CN-122001984-A - Multi-path concurrent dispatching voice outbound platform

CN122001984ACN 122001984 ACN122001984 ACN 122001984ACN-122001984-A

Abstract

The invention relates to the technical field of voice communication and discloses a multipath concurrent dispatching voice outbound platform which comprises an initial channel allocation module, an execution perception module, a strategy attribution learning module and a dynamic dispatching optimization module, wherein the initial channel allocation module is used for making an initial allocation scheme based on channel state data and allocating voice outbound tasks to be executed to corresponding voice outbound channels according to the initial allocation scheme, the execution perception module is used for evaluating congestion matching events between the voice outbound tasks and the corresponding voice outbound channels, the strategy attribution learning module is used for carrying out tracing analysis on the congestion matching event causes by using a deep attribution technology and generating dispatching instructions of the voice outbound tasks on the congestion channels based on tracing analysis results, and the dynamic dispatching optimization module is used for carrying out dynamic dispatching on the voice outbound tasks on the congestion channels according to the dispatching instructions. According to the invention, dynamic load balancing under a multipath concurrency scene is realized by means of similar channel migration, cross-level scheduling and the like, so that the concurrency processing capacity and the running stability of the voice outbound platform are enhanced.

Inventors

Huang Zhuokeng
ZHANG MINGXUE
WANG JINQUAN
HE WENJUN

Assignees

深圳市厚利联明信息技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260210

Claims (10)

1. A multiple concurrent dispatch voice outbound platform, the platform comprising: The initial channel allocation module is used for acquiring channel state data of each current voice outbound channel, formulating an initial allocation scheme based on the channel state data, and allocating the voice outbound task to be executed to the corresponding voice outbound channel according to the initial allocation scheme; The execution perception module is used for tracking state evolution signals of each voice outbound channel in a task execution period, identifying load abnormal signals from the state evolution signals, and evaluating congestion matching events between the voice outbound tasks and the corresponding voice outbound channels according to the load abnormal signals; the strategy attribution learning module is used for receiving the congestion matching event output by the link execution perception module, carrying out traceability analysis on the congestion matching event cause by utilizing the deep attribution technology, and generating a scheduling instruction of a voice outbound task on a congestion channel based on the traceability analysis result; The dynamic scheduling optimization module is used for executing dynamic scheduling on the voice outbound tasks on the crowded channels according to the scheduling instructions, wherein the dynamic scheduling comprises a similar channel load migration action and a cross-stage channel scheduling action which are used for maintaining the current allocation action and migrating the voice outbound tasks on the crowded channels to alternative voice outbound channels with similar channel characteristics, and the similar channel load migration action and the cross-stage channel scheduling action are realized under the condition that the task execution environments are consistent.
2. The multiple concurrent dispatch voice outbound platform of claim 1 wherein the execution awareness module, when tracking the channel state evolution signals of each voice outbound channel during the task execution period and identifying the load fluctuation signals therefrom, evaluates congestion matching events between the voice outbound task and the corresponding voice outbound channel based on the load fluctuation signals, comprises: Synchronously extracting state signals of each voice outbound channel in a task execution period, and sequentially carrying out noise reduction and smoothing treatment on the state signals to generate a time sequence feature sequence corresponding to each voice outbound channel; Sequentially comparing the time sequence characteristic value with a state datum line generated according to a history rule, judging that a corresponding voice outbound channel has a load abnormal signal when the time sequence characteristic value continuously deviates from the state datum line, otherwise, judging that the corresponding voice outbound channel does not have the load abnormal signal, executing a voice outbound task according to a current distribution scheme and generating a real-time execution result; Carrying out association binding on voice outbound tasks on voice outbound channels with load abnormal signals, and comprehensively evaluating whether congestion matching events between the voice outbound tasks and the corresponding voice outbound channels are formed according to preset rules; if yes, marking the voice outbound channel corresponding to the determined congestion matching event as a congestion channel, and if not, marking the corresponding voice outbound channel as a congestion channel.
3. The multiple concurrency dispatch voice outbound platform of claim 1, wherein the policy attribution learning module comprises: The scene reconstruction module is used for identifying key variables and causal directions thereof which cause the occurrence of the congestion matching event from the congestion matching event, and reconstructing a state attribution map before the occurrence of the congestion matching event by combining the key variables and the causal directions; The strategy deconstructing module is used for deconstructing and analyzing an initial allocation scheme executed when the congestion matching event occurs in the game scene defined by the state attribution map, identifying the contribution degree of each decision dimension in the initial allocation scheme and outputting a failure component list; The scheduling instruction generation module is used for taking the state attribution map and the failure component list as input, evaluating heterogeneous processing effects of different candidate scheduling actions by utilizing a causal forest model based on the heterogeneous processing effects, and further generating a scheduling instruction for a voice outbound task on a crowded channel.
4. A multi-path concurrently scheduled voice outbound platform as claimed in claim 3 wherein the policy deconstructing module deconstructs the initial allocation scheme executed when the congestion matching event occurs in the game scenario defined by the state attribution map, and the identifying the contribution of each decision dimension in the initial allocation scheme and outputting the list of invalidation components comprises: mapping key variables causing congestion matching events in the state attribution map with each decision dimension in the initial allocation scheme, and defining game scenes according to mapping results; Mapping each decision dimension into a game participant in a game scene, mapping a strategy alliance corresponding to any decision dimension subset into the congestion degree caused by the strategy alliance through a preset feature function, and taking the congestion degree as the payment value of the current strategy alliance; calculating the marginal change of the payment value when each decision dimension is added into different strategy alliances, calculating the saproliferation value of the corresponding decision dimension in a weighted average mode, and sequencing the saprolion value to generate a sequencing list of the decision dimension; And (3) comparing the saprolimus value of each decision dimension with a preset threshold value, and screening the decision dimension which causes congestion matching event occurrence from the ordered list based on the comparison result and integrating the decision dimension into a failure component list.
5. The multi-path concurrent dispatch voice outbound platform of claim 4 wherein the dispatch instruction generation module, when using the state attribution map and the list of failed components as inputs, evaluates heterogeneous processing effects of different candidate dispatch actions using a causal forest model based on heterogeneous processing effects, thereby generating dispatch instructions for voice outbound tasks on congested tunnels, comprises: Acquiring a preset historical state attribution map and a corresponding failure component list to construct a historical training sample which comprises characteristic variables representing game scenes, actually executed scheduling actions as processing variables and improved results of congestion matching events after the actions are executed as result variables; Synchronously constructing an auxiliary estimator comprising a tendency score and a base line result expectation, and combining the auxiliary estimator and a historical training sample to train a causal forest model; Inputting a state attribution map corresponding to the current crowded channel and a failure component list into a causal forest model after training is completed, and calculating and comparing the processing effect of crowded matching events and confidence intervals thereof after different candidate scheduling actions are executed by using the causal forest model; and screening the optimal scheduling actions based on the processing effect of the congestion matching event after the execution of the different candidate scheduling actions and the confidence interval thereof, generating scheduling instructions and outputting the scheduling instructions to the dynamic scheduling optimization module.
6. The multiple concurrent dispatch voice call platform of claim 5, wherein the filtering the optimal dispatch actions based on the processing effect of congestion matching events and confidence intervals thereof after execution of different candidate dispatch actions comprises: Generating global dominant probability of any one candidate scheduling action over the other candidate scheduling action based on the processing effect of the congestion matching event after the execution of the different candidate scheduling actions and the confidence interval thereof; Calculating the risk exposure probability of the processing effect of each candidate scheduling action in a zero benefit or negative benefit interval, wherein the risk exposure probability is superior to the joint probability lower bound of all other candidate scheduling actions, based on the global dominant probability; Inputting the global dominant probability and the risk exposure into a preset Bayesian decision function, solving the maximum expected utility of each candidate scheduling action through the preset Bayesian decision function, and screening the optimal scheduling action based on the maximum expected utility.
7. The multi-path concurrent dispatch voice outbound platform of claim 1 wherein the similar path load migration actions are used to migrate voice outbound tasks on congested paths to alternative voice outbound paths with similar path characteristics to achieve load balancing while maintaining consistent task execution environments; The cross-class channel scheduling action is used for scheduling the voice outbound task with high priority from the current crowded channel to the voice outbound channel with high service level.
8. The multiple concurrent dispatch voice outbound platform of claim 7 wherein the migration of voice outbound tasks on congested channels to alternative voice outbound channels with similar channel characteristics comprises: Generating a plurality of virtual nodes for the feature labels of each voice outbound channel, mapping the virtual nodes onto a hash ring through a hash function, and associating the real-time load state and the feature labels of the corresponding voice outbound channels with each virtual node; extracting task labels of voice outbound tasks to be migrated on the crowded channels, calculating hash values, positioning initial virtual nodes on the hash ring according to the hash values, and traversing the hash ring clockwise from the initial virtual nodes to obtain a candidate channel set similar to the source channel; And selecting the voice outbound channel with the lowest real-time load from the candidate channel set as a target channel, migrating the voice outbound task on the crowded channel to the target channel, and synchronously updating the load state of the virtual node corresponding to the target channel in the hash ring to complete migration closed loop.
9. The multi-path concurrently scheduled voice outbound platform of claim 8, wherein locating the starting virtual node on the hash ring based on the hash value, traversing the hash ring clockwise from the starting virtual node, obtaining a set of candidate channels similar to the source channel comprises: In the process of traversing the hash ring clockwise, calculating the feature matching degree between the feature labels and the task labels of each path node association channel and the upper confidence limit of the load state; Screening voice outbound channels corresponding to the characteristic matching degree, wherein the upper confidence limit of the characteristic matching degree exceeds a preset threshold value, and the upper confidence limit of the load state is lower than the preset threshold value, and integrating the voice outbound channels to generate a candidate channel set.
10. A method of multi-path concurrent dispatch voice outbound using the multi-path concurrent dispatch voice outbound platform of any one of claims 1-9, the method comprising: collecting channel state data of each current voice outbound channel, formulating an initial allocation scheme based on the channel state data, and allocating the voice outbound tasks to be executed to the corresponding voice outbound channels according to the initial allocation scheme; tracking state evolution signals of each voice outbound channel in a task execution period, identifying load abnormal signals from the state evolution signals, and evaluating congestion matching events between the voice outbound task and the corresponding voice outbound channel according to the load abnormal signals; The congestion matching event output by the link execution perception module is received, the congestion matching event cause is subjected to traceability analysis by using a deep attribution technology, and a dispatching instruction of a voice outbound task on a congestion channel is generated based on the traceability analysis result; and executing dynamic scheduling on the voice outbound task on the crowded channel according to the scheduling instruction.

Description

Multi-path concurrent dispatching voice outbound platform Technical Field The invention relates to the technical field of voice communication, in particular to a voice outbound platform with multiple paths of concurrent scheduling. Background The voice outbound platform is an automatic or semi-automatic voice call system built based on a communication technology, a computer system and a voice interaction technology, and can realize the functions of batch outbound, intelligent voice interaction, seat butt joint, call record management, data analysis and the like, so that the voice outbound platform is widely applied to scenes of enterprise product marketing, customer service return visit, government affair notification, financial prompting and the like, which need to initiate voice calls in batches. If the platform cannot realize multi-path concurrent scheduling in outbound, the system cannot process a plurality of call tasks at the same time, so that a large number of outbound tasks are trapped in queuing, the overall outbound efficiency is greatly reduced, if the platform cannot perform dynamic task scheduling based on the actual load condition of the channels, for example, the problem of unbalanced resource allocation is aggravated if the platform performs dynamic task scheduling on the crowded channels, on one hand, the call congestion of the crowded channels cannot be relieved, and on the other hand, the idle channels cannot be reasonably utilized, thereby not only continuously aggravating the queuing backlog of outbound tasks, but also losing the capability of dynamically adjusting task allocation according to the load of the real-time channels by the platform, and further weakening the flexibility and scene suitability of the system scheduling. For the problems in the related art, no effective solution has been proposed at present. Disclosure of Invention Aiming at the problems in the related art, the invention provides a voice outbound platform with multiple paths of concurrent scheduling so as to overcome the technical problems in the prior related art. For this purpose, the invention adopts the following specific technical scheme: according to a first aspect of the present invention there is provided a multi-path concurrently scheduled voice outbound platform comprising: The initial channel allocation module is used for acquiring channel state data of each current voice outbound channel, formulating an initial allocation scheme based on the channel state data, and allocating the voice outbound task to be executed to the corresponding voice outbound channel according to the initial allocation scheme; The execution perception module is used for tracking state evolution signals of each voice outbound channel in a task execution period, identifying load abnormal signals from the state evolution signals, and evaluating congestion matching events between the voice outbound tasks and the corresponding voice outbound channels according to the load abnormal signals; the strategy attribution learning module is used for receiving the congestion matching event output by the link execution perception module, carrying out traceability analysis on the congestion matching event cause by utilizing the deep attribution technology, and generating a scheduling instruction of a voice outbound task on a congestion channel based on the traceability analysis result; The dynamic scheduling optimization module is used for executing dynamic scheduling on the voice outbound tasks on the crowded channels according to the scheduling instructions, and the dynamic scheduling comprises a similar channel load migration action and a cross-stage channel scheduling action which maintain the current allocation action and are used for migrating the voice outbound tasks on the crowded channels to alternative voice outbound channels with similar channel characteristics, and load balancing is realized under the condition that the task execution environments are consistent. Preferably, the execution sensing module tracks the channel state evolution signal of each voice outbound channel in the task execution period, identifies a load fluctuation signal therefrom, and evaluates a congestion matching event between the voice outbound task and the corresponding voice outbound channel according to the load fluctuation signal, wherein the method comprises the following steps: Synchronously extracting state signals of each voice outbound channel in a task execution period, and sequentially carrying out noise reduction and smoothing treatment on the state signals to generate a time sequence feature sequence corresponding to each voice outbound channel; Sequentially comparing the time sequence characteristic value with a state datum line generated according to a history rule, judging that a corresponding voice outbound channel has a load abnormal signal when the time sequence characteristic value continuously deviates from the state datum line, other