry decisions on failure messages and optimizing decision tree parameters;

[0402] Resource scheduling agent: responsible for optimizing the allocation of system resources and balancing the resource demands of messages with different priorities;

[0403] These agents are distributed across different physical nodes, each possessing independent learning capabilities and decision-making authority. They collaborate to optimize the overall system performance. This distributed deployment improves system reliability, avoids single points of failure, and allows each agent to focus on its own domain, providing more refined decision-making capabilities.

[0404] Step S802: Establish a minimal communication mechanism among agents in the multi-agent framework, exchange only local reward information, and achieve secure data exchange through an encrypted communication channel to obtain a distributed cooperative network.

[0405] In this embodiment, the system designs an efficient inter-agent communication mechanism, adopting the principle of minimizing communication. Agents only exchange necessary local reward information, rather than complete states, significantly reducing communication overhead. The system implements Federated Learning-style distributed model updates, where each agent learns locally and periodically exchanges model parameters or gradient information.

[0406] The system employs encrypted communication channels to ensure the security of data exchange between agents, preventing the leakage or tampering of sensitive information. The system also incorporates fault-tolerant design; even if a single agent fails, it will not affect the overall system operation. Other agents can continue to function and, upon the recovery of the failed agent, will help it quickly synchronize its latest state.

[0407] Through this design, the system forms an efficient, secure, and reliable distributed collaborative network, in which each agent can maintain relative independence while working together to optimize system performance.

[0408] Step S803: In the distributed cooperative network, a participant-critic network structure is constructed for each agent. The participant network is used to learn the optimal policy mapping and output action selection, while the critic network is used to evaluate the value of actions and provide policy gradient guidance.

[0409] In this embodiment, the system constructs an Actor-Critic network structure for each agent, which is a reinforcement learning architecture that combines policy gradients and value functions. In this structure:

[0410] Actor Network: Responsible for learning the optimal policy mapping, which is a mapping function directly from state to action, and outputs the specific action selection;

[0411] The critic network is responsible for evaluating the value of actions, learning the state-action value function, and providing policy gradient guidance to the participant network.

[0412] The network architecture employs an LSTM+attention mechanism suitable for time-series decision-making, capable of capturing long-term dependencies in time-series data and focusing on the most relevant features. The system implements a hierarchical training strategy, first training on local data and then aggregating the model through a consensus mechanism to improve learning efficiency.

[0413] The system also sets differentiated learning rates for different agents, adjusting learning parameters based on the rate of environmental change and task complexity. For example, a channel evaluation agent might use a lower learning rate to maintain stability, while a retry strategy agent might use a higher learning rate to adapt quickly to changes.

[0414] This dual-network architecture enables the system to learn "what to do" (strategy) and "how good" (value) simultaneously, improving learning efficiency and strategy quality.

[0415] Step S804: In the participant-critic network structure, design a hybrid reward structure that includes individual rewards and team rewards. Maintain policy diversity through policy entropy regularization and establish a collaborative behavior incentive mechanism.

[0416] In this embodiment, the system is designed with a hybrid reward structure, which consists of two parts:

[0417] Individual rewards: Rewards based on the agent's own task completion performance, such as channel evaluation agents receiving rewards based on the accuracy of their channel selection;

[0418] Team rewards: Shared rewards based on the overall system performance, such as improved overall success rate and reduced costs;

[0419] This hybrid reward structure encourages agents to optimize their own tasks while promoting overall collaboration, thus preventing agents from sacrificing overall system performance in order to maximize individual rewards.

[0420] The system implements policy entropy regularization, which encourages agents to maintain policy diversity by adding a policy entropy term to the objective function, thus preventing premature convergence to suboptimal policies. This mechanism is particularly suitable for dynamic environments, enabling the system to continuously explore new possibilities.

[0421] The system also establishes a collaborative behavior incentive mechanism, specifically rewarding collaborative behaviors that benefit overall system performance, such as resource concession and information sharing. For example, when a resource scheduling agent yields resources for high-priority messages during peak periods, it receives an additional reward.

[0422] Through this design, the system balances competition and cooperation, maintaining the autonomy of each agent while ensuring the consistency of the overall goal.

[0423] Step S805: Implement distributed model updates through consensus algorithms, support dynamic adjustment of the number of agents according to the system scale, realize cross-scenario knowledge transfer, and obtain global optimization parameters.

[0424] In this embodiment, the system employs a consensus algorithm to achieve distributed model updates. Each agent periodically shares its model parameters or gradients, and a consensus model is formed through weighted averaging or other aggregation methods. This approach retains the efficiency of local learning while incorporating global knowledge, thus improving the quality of learning.

[0425] The system supports dynamic adjustment of the number of agents based on system scale. During large-scale deployments, the number of agents can be increased to improve processing power and decision-making accuracy; conversely, the number of agents can be reduced in small-scale scenarios to lower computational overhead. This flexible architecture enables the system to adapt to business needs of varying scales.

[0426] The system enables cross-scenario knowledge transfer, allowing agents to apply knowledge learned in one scenario to a new one, accelerating the learning process. For example, SMS scheduling strategies learned on an e-commerce platform can be transferred to a financial scenario, quickly adapting to the new environment after minor adjustments.

[0427] Through these mechanisms, the system can continuously optimize parameters, adapt to environmental changes, and achieve true adaptive learning. The final globally optimized parameters reflect both the expertise of each agent and maintain overall consistency.

[0428] Step S806: Based on the global optimization parameters, the channel evaluation agent outputs the optimized channel score weight coefficient, the routing decision agent outputs the optimized channel selection strategy, the retry strategy agent outputs the optimized retry decision parameters, and the resource scheduling agent outputs the optimized resource allocation strategy, thereby obtaining the optimized weight coefficient and scheduling strategy.

[0429] In this embodiment, the system outputs the optimization results for each specialized intelligent agent in its responsible domain based on global optimization parameters:

[0430] The channel evaluation agent outputs optimized channel score weight coefficients (w1~W4), making the channel health score calculation more accurate;

[0431] The routing decision agent outputs an optimized channel selection strategy, including matching algorithm parameters and utility function weights;

[0432] The retry strategy agent outputs optimized retry decision parameters, including decision tree threshold, retry interval strategy, etc.

[0433] The resource scheduling agent outputs optimized resource allocation strategies, including resource ratios for messages of different priorities and flow control parameters.

[0434] These optimization results are integrated into a complete and consistent scheduling strategy configuration. Compared with single-agent solutions, multi-agent architecture can provide more refined and specialized optimization results, with each agent playing its maximum role in its area of ​​expertise.

[0435] Experimental results show that, under large-scale deployment, this solution can reduce the overall system decision latency by 30% while improving decision accuracy by 5-8%. More importantly, this decentralized architecture improves the system's reliability and scalability, enabling it to adapt to more complex and dynamic business scenarios.

[0436] This invention also provides an optimal matching algorithm applied to a candidate set of channels, calculating the utility value of each channel through a utility function, selecting the channel with the highest utility value, and obtaining the matching result between the message and the channel, including:

[0437] Step S901: Based on the state information of the channel candidate set, construct a dual deep Q network architecture, including a main network responsible for action selection and a target network for calculating the target Q value, and integrate multi-dimensional information such as channel health score, message characteristics and system load as state representation.

[0438] In this embodiment, the system constructs a Double Deep Q-Network (DDQN) architecture, which includes two complementary networks:

[0439] Main network (Q-Network): Responsible for action selection, interacts directly with the environment, and updates frequently;

[0440] Target Network: Used to calculate the target Q-value, providing a stable learning target, and updated at a low frequency;

[0441] The network topology employs a deep residual network structure, containing 6-8 convolutional layers and fully connected layers, effectively handling high-dimensional input data. The system integrates multi-dimensional information as a state representation, including:

[0442] Channel Health Score: Real-time health score for each channel and its components (success rate, latency, cost, quota).

[0443] Message characteristics: priority, content type, length, receiver characteristics, etc.;

[0444] System load: current queue length, processing latency, resource utilization, etc.

[0445] Time characteristics: time period, date type (weekday / weekend / holiday), etc.;

[0446] The system defines a fine-grained action space, including possible channel selection and parameter configuration combinations, such as selecting a specific channel, setting specific transmission parameters, and adjusting retry strategies.

[0447] This dual-network architecture effectively solves the overestimation problem in traditional Q-learning, improving the stability and efficiency of learning. The state representation, which integrates multi-dimensional information, enables the system to comprehensively consider various factors and make more accurate decisions.

[0448] Step S902: Implement a greedy strategy optimization operation on the dual deep Q network architecture, select the action with the highest current Q value, retain the ε probability for random exploration, gradually reduce the exploration probability according to a preset ratio as the learning progresses, and output the greedy action selection strategy.

[0449] In this embodiment, the system implements a greedy strategy optimization operation, employing an ε-greedy strategy to balance exploration and exploitation:

[0450] Greedy action selection: With a probability of (1-ε), the system selects the action with the highest current Q value, that is, to maximize the expected reward;

[0451] Random exploration: With a probability of ε, the system randomly selects an action to explore the unknown state-action space;

[0452] The system implements an adaptive ε decay mechanism. Initially, a high ε value (e.g., 0.3) is set to encourage exploration. As learning progresses, the ε value is gradually reduced according to a preset ratio, eventually dropping to a lower level (e.g., 0.05) to make better use of the learned knowledge.

[0453] The system also implements a prioritized experience replay mechanism, assigning higher sampling probabilities to important experience samples (such as those causing large errors) to accelerate the learning of key knowledge. The system maintains an experience pool sorted by TD error, prioritizing the sampling of high-error samples from it for learning.

[0454] By optimizing the system using this greedy strategy, the system can strike a balance between exploring new strategies and utilizing known effective strategies. This prevents it from converging to a suboptimal solution too early and allows it to make full use of the learned knowledge, thereby improving decision-making efficiency.

[0455] Step S903: Perform time series analysis based on the greedy action selection strategy to identify traffic characteristics in different time periods, make routing decisions considering the network topology relationship between channels, monitor the bandwidth, latency and reliability of each link in real time, and generate a link activation scheme based on traffic and topology.

[0456] In this embodiment, the system collects historical traffic data for each time period, identifies periodic traffic patterns (such as daily peaks and weekend characteristics) and sudden traffic characteristics through an LSTM time series model, and predicts the distribution of traffic demand in future time periods, providing a basis for resource planning.

[0457] The system constructs a network topology map between channels based on traffic prediction results, records the geographical location, network latency and bandwidth capacity of each channel, calculates the path cost and reachability matrix between channels, and forms a complete topology relationship model.

[0458] The system combines a topology model to monitor the current load rate, average latency, and packet loss rate of each link in real time, identify potential network bottlenecks and fault risks, and provide real-time basis for routing decisions.

[0459] The system implements a predictive activation strategy based on link status assessment, which activates backup links and expands channel capacity in advance during peak traffic periods, and releases redundant resources and reduces operating costs during off-peak traffic periods, thereby optimizing resource utilization efficiency.

[0460] The system combines predictive activation strategies with real-time traffic scheduling to dynamically adjust the weight allocation and load balancing parameters of each link, forming a complete link activation scheme based on traffic and topology. This scheme can predictively manage system resources, expand capacity in advance during peak periods, and release resources in a timely manner during off-peak periods, significantly improving system throughput and resource utilization.

[0461] In this embodiment, a time series analysis is performed based on a greedy action selection strategy to identify traffic characteristics at different time periods. Routing decisions are made considering the network topology relationships between channels. The bandwidth, latency, and reliability of each link are monitored in real time. The specific implementation of the link activation scheme based on traffic and topology is as follows:

[0462] First, the system collects historical traffic data for various time periods, including hourly traffic within 24 hours, daily traffic variations for each day of the week, and special traffic patterns during holidays. The system analyzes this data using a Long Short-Term Memory (LSTM) time-series model. This model comprises a three-layer LSTM structure with 128 neurons per layer and is equipped with an attention mechanism to capture traffic characteristics at key time points. Through this model, the system can identify the dual-peak patterns of weekday traffic (9:00-11:00 and 14:00-16:00), the gradual distribution of traffic on weekends, and the sudden surges in traffic before and after holidays. Based on the LSTM model trained on historical data, the system can predict the distribution of traffic demand for each time period within the next 24 hours with an accuracy rate exceeding 92%, generating traffic prediction results.

[0463] Based on traffic prediction results, the system constructs a network topology map between channels. This topology map adopts a weighted directed graph structure, where nodes represent SMS channel servers and edges represent connections between channels. The system records the geographical location, network latency (end-to-end latency obtained through ICMP probing, typically within the range of 10-200ms), and bandwidth capacity (e.g., 100Mbps, 1Gbps, etc.) of each channel. The system uses Dijkstra's algorithm to calculate the shortest path cost between channels and constructs an N×N reachability matrix (N being the number of channels). The matrix elements represent the reachability and cost from the source channel to the target channel, resulting in a complete topology model.

[0464] The system combines a topology model with distributed probe nodes to monitor the performance metrics of each link in real time, including current load rate (utilization percentage), average latency (milliseconds), and packet loss rate (percentage). The system employs a sliding window algorithm (window size of 5 minutes, step size of 1 minute) to analyze the changing trends of these metrics and uses anomaly detection algorithms (based on Z-score and DBSCAN clustering) to identify potential network bottlenecks (such as links with a load rate exceeding 85%) and fault risks (such as links with a sudden increase in packet loss rate exceeding 5%). The system categorizes the detection results into three levels: normal, warning, and dangerous, generating a link status assessment report.

[0465] Based on link status assessment, the system implements a predictive activation strategy. Thirty minutes before a predicted traffic peak (e.g., 10:00 AM on a weekday), the system automatically activates backup links (typically 50% of the primary link's capacity) and expands channel capacity (by calling the cloud service provider's elastic scaling function via API). Specific implementations include preheating backup servers, increasing the number of processing threads, and expanding the database connection pool. During low traffic periods (e.g., 2:00 AM - 5:00 AM), the system gradually releases redundant resources (e.g., shutting down some service instances) and switches to an economy channel to reduce operating costs. The system evaluates the strategy's effectiveness using resource utilization and cost-effectiveness metrics and records historical decision results for continuous optimization.

[0466] Finally, the system combines predictive activation strategies with real-time traffic scheduling to form a complete link activation scheme. The system uses a load balancing algorithm combining weighted round-robin and minimum connection count, dynamically adjusting the weight allocation of each link based on its status (weight range 1-100, step size 5). When a link's performance degrades, the system automatically reduces its weight and transfers traffic to a better-performing link. The system also implements an adaptive load balancing parameter adjustment mechanism; for example, connection timeout (range 5-30 seconds) and retry interval (range 1-10 seconds) are automatically adjusted based on network conditions. Through this dynamic adjustment mechanism, the system can optimize resource utilization efficiency while ensuring message delivery rate, achieving an intelligent link activation scheme based on traffic and topology.

[0467] Step S904: Establish a hierarchical scheduling architecture based on the link activation scheme based on traffic and topology, including a master scheduler responsible for global resource allocation and a sub-scheduler focused on local optimization, to achieve efficient state sharing and decision coordination, and obtain a multi-agent DDQN allocator.

[0468] In this embodiment, the system establishes a hierarchical scheduling architecture, forming a multi-layered decision-making system:

[0469] Master Scheduler: Responsible for global resource allocation and policy coordination, with a system-level perspective, and formulates overall policies;

[0470] Sub-schedulers: Focus on local optimization for specific regions or business lines, offering more granular control capabilities;

[0471] The system implements a hierarchical decision-making process. Global decision-making determines the main strategies and resource allocation schemes, while local decision-making processes the specific execution details. The two work together to ensure both global optimization and local characteristics.

[0472] The system establishes an efficient communication mechanism between schedulers to achieve state sharing and decision coordination, reducing conflicts and resource waste. The system defines clear priority rules to handle potential decision conflicts and ensure a consistent response in emergency situations.

[0473] This hierarchical architecture creates a multi-agent DDQN allocator, where each level of scheduler can work independently or collaboratively, improving system decision-making efficiency and resource utilization. This architecture is particularly suitable for large-scale, geographically distributed SMS dispatching systems, simultaneously considering both global optimization and local characteristics.

[0474] Step S905: Apply neural network quantization and GPU acceleration technology to the multi-agent DDQN allocator to implement a caching mechanism for scene decision results, process similar requests in batches, and obtain the matching results of the message and channel.

[0475] In this embodiment, the system optimizes the performance of the multi-agent DDQN allocator by applying neural network quantization technology to convert floating-point weights into low-precision integers (such as 8-bit or 4-bit), which greatly reduces the model size and computational complexity while maintaining decision accuracy.

[0476] The system leverages GPU / TPU accelerators to enhance inference speed and parallelize neural network computations, achieving performance improvements of several to tens of times, especially when processing batch requests. For scenarios that do not require real-time response, the system can further improve energy efficiency using dedicated AI acceleration chips.

[0477] The system implements a caching mechanism for scenario-based decision results. Decision results for common scenarios are cached, and when a similar request is encountered, the cached result is returned directly, avoiding duplicate calculations. The system uses an efficient similarity calculation algorithm to quickly identify reusable decision results.

[0478] The system processes similar requests in batches instead of one by one, fully utilizing the parallel computing capabilities of the hardware to improve system throughput. This batch processing mechanism is particularly effective for bulk marketing messages, significantly improving processing efficiency.

[0479] Through these optimization techniques, the greedy DDQN scheduler can make optimal scheduling decisions at the millisecond level, reducing decision latency by more than 60% compared to the rule-based selection in the original scheme, while increasing system throughput by 45% and resource utilization by 30%.

[0480] The present invention also provides: executing the off-peak sending plan to each region in the region grouping results during a specified off-peak window, recording compression rate, sending cost, and delivery rate data, calculating the cost savings compared to the standard sending method, and before obtaining the cost savings data, further comprising:

[0481] Step S1001: Based on the idle time sending plan, construct a dual-sum duel depth Q-Learning architecture, use two independent Q networks to reduce overestimation bias, decompose the Q value into a state value function V(s) and an advantage function A(s,a), and calculate the Q value to obtain the D3QL network structure.

[0482] In this embodiment, the system constructs a dual-duel deep Q-Learning (D3QL) architecture, which is an innovative reinforcement learning network structure that combines the advantages of dual Q-Learning and duel networks.

[0483] Dual Q-Learning uses two independent Q-networks to reduce overestimation bias. One network is used to select actions, and the other network is used to evaluate the value of the selected actions, thus avoiding the positive bias problem in traditional Q-Learning.

[0484] The duel network decomposes the Q-value into a state value function V(s) and an advantage function A(s,a):

[0485] Q(s,a) = V(s) + A(s,a) - mean[A(s,a')]

[0486] Where V(s) represents the value of being in state s, regardless of the action chosen; A(s,a) represents the advantage of choosing action a relative to the average level in state s; mean[A(s,a')] is the average advantage value of all actions.

[0487] This decomposition enables the network to learn the intrinsic value of states and the relative advantages of different actions separately, improving learning efficiency and generalization ability. The system adopts a hierarchical state representation, constructing a state vector that includes system states, service requirements, and resource constraints, comprehensively capturing environmental information.

[0488] The system implements a progressive training strategy, learning from simple to complex scenarios to accelerate network convergence. Through these techniques, the system achieves an efficient D3QL network structure, providing decision support for subsequent service placement and resource scheduling.

[0489] Step S1002: Implement a multi-level service deployment strategy using the D3QL network structure, dividing the processing logic into a core layer, an edge layer, and a terminal layer. The core layer is used to process key business logic and data storage, the edge layer is used to process localized message processing and caching, and the terminal layer is used to implement preprocessing and status detection, thus determining the layered service architecture.

[0490] In this embodiment, the system utilizes the D3QL network structure to implement a multi-level service deployment strategy, distributing SMS processing services across three levels:

[0491] Core layer: Located in the central data center, it handles critical business logic, global decision-making, and data storage, and has the highest computing power and reliability;

[0492] Edge layer: Located at the network edge nodes (such as regional data centers), it handles localized message processing and caching, reduces network latency, and improves response speed;

[0493] Terminal layer: Located closest to the user, it performs preprocessing and status detection, such as message format verification and terminal status detection.

[0494] This layered architecture combines edge computing with cloud computing, leveraging the powerful computing capabilities of the cloud while utilizing the low latency of edge nodes. The system dynamically determines which layer each message processing task should be executed at, intelligently scheduling tasks based on message priority, computational complexity, and network conditions.

[0495] For example, for P0-level verification code messages, the system may process and send them directly at the edge layer to avoid delays at the central node; while for marketing messages that require complex analysis, they may be processed at the core layer and then distributed to the edge layer for sending.

[0496] This layered service architecture significantly reduces end-to-end latency, improves resource utilization efficiency, and maintains consistency and manageability of processing logic.

[0497] Step S1003: Based on the intelligent prediction of service demand distribution in the layered service architecture, deploy service instances in advance at the optimal location, and dynamically adjust the distribution of service instances according to load changes to obtain an intelligent service placement strategy.

[0498] In this embodiment, the system implements an intelligent service placement strategy based on a hierarchical service architecture and predicts the future distribution of service demand through a D3QL network, including message processing needs in different regions and at different times.

[0499] The system pre-deploys service instances in areas predicted to have high demand to ensure sufficient processing capacity when peak demand arrives. For example, if a large number of marketing messages are predicted to be sent in a certain region, additional processing instances are deployed in advance on edge nodes in that region.

[0500] The system dynamically adjusts the distribution of service instances based on actual load changes, automatically scaling up when the load increases and automatically scaling down when the load decreases, maintaining an optimal balance in resource utilization. The system considers service migration costs, comprehensively evaluating computing resources, network bandwidth, and migration overhead when deciding whether to scale up existing instances or deploy instances in new locations.

[0501] The system also implements a service affinity strategy to maintain the geographical proximity of related services and reduce communication latency between services. For example, message processing services and data caching services are deployed on the same edge node to avoid cross-node data access.

[0502] Through this intelligent service placement strategy, the system can predictively manage resource distribution, minimizing resource costs while ensuring service quality, reducing average response time by 40%, and increasing resource utilization by 50%.

[0503] Step S1004: Define differentiated service quality requirements for different service types according to the intelligent service placement strategy, reserve a dedicated resource pool for the first priority service, and dynamically adjust QoS parameters according to load and service importance.

[0504] In this embodiment, the system implements a strict Quality of Service (QoS) guarantee mechanism based on an intelligent service placement strategy. The system defines differentiated QoS requirements for different service types, such as:

[0505] P0-level CAPTCHA messages: 99.99% availability, delivery in ≤3 seconds, and a loss rate of ≤0.1%.

[0506] P1 level notification messages: 99.9% availability, delivery time ≤15 seconds, loss rate ≤0.5%;

[0507] P2 level marketing messages: 99% availability, delivery within ≤30 minutes, and a loss rate of ≤2%;

[0508] The system reserves a dedicated resource pool for first-priority services (such as P0-level CAPTCHAs) to ensure that these critical services receive sufficient processing resources even during periods of high system load. This resource reservation mechanism employs an elastic quota model, allowing other services to use the reserved resources during periods of low load, but enabling immediate reclamation when needed.

[0509] The system dynamically adjusts QoS parameters based on current load and business importance. When system pressure increases, it may appropriately relax the QoS requirements for non-critical services to ensure the normal operation of core services. For example, during peak traffic periods, the system may temporarily adjust the delivery time requirement for P2 level messages from 30 minutes to 60 minutes to free up resources for higher priority messages.

[0510] The system implements QoS monitoring and alarm mechanisms, tracks the service quality indicators of various services in real time, triggers early warnings when indicators approach the threshold, triggers alarms when service level agreements (SLAs) are violated, and automatically takes remedial measures, such as increasing resources, adjusting routes, or downgrading.

[0511] Through this differentiated QoS management, the system can ensure the service quality of critical services while reasonably handling non-critical services, achieving the optimal balance of overall service quality, even with limited resources.

[0512] Step S1005: Adopt a microservice architecture and containerized deployment method, use Kubernetes to manage the dynamic scaling of service instances, minimize the state dependencies between services, and test the system's adaptability through fault injection.

[0513] In this embodiment, the system adopts a modern microservice architecture, breaking down the SMS dispatch system into multiple independent functional services, such as channel management service, message routing service, retry decision service, and monitoring service. Each service is responsible for a specific function and communicates with each other through standard APIs, achieving a loosely coupled design.

[0514] The system employs a containerized deployment approach, packaging each microservice into container images to ensure environmental consistency and deployment flexibility. Kubernetes is used as the container orchestration platform to manage the lifecycle of service instances, enabling automated deployment, scaling, and fault recovery.

[0515] The system is designed with a horizontally scalable architecture, addressing load growth by increasing the number of service instances rather than increasing the size of individual instances, thus improving system scalability and elasticity. The system implements automatic scaling based on metrics, automatically adjusting the number of service instances according to indicators such as CPU utilization, memory usage, and request queue length.

[0516] The system minimizes state dependencies between services, adopts a stateless design principle, and stores state information in a distributed cache or database, allowing service instances to scale freely without affecting system functionality. For states that must be maintained, the system uses a distributed state management solution to ensure state consistency and reliability.

[0517] The system tests its adaptability through fault injection, simulating various failure scenarios (such as node crashes, network partitions, and service timeouts) to verify its automatic recovery capabilities and the effectiveness of its service degradation strategies. This "chaos engineering" practice enables the system to maintain stable operation when faced with real-world failures.

[0518] Through these modern architectures and deployment methods, the system achieves unprecedented scalability, resilience, and reliability, enabling it to easily meet the needs of various scenarios, from small businesses to ultra-large-scale applications.

[0519] This invention also provides a strategy for placing intelligent services using a greedy multiple access scheme, including:

[0520] Step S1101: Based on the intelligent service placement strategy, key messages are transmitted in parallel using multiple independent channels simultaneously. Forward error correction technology is applied to enhance transmission reliability and generate a multi-channel parallel transmission mechanism.

[0521] In this embodiment, the system implements a multi-channel parallel transmission mechanism for critical messages (such as P0-level verification codes) based on an intelligent service placement strategy. The system simultaneously uses multiple independent channels (usually 2-3) to send the same message. As long as any channel successfully delivers the message, it is considered a successful transmission, significantly improving the reliability of critical message delivery.

[0522] The system employs forward error correction (FEC) technology to enhance transmission reliability. For particularly important messages, the system encodes the original message into multiple redundant segments, so that even if some segments are lost during transmission, the complete message can be reconstructed from the remaining segments. This technology is particularly suitable for scenarios with unstable network conditions.

[0523] The system implements intelligent redundancy control, dynamically adjusting the redundancy level based on message importance, network conditions, and historical delivery rates. For example, it may use only two channels when network conditions are good, but increase to three or more channels when network fluctuations are severe.

[0524] The system also implements a cross-carrier distributed strategy, ensuring that multiple channels used in parallel belong to different carriers or network paths, maximizing path independence and avoiding the risk of single points of failure. For example, sending the same message simultaneously using channels from different carriers.

[0525] Through this multi-channel parallel transmission mechanism, the system provides nearly 100% reliability for critical messages, maintaining a 99.99% delivery rate even under severe network fluctuations.

[0526] Step S1102: Utilize the multi-channel parallel transmission mechanism to evaluate the success rate, latency, and stability of each transmission path in real time, dynamically adjust the traffic allocation ratio based on path quality, prioritize the use of the best-performing path, and obtain an adaptive path selection strategy.

[0527] In this embodiment, the system implements an adaptive path selection strategy based on a multi-channel parallel transmission mechanism. The system continuously monitors and evaluates the key performance indicators of each transmission path (channel) in real time, including:

[0528] Success rate: The percentage of messages that are successfully delivered;

[0529] Delay: The time from sending a request to receiving an acknowledgment;

[0530] Stability: The degree of fluctuation in performance indicators;

[0531] Throughput capacity: The number of messages that can be processed per unit of time;

[0532] Based on these real-time evaluation results, the system dynamically adjusts the traffic allocation ratio for each path. The best-performing path receives a higher traffic share, while the worst-performing path receives less traffic or is temporarily not used. This dynamic adjustment is performed every 5-15 minutes to ensure that the system always uses the currently optimal transmission path.

[0533] The system implements predictive path assessment, taking into account not only current performance but also historical performance trends and predicted future states. For example, if the performance of a certain channel shows a gradual downward trend, the system will reduce the use of that channel in advance to avoid potential service quality degradation.

[0534] The system also considers path diversity; even if some paths currently have slightly lower performance, a certain amount of traffic will be reserved to continuously acquire performance data and maintain the path's activity. This strategy avoids the single point of dependence risk caused by a "winner-takes-all" scenario.

[0535] Through this adaptive path selection strategy, the system can intelligently utilize multiple transmission paths, minimizing latency and resource consumption while ensuring delivery reliability.

[0536] Step S1103: Based on the adaptive path selection strategy, taking into account channel capacity, network latency and transmission cost, a weighted round-robin algorithm is applied to achieve intelligent load distribution, resulting in a load balancing algorithm.

[0537] In this embodiment, the system implements a more refined load balancing algorithm based on an adaptive path selection strategy. The system comprehensively considers multiple key factors when making load allocation decisions:

[0538] Channel capacity: The maximum processing capacity and current load level of each channel;

[0539] Network latency: end-to-end transmission latency and latency stability of each channel;

[0540] Transmission cost: The cost of sending a single message on each channel and the overall economic efficiency;

[0541] Priority strategy: Resource allocation strategy for messages with different priorities;

[0542] The system employs an improved weighted round-robin algorithm to achieve intelligent load balancing. Unlike simple round-robin, weighted round-robin assigns different weights based on channel performance metrics, with higher-performing channels receiving higher weights and thus processing more messages. The system dynamically adjusts these weights to reflect real-time changes in channel performance.

[0543] The system implements a smooth load shifting mechanism. When traffic allocation needs to be adjusted, a gradual shift is used instead of an abrupt switch to avoid sudden pressure or traffic fluctuations on the channels. For example, when the load share of a channel is reduced from 40% to 20%, it will be adjusted gradually over several minutes, rather than changing instantaneously.

[0544] The system also enables load prediction and advance adjustment. Based on historical traffic patterns, it predicts future load and adjusts resource allocation in advance to avoid delays caused by temporary adjustments during peak load periods. For example, before predicting an upcoming traffic peak, the system will increase the weight of high-performance channels in advance.

[0545] Through this sophisticated load balancing algorithm, the system can maximize resource utilization efficiency, maintain a reasonable distribution of load across channels, and avoid overload or waste of resources on a single channel.

[0546] Step S1104: Establish a channel failure detection and switching mechanism based on the load balancing algorithm, and automatically switch to the backup channel when the main channel is detected to be abnormal.

[0547] In this embodiment, the system establishes a robust channel fault detection and switching mechanism based on a load balancing algorithm. The system implements multi-level fault detection:

[0548] Proactive health checks: Regularly send probe requests to each channel to verify channel availability;

[0549] Passive performance monitoring: Analyze the success rate, latency, and error patterns of actual message sending;

[0550] Anomaly pattern recognition: Using machine learning algorithms to identify abnormal performance patterns and predict potential failures;

[0551] External alarm integration: Receive and process fault alarms from carriers or monitoring systems;

[0552] The system defines clear fault judgment criteria, including hard faults (such as connection interruption and authentication failure) and soft faults (such as decreased success rate and increased latency). The system sets multiple threshold levels, such as minor anomalies (success rate drops below 95%), severe anomalies (success rate drops below 85%), and complete failures (success rate drops below 50% or connection is interrupted).

[0553] The system implements an intelligent fault response strategy:

[0554] For minor anomalies, reduce the traffic allocation to that channel and increase the monitoring frequency;

[0555] For severe anomalies, most traffic will be diverted to the healthy channel, while a small amount of traffic will be reserved for status detection.

[0556] In the event of a complete failure, immediately switch all traffic to the backup channel and trigger an alarm notification.

[0557] The system maintains historical records of channel status, enabling historical reliability assessments and recovery strategies. For channels that frequently fail, the system lowers their priority and uses them cautiously even after recovery.

[0558] Through this advanced fault detection and switching mechanism, the system can respond quickly when a problem occurs in the channel, minimize service interruption, and ensure the continuous and reliable transmission of messages.

[0559] Step S1105: Combine the channel fault detection and switching mechanism with the multi-channel parallel transmission mechanism, the adaptive path selection strategy and the load balancing algorithm to obtain the greedy multiple access scheme.

[0560] In this embodiment, the system organically integrates the aforementioned mechanisms to form a complete greedy multiple access scheme. This scheme integrates technologies such as multi-channel parallel transmission, adaptive path selection, intelligent load balancing, and fault detection switching to form a coordinated end-to-end solution.

[0561] The system implements a unified decision-making framework to coordinate the operation of various subsystems, ensuring consistency in decisions and optimal allocation of resources. For example, when the adaptive path selection strategy decides to increase the usage ratio of a certain channel, the load balancing algorithm will adjust the weight accordingly, and the fault detection mechanism will strengthen the monitoring of that channel.

[0562] The system establishes a global resource view to monitor the status, capacity, and performance of each channel in real time, providing a basis for decision-making regarding greedy multiple access schemes. The system maintains a unified performance indicator system, enabling different mechanisms to be evaluated and decided upon based on the same standards.

[0563] The system implements adaptive strategy selection, dynamically choosing the most suitable transmission strategy based on different scenarios and message types:

[0564] For high-priority CAPTCHAs, a fully parallel transmission strategy may be adopted, using multiple channels simultaneously;

[0565] For regular notification messages, a primary / backup switchover strategy may be adopted, setting up one primary channel and multiple backup channels;

[0566] For marketing messages, a cost-first strategy may be adopted, prioritizing the use of cost-effective channels;

[0567] The system also enables end-to-end performance monitoring and optimization feedback, continuously evaluates the effectiveness of the greedy multiple access scheme, identifies opportunities for improvement, and continuously optimizes decision parameters through reinforcement learning.

[0568] Through this highly integrated greedy multiple access scheme, the system achieves extremely high message delivery reliability and efficiency. Even in complex and ever-changing network environments, the system can maintain stable service quality, providing a 99.99% delivery guarantee for critical messages, while optimizing resource utilization and operating costs.

[0569] In this embodiment, the specific application of the greedy multiple access scheme can be illustrated through the following example:

[0570] A financial institution needs to send a large number of verification code SMS messages. The system combines a channel failure detection and switching mechanism with a multi-channel parallel transmission mechanism, an adaptive path selection strategy, and a load balancing algorithm to form a complete greedy multiple access scheme. For example, when a user makes a transfer of 50,000 yuan, the system identifies it as a high-risk transaction and requires sending a verification code SMS. At this time, the greedy multiple access scheme first activates the multi-channel parallel transmission mechanism, simultaneously sending the same verification code through three independent channels from different operators. The system applies forward error correction coding technology to the verification code message, encoding the original 6-digit verification code "847295" into three data packets containing redundant information. Even if one data packet is partially damaged during transmission, the receiver can reconstruct the complete verification code from the remaining information.

[0571] Meanwhile, the system evaluates the performance of the three channels in real time based on an adaptive path selection strategy. In this example, the system detects that the current success rate of Carrier 1 channel is 99.7%, with an average latency of 1.2 seconds; the success rate of Carrier 2 channel is 98.5%, with an average latency of 1.8 seconds; and the success rate of Carrier 3 channel is 99.2%, with an average latency of 1.5 seconds. Based on this real-time data, the system dynamically adjusts the traffic allocation ratio, allocating 50% of the traffic to Carrier 1 channel, 30% to Carrier 3 channel, and 20% to Carrier 2 channel. This allocation ensures that most messages are transmitted through the best-performing channel, while keeping the other channels active to obtain continuous performance data.

[0572] At the load balancing level, the system uses a weighted round-robin algorithm to achieve intelligent load distribution. For example, during the peak period of 10:00 AM on a weekday, the system detects that the load rate of Carrier 1 channel reaches 75%, close to the warning threshold, while the load rate of Carrier 3 channel is only 45%. The system immediately adjusts the weights, transferring some messages originally planned to be sent through the mobile channel to Carrier 3 channel to ensure load balance across channels and avoid performance degradation caused by overload of a single channel.

[0573] In this example, the system also implemented a channel failure detection and switching mechanism. When the verification code message was sent in parallel through three channels, Carrier 2 channel returned a success response first, taking 1.5 seconds; while Carrier 1 channel still hadn't returned a response after 2 seconds. The system's health check mechanism was immediately triggered, detecting network fluctuations in a certain area node of Carrier 1 channel, with the success rate plummeting from 99.7% to 85% in the past 5 minutes. The system immediately marked this channel as "warning" and automatically switched most of the traffic originally allocated to this channel to Carrier 3 and Carrier 2 channels, while reserving a small amount of traffic for continuous monitoring of the mobile channel's recovery status.

[0574] Through this integrated solution, even under poor network conditions, the financial transaction verification code was successfully delivered to the user's mobile phone within 1.5 seconds, far below the 3-second threshold set by the system for P0-level messages, ensuring user experience and transaction security. System records show that after adopting the greedy multiple access scheme, the financial institution's verification code delivery reliability increased from 99.5% to 99.98%, and the average delivery time decreased from 2.3 seconds to 1.4 seconds, greatly improving the security of financial transactions and user satisfaction.

[0575] The present invention also provides an intelligent SMS scheduling device based on multi-dimensional dynamic optimization, including a memory and a processor, wherein the processor is used to execute the method described in any of the above-mentioned embodiments.

[0576] In one specific embodiment, the device includes:

[0577] Memory: Used to store program instructions, system configuration, historical data, and runtime data;

[0578] Processor: Used to execute program instructions and implement the steps of the above methods;

[0579] Communication interface: Used for data exchange with external systems (such as business systems, SMS channels);

[0580] Database connectors: used to connect to various databases used by the operating system;

[0581] Monitoring module: Used to monitor the system's operating status and performance indicators;

[0582] The processor executes program instructions stored in memory to perform functions such as channel scoring, message routing, transmission monitoring, retry decision-making, batch processing, and reinforcement learning. The system adopts a modular design, allowing each functional module to be independently upgraded and expanded, thus improving the system's maintainability and scalability.

[0583] The intelligent SMS scheduling method and apparatus of the present invention solve the problems of insufficient dynamic adaptability, serious resource waste, lack of multi-objective coordination and high operation and maintenance costs of traditional SMS scheduling systems by multi-dimensional dynamic channel scoring, intelligent retry decision based on terminal status perception, cost-time balance batch processing and closed-loop self-optimization system. It significantly improves the SMS delivery rate and reduces the invalid retry rate and sending cost.

[0584] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent short message dispatching method based on multi-dimension dynamic optimization, characterized in that, The method comprises the following steps: obtaining performance data of a plurality of short message channels within a preset time, the performance data including success rate, delay, cost and quota usage, performing cleaning and normalization processing on the performance data, calculating health scores of the plurality of short message channels based on a weight dynamic adjustment model, and obtaining channel health evaluation results; receiving a message to be sent and a priority identifier thereof, the priority identifier including P0-level verification code messages, P1-level notification messages and P2-level marketing messages, determining scheduling strategy requirements according to the priority identifier, performing channel screening and optimal matching based on the channel health evaluation results and the scheduling strategy requirements, and obtaining a matching result of the message and the channel; performing message sending based on the matching result of the message and the channel, monitoring sending status and channel response in real time, receiving and analyzing sending result data, and obtaining sending result information; obtaining failed message data in the sending result information, performing terminal state detection through operator base station signaling, applying a decision tree model to determine a retry strategy, obtaining a retry decision result, and screening P2-level marketing messages that fail to retry from the retry decision result; performing Huffman coding compression processing on the P2-level marketing messages that fail to retry, performing regional grouping according to receiver base station information, performing batch sending operation of the P2-level marketing messages that have been compressed in an idle time window, obtaining sending result information after the operation is completed, and corresponding cost saving data.

2. The method of claim 1, wherein, The method further comprises: constructing a comprehensive performance evaluation index using the sending result information and the cost saving data, constructing a reward function model based on the comprehensive performance evaluation index, applying a reinforcement learning algorithm to optimize parameters, generating optimized weight coefficients and scheduling strategies, and updating short message scheduling operations based on the scheduling strategies.

3. The method of claim 1, wherein, The method further comprises: performing cleaning and normalization processing on the performance data, calculating health scores of the plurality of short message channels based on a weight dynamic adjustment model, and obtaining channel health evaluation results, including: performing outlier detection and cleaning processing on the performance data to obtain a cleaned data set; performing normalization processing on success rate, delay, cost and quota usage indicators in the cleaned data set, and uniformly converting multi-dimensional index data to the 0-1 interval to obtain a normalized data set; selecting a preset weight template based on time period characteristics according to current time period characteristics, determining success rate weight, delay weight, cost weight and quota weight, and obtaining weight coefficient configuration; 4. The method of claim 1, wherein, based on the normalized data set and the weight coefficient configuration, calculating health score values of the plurality of short message channels based on a weight dynamic adjustment model, and obtaining the channel health evaluation results. The method further comprises: receiving and analyzing the content of the message to be sent, identifying the priority identifier corresponding to the content of the message to be sent, and obtaining a priority classification result; determining a corresponding scheduling policy requirement according to the priority classification result, to obtain a policy parameter configuration, wherein the scheduling policy requirement comprises: a P0-level message requirement of a health score greater than or equal to 90 points and a delay less than or equal to 3 seconds, a P1-level message requirement of a health score greater than or equal to 70 points and a delay less than or equal to 15 seconds, and a P2-level message requirement of a health score greater than or equal to 50 points and a priority cost factor; performing multi-level screening in combination with the channel health evaluation result and the policy parameter configuration, respectively performing screening based on a health score threshold, screening based on a specific dimension hard condition, and availability screening based on quota management, to obtain a channel candidate set meeting the conditions; applying an optimal matching algorithm to the channel candidate set, calculating an utility value of each channel through an utility function, selecting a channel with the maximum utility value, and obtaining a matching result of the message and the channel.

5. The method of claim 1, wherein, performing message sending based on the matching result of the message and the channel, monitoring a sending state and a channel response in real time, and obtaining sending result information, comprising: preparing a message sending parameter according to the matching result of the message and the channel, performing content format conversion and security signature processing on the message sending parameter, constructing a request parameter meeting a channel API specification, and obtaining a sending request configuration; based on the sending request configuration, initiating a message sending request to a selected channel through an asynchronous non-blocking HTTP client, recording a request initiation timestamp and a tracking identifier, to obtain a sending execution state; monitoring the sending execution state in real time, tracking a complete cycle of network connection establishment, request sending, and response receiving, recording key time nodes and state changes, and obtaining state tracking data; receiving and analyzing sending result data returned by the channel, mapping a specific status code of each channel into a system unified status definition, performing success determination, temporary failure determination, and permanent failure determination, and obtaining a standardized sending result; updating a channel performance index according to the standardized sending result, recalculating a channel health score, and feeding back to a channel scoring system, to obtain the sending result information.

6. The method of claim 1, wherein, obtaining failure message data in the sending result information, performing terminal state detection through an operator base station signaling, applying a decision tree model to determine a retry strategy, and obtaining a retry decision result, comprising: obtaining failure message data in the sending result information, performing mapping analysis on error codes returned by the channel, and obtaining a failure cause classification result, wherein the failure cause classification result comprises: channel failure, terminal problem, content problem, or quota problem; based on the failure cause classification result, querying a terminal real-time state through an operator base station signaling or a manufacturer API, to obtain terminal state data, wherein the terminal state data comprises: online accessibility, shutdown inaccessible, weak and unstable signal, or roaming state; inputting the failure cause classification result and the terminal state data into the decision tree model, performing multi-layer condition determination based on a failure type, a terminal state, a message priority, and a historical retry number, and obtaining a retry strategy type; performing a corresponding retry operation according to the retry strategy type, comprising: immediate retry, channel switching, delay retry, timing retry, content adjustment, or termination retry, and obtaining a retry execution result; Record the retry execution result, update the retry effect statistics and decision model parameters, calculate the success rate indicators of different retry strategies, and obtain the retry decision result.

7. The method of claim 1, wherein, The retry failed P2 level marketing messages are subjected to Huffman coding compression processing, are grouped by region according to the base station information of the receiving party, are subjected to batch sending operation of the P2 level marketing messages after compression processing in the idle time window, and sending result information and corresponding cost saving data are obtained after the operation is completed, including: P2 level marketing messages are collected from the to-be-sent messages and the retry failed P2 level marketing messages, marketing nature message content is identified through keyword matching, time window aggregation is performed, and a P2 level message set is obtained; The P2 level message set is subjected to content compression processing by applying a Huffman coding algorithm, a character frequency dictionary and an encoding mapping table are constructed, the original messages are converted into compressed binary streams, and compressed message data are obtained; According to the mobile phone number of the receiving party, the home location and the operator information are analyzed, the operators are grouped into one level according to the operator type, the operators are grouped into two levels according to the provincial administrative division, and the operators are grouped into three levels according to the city or region, and a regional grouping result is obtained, each region corresponding to a group of to-be-sent compressed message data; An idle time window is set as the period of 00:00-06:00, a batch sending plan is made for the compressed message data corresponding to the regional grouping result, load balancing distribution is performed by using the operator discount channel, and an idle time sending plan is obtained; In the specified idle time window, the idle time sending plan is executed for each region in the regional grouping result, the compression rate, the sending cost and the delivery rate data are recorded, the cost saving range compared with the standard sending mode is calculated, and the cost saving data are obtained.

8. The method of claim 1, wherein, The method further comprises: adopting a decentralized multi-agent reinforcement learning architecture, and applying a reinforcement learning algorithm to optimize parameters, to generate optimized weight coefficients and scheduling strategies, including: A multi-agent framework is established, and a plurality of distributed agents are deployed in the framework, including a channel evaluation agent, a routing decision agent, a retry strategy agent and a resource scheduling agent, and each agent is distributed in different physical nodes; A minimum communication mechanism between agents in the multi-agent framework is established, only local reward information is exchanged, secure data exchange is realized through an encrypted communication channel, and a distributed collaboration network is obtained; In the distributed collaboration network, a participant-critic network structure is constructed for each agent, the participant network is used to learn the optimal strategy mapping and output action selection, and the critic network is used to evaluate the action value and provide strategy gradient guidance; In the participant-critic network structure, a mixed reward structure is designed, including individual reward and team reward, strategy entropy regularization is used to maintain strategy diversity, and a collaboration behavior incentive mechanism is established; Distributed model updating is realized through a consensus algorithm, the number of agents is dynamically adjusted according to the system size, cross-scene knowledge transfer is realized, and global optimization parameters are obtained. Based on the global optimization parameters, the channel evaluation agent outputs optimized channel score weight coefficients, the routing decision agent outputs optimized channel selection strategies, the retry strategy agent outputs optimized retry decision parameters, and the resource scheduling agent outputs optimized resource allocation strategies, to obtain the optimized weight coefficients and scheduling strategies.

9. The method of claim 4, wherein, An optimal matching algorithm is applied to the channel candidate set, the utility values of each channel are calculated through an utility function, the channel with the maximum utility value is selected, and a matching result of the message and the channel is obtained, including: Based on the state information of the channel candidate set, a double deep Q network architecture is constructed, including a main network responsible for action selection and a target network for calculating target Q values, and multi-dimensional information such as channel health score, message features and system load is fused as state representation; A greedy strategy optimization operation is performed on the double deep Q network architecture, the action with the highest current Q value is selected, and an epsilon probability is reserved for random exploration. The exploration probability is gradually reduced by a preset ratio as the learning progresses, and a greedy action selection strategy is output; Based on the greedy action selection strategy, time series analysis is performed to identify traffic characteristics in different time periods, routing decisions are made considering the network topology relationship between channels, and real-time monitoring of the bandwidth, delay and reliability of each link is performed to generate a link activation scheme based on traffic and topology; A hierarchical scheduling architecture is established according to the link activation scheme based on traffic and topology, including a main scheduler responsible for global resource allocation and a sub-scheduler focusing on local optimization, efficient state sharing and decision coordination are realized, and a multi-agent DDQN allocator is obtained; Neural network quantization and GPU acceleration techniques are applied to the multi-agent DDQN allocator to realize a caching mechanism for scenario decision results, and similar requests are processed in batches to obtain the matching result of the message and the channel.

10. The method of claim 9, wherein, Based on the greedy action selection strategy, time series analysis is performed to identify traffic characteristics in different time periods, routing decisions are made considering the network topology relationship between channels, and real-time monitoring of the bandwidth, delay and reliability of each link is performed to generate a link activation scheme based on traffic and topology, including: Historical traffic data in each period is collected, periodic traffic patterns and burst traffic characteristics are identified through an LSTM time series model, future traffic demand distribution is predicted, and a traffic prediction result is obtained; Based on the traffic prediction result, a network topology graph between channels is constructed, the geographical location, network delay and bandwidth capacity of each channel are recorded, the path cost and reachability matrix between channels are calculated, and a topology relationship model is obtained; Real-time monitoring of the current load rate, average delay and packet loss rate of each link is performed in combination with the topology relationship model to identify potential network bottlenecks and fault risks, and a link state evaluation is obtained; Based on the link state evaluation, standby links are activated in advance and channel capacity is expanded during traffic peak periods, and redundant resources are released and operating costs are reduced during traffic trough periods, to obtain a predictive activation strategy; The predictive activation strategy is combined with real-time traffic scheduling to dynamically adjust the weight allocation and load balancing parameters of each link, and the link activation scheme based on traffic and topology is obtained.

11. The method of claim 7, wherein, Before the designated idle time window to each region in the geographical grouping results are performed idle time sending plan, record compression rate, sending cost and delivery rate data, calculate the cost saving range compared with the standard sending mode, get the cost saving data, it further includes: Based on the idle time sending plan to build a double and duel deep Q-Learning architecture, using two independent Q network to reduce overestimation bias, decompose the Q value into state value function V(s) and advantage function A(s, a), and calculate the Q value, get the D3QL network structure; Using the D3QL network structure to realize multi-level service deployment strategy, the processing logic is divided into core layer, edge layer and terminal layer, wherein the core layer is used to process key business logic and data storage, the edge layer is used to process localized message processing and cache, and the terminal layer is used to realize preprocessing and state detection, determine the hierarchical service architecture; Based on the intelligent prediction service demand distribution in the hierarchical service architecture, deploy service instances in advance at the optimal location, dynamically adjust the service instance distribution according to the load change, get the intelligent service placement strategy; According to the intelligent service placement strategy, define the differentiated service quality requirements for different business types, reserve dedicated resource pool for first priority service, dynamically adjust QoS parameters according to load and business importance; Adopting micro service architecture and containerized deployment method, using Kubernetes to manage dynamic scaling of service instances, minimizing state dependence between services, and adapting to the system through fault injection test.

12. The method of claim 11, wherein, The method further includes: adopting a greedy multiple access scheme to obtain an intelligent service placement strategy, including: Based on the intelligent service placement strategy, use multiple independent channels for parallel transmission of critical messages at the same time, apply forward error correction code technology to enhance transmission reliability, and generate a multi-channel parallel transmission mechanism; Using the multi-channel parallel transmission mechanism to evaluate the success rate, delay and stability indicators of each transmission path in real time, dynamically adjusting the traffic distribution ratio according to the path quality, and preferentially using the best performance path, to get an adaptive path selection strategy; Based on the adaptive path selection strategy, considering the channel capacity, network delay and transmission cost, applying a weighted round robin algorithm to realize intelligent load distribution, to get a load balancing algorithm; According to the load balancing algorithm, a channel fault detection and switching mechanism is established, which automatically switches to the backup channel when the main channel is detected to be abnormal; The channel fault detection and switching mechanism is combined with the multi-channel parallel transmission mechanism, the adaptive path selection strategy and the load balancing algorithm to get the greedy multiple access scheme.

13. An intelligent short message dispatching device based on multi-dimension dynamic optimization, characterized in that, It includes a memory and a processor, wherein the processor is used to execute the method of any one of claims 1-12.

Images