Method and device for transmitting low-redundancy compressed signal creation queuing number data

Abstract

This invention discloses a low-redundancy compressed data transmission method and apparatus for queuing and calling systems in the field of data transmission technology. The method includes: collecting single-service time series and queue waiting time series from multiple transmission endpoints connected to the cloud; extracting time features and inputting an adaptive timing window module to determine the timing window length; constructing endpoint state vectors and calculating state vector difference data; outputting prediction residual data through a lightweight prediction model, compressing the data, and outputting compressed encoded data; and performing redundancy difference analysis on the cloud side based on a preset redundancy level and updating the timing window length based on multiple redundancy differences. This invention solves the technical problem in existing technologies where a unified compression strategy cannot adapt to the differences in real-time waiting time perception between different transmission endpoints, leading to an inability to dynamically balance data redundancy and freshness. It achieves the technical effect of improving data transmission and compression efficiency while minimizing data redundancy.

Description

Technical Field

[0001] This invention relates to the field of data transmission technology, specifically to a low-redundancy compressed queuing and calling data transmission method and apparatus. Background Technology

[0002] In the context of information technology innovation, queuing and calling systems are widely used in government, financial, and medical service halls. With the increasing number of terminal devices and the growing demand for real-time data, how to reduce redundancy while ensuring data transmission efficiency is a critical issue that current queuing and calling systems urgently need to address. In traditional queuing and calling systems, when multiple transmission ends (clients) interact with the cloud, a unified compression strategy is typically used to encode and transmit state vectors. However, in actual operation, due to factors such as hardware performance, network conditions, and workload, the single service time and queue waiting time of each transmission end vary significantly, resulting in different data freshness requirements and tolerable redundancy for each transmission end. Specifically, for transmission ends with longer waiting times, the state vector changes more gradually, resulting in relatively high redundancy. Larger timing windows are suitable for compression. Conversely, for transmission ends with shorter waiting times, state changes are more frequent, requiring higher data freshness and necessitating smaller timing windows to reduce redundancy. Existing data transmission methods fail to adequately consider the heterogeneity of time characteristics, making it unsuitable for real-time waiting time perception across different transmission ends. This leads to excessive redundancy or loss of critical information on some endpoints. Furthermore, the timing window length cannot be dynamically adjusted based on the real-time service and waiting time characteristics of the transmission ends, affecting the accuracy of state vector difference calculation. Additionally, the cloud cannot perform closed-loop optimization of compression parameters for each transmission end based on the actual redundancy differences in received data, impacting the overall data transmission efficiency of the system.

[0003] Therefore, in the current related technologies, there is a technical problem that the unified compression strategy cannot adapt to the differences in real-time waiting time perception at different transmission ends, resulting in an inability to dynamically balance data redundancy and freshness. Summary of the Invention

[0004] This application provides a low-redundancy compressed data transmission method and apparatus for queuing and calling in the domestic IT industry. It solves the technical problem in the prior art that the uniform compression strategy cannot adapt to the differences in real-time waiting time perception of different transmission ends, resulting in an inability to dynamically balance data redundancy and freshness. It achieves the technical effect of improving data transmission and compression efficiency and minimizing data redundancy.

[0005] This application provides a low-redundancy compressed data transmission method for queuing and calling systems in China. The method includes: collecting data from multiple transmission endpoints connected to the cloud side; collecting a single service time series and a queue waiting time series from each of the multiple transmission endpoints; extracting the time features of the single service time series and the queue waiting time series; inputting the time features into an endpoint adaptive timing window module to determine the timing window length of each transmission endpoint; constructing an endpoint state vector based on the timing window length; calculating the state vector difference data between adjacent time slices; outputting prediction residual data through a lightweight prediction model; compressing the state vector difference data and the prediction residual data to output compressed encoded data; and the cloud side performing redundancy difference analysis on the multiple compressed encoded data corresponding to the multiple transmission endpoints according to a preset redundancy, and updating the timing window length based on the multiple redundancy differences.

[0006] In a possible implementation, the time feature input to the end-side adaptive timing window module determines the timing window length for each transmission end. The method includes: constructing single-service time series samples and queue waiting time series samples; slicing the single-service time series samples and queue waiting time series samples to construct a basic training dataset; extracting time feature training samples and corresponding timing window length label samples from the basic training dataset; constructing a lightweight neural network; training the time feature training samples and timing window length label samples using the lightweight neural network to obtain a trained end-side adaptive timing window module; and inputting the time feature into the end-side adaptive timing window module to calculate the timing window length for each transmission end, wherein the time feature includes average service time, average waiting time, fluctuation, and queue change frequency.

[0007] In a possible implementation, the method for constructing time-series window length label samples includes: obtaining a candidate time-series window set; performing simulated compression on the basic training dataset based on the candidate time-series window set to obtain a compressed simulated dataset; calculating the delay penalty index of the compressed simulated dataset using a delay penalty function; performing a minimization search on the candidate time-series window set according to the delay penalty index to output a preferred time-series window set for each basic training dataset; and performing labeling processing on the preferred time-series window set to construct time-series window length label samples.

[0008] In a possible implementation, an end-side state vector is constructed for each transmission end based on the time window length, and state vector differential data of adjacent time slices is calculated. The method includes: dynamically slicing the time axis based on the time window length to output continuous time slice intervals; extracting traffic volume indicators, efficiency indicators, queuing status indicators, and trend indicators of the continuous time slice intervals, and outputting end-side state vectors; comparing the end-side state vectors of adjacent time slices in the continuous time slice intervals field by field to extract state vector differential data based on differential masks and differential values.

[0009] In a possible implementation, the prediction residual data is output through a lightweight prediction model. The method includes: inputting the end-side state vector of the previous time slice in the continuous time slice interval into the lightweight prediction model to obtain the prediction end-side state vector of the continuous time slice, wherein the lightweight prediction model is a Markov probability prediction model based on state transition rules; comparing the prediction end-side state vector with the actual end-side state vector field by field to obtain prediction error data; and extracting the residual mask and residual value from the prediction error data to output the prediction residual data.

[0010] In a possible implementation, compressed data is output based on the state vector difference data and the prediction residual data. The method includes: fusing the state vector difference data and the prediction residual data to construct a fused data structure; performing bit-level compressed encoding on the difference mask and residual mask of the fused data structure; performing variable-length compressed encoding on the difference value and residual value of the fused data structure; and outputting compressed data.

[0011] In a possible implementation, the cloud side performs redundancy difference analysis on multiple compressed encoded data corresponding to the multiple transmission ends based on a preset redundancy. The method includes: obtaining a preset target redundancy on the cloud side; performing redundancy analysis on the multiple compressed encoded data corresponding to the multiple transmission ends to obtain multiple redundancies; performing redundancy difference analysis based on the preset target redundancy and the multiple redundancy corresponding to the multiple transmission ends to obtain individual redundancy difference and global redundancy difference; obtaining multiple redundancy differences corresponding to the multiple transmission ends based on the individual redundancy difference and the global redundancy difference; and updating the timing window length based on each redundancy difference.

[0012] This application also provides a low-redundancy compressed queuing and calling data transmission device, the device comprising: a time series acquisition module, used to acquire multiple transmission end-sides connected to the cloud side, and acquire the single service time series and queue waiting time series of each of the multiple transmission end-sides; a time feature extraction module, used to extract the time features of the single service time series and the queue waiting time series, and input the time features into the end-side adaptive time series window module to determine the time series window length of each transmission end-side; a differential data calculation module, used to construct the end-side state vector of each transmission end-side based on the time series window length, and calculate the state vector differential data of adjacent time slices; a data compression processing module, used to output prediction residual data through a lightweight prediction model, and perform compression processing on the state vector differential data and the prediction residual data to output compressed encoded data; and a redundancy difference analysis module, used by the cloud side to perform redundancy difference analysis on the multiple compressed encoded data corresponding to the multiple transmission end-sides according to a preset redundancy, and update the time series window length according to the multiple redundancy differences.

[0013] This application proposes a low-redundancy compressed queuing and calling data transmission method and apparatus. The method collects single-service time series and queue waiting time series from each of multiple transmission endpoints connected to the cloud. It extracts time features and inputs them into an adaptive timing window module to determine the timing window length. It constructs an endpoint state vector and calculates state vector difference data. A lightweight prediction model outputs prediction residual data, which is then compressed to output compressed encoded data. The cloud side performs redundancy difference analysis based on a preset redundancy level and updates the timing window length based on multiple redundancy differences. This solves the technical problem in existing technologies where a unified compression strategy cannot adapt to the differences in real-time waiting time perception among different transmission endpoints, leading to an inability to dynamically balance data redundancy and freshness. This achieves the technical effect of improving data transmission and compression efficiency while minimizing data redundancy. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the apparatus according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 A schematic diagram of the low-redundancy compressed data transmission method for queuing and calling numbers provided in this application embodiment.

[0016] Figure 2A schematic diagram of the structure of the low-redundancy compressed queuing and calling data transmission device provided in the embodiments of this application.

[0017] Figure labeling: Time series acquisition module 10, time feature extraction module 20, differential data calculation module 30, data compression processing module 40, redundancy difference analysis module 50. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0019] This application provides a low-redundancy compressed data transmission method for queuing and calling systems in the context of domestic IT innovation, such as... Figure 1 As shown, the method includes: Step S100: Collect data from multiple transmission endpoints connected to the cloud side, and collect the single service time sequence and queue waiting time sequence for each of the multiple transmission endpoints.

[0020] Preferably, the system acquires multiple transmission terminals connected to the cloud side in the queuing and calling system, which may include window calling terminals, queuing and ticket taking terminals, and business processing terminals. Then, it periodically collects the single service time series and queue waiting time series of each transmission terminal from system log records and ticket taking system data. The single service time series refers to the time spent by each transmission terminal in processing a business from start to finish, and the data sequence is formed by arranging them in chronological order. It reflects the business processing efficiency of that window. For example, the single service time series of transmission terminal A is [180 seconds, 210 seconds, 165 seconds, 195 seconds, 225 seconds], corresponding to the business of account opening, transfer, loss reporting, payment, and loan consultation. The service time of different types of business is different. Simple business is fast and complex business is slow. It is used to judge the data change frequency and redundancy of that terminal. Queue waiting time series refers to the waiting time for each user from obtaining a number to being served by the transmission terminal. It is a data sequence arranged in chronological order, reflecting the congestion level of the window queue. For example, the queue waiting time series of transmission terminal B is [450 seconds, 380 seconds, 510 seconds, 290 seconds, 420 seconds], which corresponds to customer 1 waiting for 7.5 minutes, customer 2 waiting for 6.3 minutes, customer 3 waiting for 8.5 minutes, customer 4 waiting for 4.8 minutes, and customer 5 waiting for 7 minutes, respectively. The longer the waiting time, the more backlog of business there is, affecting the data freshness requirement.

[0021] Step S200: Extract the time features of the single service time series and the queue waiting time series, and input the time features into the adaptive timing window module on the input end to determine the timing window length of each transmission end.

[0022] Step S200 further includes: constructing single-service time-series samples and queue waiting time-series samples; slicing the single-service time-series samples and queue waiting time-series samples to construct a basic training dataset; extracting time feature training samples and corresponding time window length label samples from the basic training dataset; constructing a lightweight neural network; training the time feature training samples and the time window length label samples based on the lightweight neural network to obtain a trained end-side adaptive time window module; inputting the time features into the end-side adaptive time window module to calculate the time window length for each transmission end, wherein the time features include average service time, average waiting time, fluctuation amount, and queue change frequency.

[0023] Preferably, the time features of single service time series and queue waiting time series are extracted, including average service time, mean waiting time, volatility, and queue change frequency. The average service time is the average time for a transmission end to process each service within a specified time window. Average service time = sum of service times for all services at that end / total number of services. This reflects the service processing efficiency of that port; a longer average service time indicates more complex services and relatively slower status changes. The mean waiting time is the average waiting time for a user from obtaining a number to being called by the transmission end within a specified time window. Mean waiting time = sum of waiting times for all users / total number of users. This reflects the queue congestion and service pressure of that port; a longer waiting time indicates more backlogged services and higher data freshness requirements. Volatility reflects the dispersion and instability of service time or waiting time in time series, usually expressed as standard deviation or variance. Volatility = √[Σ(service time per instance - average service time)]. 2 [Sampling quantity]; Queue change frequency is the number of times the queue length (number of people in the queue) changes per unit time. Queue change frequency = number of queue state changes / observation time, reflecting the dynamic activity level of the queue. Queue state changes include new customers taking numbers and joining the queue (queue +1), customers being called for service, or customers leaving midway (queue -1).

[0024] Preferably, single-service time-series samples and queue waiting time-series samples are constructed from the original time-series data. These samples are then sliced ​​at fixed time lengths to obtain multiple sample segments, each serving as an independent training sample to determine the basic training dataset. For each slice of the basic training dataset, its time feature training sample is calculated as input, and the corresponding time-series window length label sample is determined as the training label. The time feature training sample includes the average service time, mean waiting time, fluctuation, and queue change frequency for each slice. The time-series window length label is calculated using a delay penalty function, and the optimal window is searched in the candidate window set to determine the complete training data pair.

[0025] Preferably, a lightweight neural network model suitable for running on the window terminal device is designed. This model is trained on time feature training samples and time-series window length label samples to learn how to predict the optimal window length from the time features, thus obtaining a trained end-side adaptive time-series window module. The neural network model has 5 input layers, which take into account the average service time, average waiting time, service fluctuation, waiting fluctuation, and queue change frequency, respectively; 16 hidden layers with ReLU activation functions; 28 hidden layers with ReLU activation functions; and 1 output layer, which outputs the time-series window length. After the trained model is deployed to each transmission window terminal, the real-time collected time features are input into the end-side adaptive time-series window module to calculate the time-series window length for each transmission terminal under the current service state.

[0026] Preferably, in a simulated government service hall scenario (including 30 transmission endpoints, covering business types such as simple consultation, medium-complex account opening, and complex loan approval), comparative tests were conducted using a fixed window (uniformly set to 60 seconds) and an adaptive window (dynamically calculated based on the endpoint's adaptive timing window module, with a window length ranging from 15 seconds to 300 seconds). The test results are shown in Table 1. Table 1. Comparison of bandwidth savings rates under different business scenarios (based on a fixed window)

[0027] For complex business terminals with smooth state changes, fixed window strategies transmit a large amount of redundant and repetitive state data. Adaptive windows, by increasing the window length, achieve bandwidth savings of over 56%. For fast business terminals with frequent state changes, dynamically reducing the window length achieves 26% bandwidth savings while ensuring data freshness, and reduces state update latency by 41%. Furthermore, the window length of each terminal adaptively adjusts according to its own service time fluctuations and queue change frequency. For example, during peak traffic periods in the morning, the window length of fast business terminals automatically shortens to capture rapid changes; during the midday off-peak period, the window length automatically increases to improve compression efficiency. This enhances the ability to accurately adapt to the dynamic states of heterogeneous terminals in a domestic IT innovation environment.

[0028] Furthermore, step S200 also includes obtaining a candidate time-series window set, performing simulated compression on the basic training dataset based on the candidate time-series window set to obtain a compressed simulated dataset; calculating the delay penalty index of the compressed simulated dataset using a delay penalty function; performing a minimization search on the candidate time-series window set according to the delay penalty index, outputting a preferred time-series window set for each basic training dataset, and performing labeling processing on the preferred time-series window set to construct time-series window length label samples.

[0029] Preferably, a set of possible time-series window length values ​​are predefined based on the business characteristics of the queuing system, serving as a candidate space to be searched. A candidate time-series window set is determined, where the smallest window can capture rapid changes and the largest window can cover long periods. Based on the candidate time-series window set, the basic training dataset is simulated for compression. For each training sample, a complete compression process is simulated using each window length in the candidate time-series window set to obtain the corresponding compressed result dataset. Then, a delay penalty function is designed to comprehensively score the compression result of each candidate window, taking into account multiple factors such as compression rate, delay, and reconstruction error. The multi-objective optimization problem is transformed into a single scalar evaluation problem, and the delay penalty index of the compressed simulation dataset is calculated. The smaller the value, the better the window. Then, a minimization search is performed in the candidate time-series window set according to the delay penalty index. That is, for each training sample, the window with the smallest delay penalty index is selected from the candidate windows as the optimal window. Finally, the preferred time-series window set is labeled to construct time-series window length label samples.

[0030] Step S300: Construct the end-side state vector of each transmission end based on the time window length, and calculate the state vector difference data of adjacent time slices.

[0031] Step S300 further includes: dynamically slicing the time axis based on the time window length to output continuous time slice intervals; extracting the business volume indicators, efficiency indicators, queuing status indicators, and trend indicators of the continuous time slice intervals to output end-side state vectors; and comparing the end-side state vectors of adjacent time slices in the continuous time slice intervals field by field to extract state vector differential data based on differential masks and differential values.

[0032] Preferably, the time axis is dynamically sliced ​​according to a determined time window length (e.g., 120 seconds), dividing the continuous time axis into several adjacent and non-overlapping time slice intervals. Each time slice interval serves as an analysis unit. Different transmission ends use different slicing granularities based on service characteristics. Then, for each time slice interval, its traffic volume indicators, efficiency indicators, queuing status indicators, and trend indicators are extracted and combined into a fixed-dimensional end-side state vector to characterize the operating status of the transmission end within that time period. Among them, the traffic volume indicators reflect the total amount of service processing within the time slice, such as the total number of completed services, the total number of customers served, the arrival rate, and the completion rate; the efficiency indicators reflect the service efficiency level within the time slice, such as the average service time, the standard deviation of service time, the window utilization rate, and the throughput; the queuing status indicators reflect the queue backlog within the time slice, such as the queue length at the beginning / end of the slice, the average queue length, the maximum queue length, and the average waiting time; and the trend indicators include queue change trend (negative values ​​indicate a decrease), service time trend (positive values ​​indicate a slowdown), arrival rate trend, and efficiency trend. The end-side state vectors of adjacent time slices in a continuous time slice interval are compared field by field. Differential masks are used to mark the fields that have changed, and differential values ​​are used to record the specific amount of change. The differential mask is a binary sequence of the same length as the state vector, where 1 indicates that the field has changed and 0 indicates that it has not changed. Finally, the differential mask and differential values ​​are integrated to determine the differential data of the state vector, thereby significantly reducing the amount of data transmitted.

[0033] Step S400: Output prediction residual data through a lightweight prediction model, and compress and output compressed encoded data based on the state vector difference data and the prediction residual data.

[0034] Step S400 further includes inputting the end-side state vector of the previous time slice in the continuous time slice interval into the lightweight prediction model to obtain the predicted end-side state vector of the continuous time slice, wherein the lightweight prediction model is a Markov probability prediction model based on state transition rules; comparing the predicted end-side state vector with the actual end-side state vector field by field to obtain prediction error data; and extracting the residual mask and residual value from the prediction error data to output the prediction residual data.

[0035] Preferably, the state changes of a queuing system typically exhibit regularity. For example, the service efficiency of a window often shows a gradual change between adjacent time slices. A lightweight Markov probability prediction model utilizes the characteristic that "the current state is only related to the previous state" for prediction. The end-side state vector of the previous time slice in a continuous time slice interval is input into the Markov probability prediction model based on state transition rules. This model, based on historically learned state transition rules, calculates the most likely state vector for the current time slice, obtaining the predicted end-side state vector for the current time slice. Assuming that the average queue length for a certain window is 5 people from 09:00 to 09:02 AM (slice 1), according to historical statistical patterns, when the queue length... When there are 5 people, the queue length in the next 2-minute slice (slice 2) has a 70% probability of becoming 4 people, a 20% probability of becoming 3 people, and a 10% probability of remaining unchanged. The Markov prediction model encodes these probability rules into a state transition matrix to predict the state of slice 2. The Markov prediction model has only one parameter, a state transition probability matrix, which occupies very little memory (usually less than a few hundred KB), has a fast inference speed, and is suitable for deployment on edge devices with limited computing resources. The output predicted edge state vector has the same dimensional structure as the actual state vector, including business volume indicators, efficiency indicators, queue status indicators, and trend indicators. However, the value of each field is the expected value or the most likely value calculated based on probability rules.

[0036] Preferably, the state vector predicted by the Markov model is compared field by field with the actual state vector calculated from the data collected in the current time slice. The prediction deviation for each field is calculated to form prediction error data, which may include numerical error or state error. Numerical error, i.e., the specific difference between the predicted value and the actual value, can be positive or negative. State error refers to whether the prediction of certain fields is accurate, and is used to determine whether correction information needs to be transmitted. For example, for "queue length," if the Markov model predicts that the queue length in the current slice is 4 people, and the actual measured queue length is 5 people, then the prediction error is +1 people, meaning the actual length is 1 person more than the prediction. For "average service time," if the predicted value is 45 seconds, and the actual value is 48 seconds, then the prediction error is +3 seconds. A tolerance threshold range is set for each field. If the prediction error is within the tolerance threshold range (e.g., the error does not exceed 5%), the prediction is considered accurate, and no additional correction is needed for that field. If the error exceeds the tolerance threshold range, the error information for that field needs to be recorded for data recovery.

[0037] Preferably, residual masks and residual values ​​are extracted from the prediction error data. The residual mask is a binary marker sequence of the same length as the state vector. Each bit in the mask represents a field in the state vector. If the prediction error of a field exceeds a preset tolerance threshold, the corresponding bit is marked as 1, indicating that "this field needs to be corrected by transmitting residual values." If the error is within the tolerance range, it is marked as 0, indicating that "the prediction is accurate enough and no correction is needed." Residual values ​​are recorded only for fields marked as 1 in the residual mask, representing the actual prediction error value—the difference between the actual value and the predicted value. For fields marked as 0 in the mask, no residual values ​​are recorded, thus saving transmission bandwidth. Finally, the prediction residual data is output. Assuming the state vector has 5 fields: queue length, average service time, window utilization, customer arrival rate, and average waiting time, the Markov model predicts the current slice state as [4 people, 45 seconds, 75%, 0.5 people / minute, 180 seconds], while the actual collected state is [5 people, 46 seconds, 76%, 0.5 people / minute, 185 seconds]. The error is calculated for each field as [+1 person, +1 second, +1%, 0, +5 seconds]. Tolerance thresholds are set: queue length ±1 person is acceptable, service time ±2 seconds is acceptable, utilization rate ±5% is acceptable, arrival rate ±0.1% is acceptable, and waiting time ±10 seconds is acceptable. Then, based on these thresholds, a queue length error of +1 person is acceptable, a service time error of +1 second is acceptable, a utilization rate error of +1% is acceptable, an arrival rate error of 0 is acceptable, and a waiting time error of +5 seconds is acceptable. If all fields are within the tolerance range, the residual mask is all 0, and no residual values ​​need to be transmitted.

[0038] Preferably, in the context of domestic IT innovation, terminal devices (such as window queuing terminals and queuing terminals based on domestic CPUs and operating systems) have relatively limited computing resources. Compared with using heavy deep learning models (such as LSTM or Transformer) for state prediction, Markov probability prediction models based on state transition rules have significant lightweight advantages. The comparison of measured data of typical domestic IT innovation terminals is shown in Table 2: Table 2. Data Comparison Examples under Loongson 3A5000 + Tongxin UOS Environment

[0039] As shown in the table above, the lightweight prediction model only needs to store one state transition probability matrix, and the number of parameters is usually in the thousands of bytes, reducing the number of parameters by more than 98%. The time taken for a single prediction is reduced from the 20ms level to less than 1ms, which can meet the requirements of queuing and calling systems for millisecond-level real-time response and avoid calling delays caused by data processing backlog. It significantly reduces the power consumption pressure on terminal battery-powered devices (such as mobile ticketing terminals) or low-power embedded devices. Furthermore, the model file is less than 30KB and can be directly embedded in the terminal acquisition program without the need for additional model loading and parsing frameworks, reducing software dependency and memory fragmentation risks.

[0040] Furthermore, step S400 also includes fusing the state vector difference data and the prediction residual data to construct a fused data structure; performing bit-level compression encoding on the difference mask and residual mask of the fused data structure; performing variable-length compression encoding on the difference values ​​and residual values ​​of the fused data structure; and outputting compressed encoded data.

[0041] Preferably, the state vector differential data and prediction residual data are redundancy removal results of two different dimensions describing the same set of state vector fields. If they are processed and transmitted separately, additional packet header overhead and duplicate field identification information will be generated. Fusion processing can organize the masks and values ​​of both together and share field index information, thereby further reducing transmission overhead. The fused data structure is constructed with "field" as the basic unit. For each field in the state vector, the fused data structure records whether the field has actually changed, whether the prediction of the field is accurate, and the corresponding change value and error value. Bit-level compression coding is used to compress the differential mask and residual mask of the fused data structure, using as few bits as possible to represent the mask information. The differential mask and residual mask are essentially binary sequences with a length equal to the dimension of the state vector, with each position having only 0 or 1. For example, for a 17-dimensional state vector, its differential mask is 17 bits, and each bit indicates whether the corresponding field has changed or whether the prediction is accurate. Bit-level compression coding is a special compression technique designed for binary sequences, using the distribution pattern of 0 and 1 in the mask sequence to reduce storage space.

[0042] Preferably, the difference and residual values ​​are typically integers or decimals with a relatively limited range and follow a certain probability distribution. For example, queue length variations are usually concentrated between -3 and +3, and service time variations are usually between -10 and +10 seconds. Some values ​​(such as 0 or ±1) occur very frequently, while some extreme values ​​(such as ±20) occur very infrequently. Variable-length compression coding is used to compress the difference and residual values ​​in the fused data structure. Different codeword lengths are used to represent values ​​with different frequencies; shorter codewords represent frequently occurring values, and longer codewords represent less frequently occurring values, thus minimizing the average code length and ultimately outputting compressed coded data. By encoding separately, compression efficiency can be optimized according to their respective distribution characteristics.

[0043] In step S500, the cloud side performs redundancy difference analysis on the multiple compressed encoded data corresponding to the multiple transmission end sides according to the preset redundancy, and updates the time window length according to the multiple redundancy differences.

[0044] Step S500 further includes: obtaining a preset target redundancy on the cloud side; performing redundancy analysis on multiple compressed encoded data corresponding to the multiple transmission ends to obtain multiple redundancy values; performing redundancy difference analysis based on the preset target redundancy and the multiple redundancy values ​​corresponding to the multiple transmission ends to obtain individual redundancy difference and global redundancy difference; obtaining multiple redundancy differences corresponding to the multiple transmission ends based on the individual redundancy difference and the global redundancy difference; and updating the timing window length based on each redundancy difference.

[0045] Preferably, a closed-loop feedback mechanism is established on the cloud side. By analyzing the difference between the actual redundancy and the target redundancy of the compressed data at each transmission end, the time window length of each end is dynamically adjusted to optimize and update the overall compression efficiency of the system. Specifically, the cloud side sets a target redundancy benchmark value based on the overall system requirements, representing the balance point between data freshness and compression rate. The target redundancy on the cloud side is used as the optimization benchmark. Redundancy calculation is performed on the compressed encoded data uploaded by each transmission end. By comparing the ratio of the compressed data size to the theoretical minimum data size, the current actual redundancy of each transmission end is calculated. This is used to quantitatively measure the proportion of redundant information in the compressed encoded data. For example, after decompressing and restoring the compressed encoded data, it is compared with the original data to calculate the ratio of data volume before and after compression. The higher the redundancy, the worse the compression effect, and the more repetitive or predictable information exists in the data; the lower the redundancy, the better the compression effect, and the more compact the data.

[0046] Preferably, the actual redundancy of each transmission end is compared with the preset target redundancy to calculate the individual redundancy difference and the global redundancy difference. These reflect the deviation of each end from the target and the overall deviation of the system, respectively. The individual redundancy difference is calculated separately for each transmission end, i.e., the actual redundancy minus the target redundancy. A positive number indicates that the redundancy of that end is higher than the target, indicating insufficient compression; a negative number indicates that the redundancy of that end is lower than the target, indicating over-compression and potential loss of necessary information freshness. The global redundancy difference is an overall deviation index obtained by comprehensively statistically analyzing the redundancy of all transmission ends and comparing it with the target redundancy. It is either the difference between the average redundancy of all ends and the target value, or the difference between the weighted average redundancy of all ends and the target value, where weighting is based on the data volume or importance of each end. The individual redundancy difference is used to finely adjust the timing window length of each end, addressing the differences between ends; the global redundancy difference is used for macroscopic calibration, addressing the overall bias of the system. Combining the two can simultaneously achieve local optimization and global balance.

[0047] Preferably, the cloud uses an adjustment function to calculate the combined individual redundancy difference and the global redundancy difference. For example, the combined redundancy difference = individual redundancy difference + α × global redundancy difference, where α is a weighting coefficient between 0 and 1, used to control the degree of influence of global feedback on individual terminals. When the system needs to emphasize overall consistency, α takes a larger value; when the system needs to respect the individual differences of each terminal, α takes a smaller value. This yields the final combined redundancy difference for each transmission terminal. Then, based on the combined redundancy difference, the timing window length for each transmission terminal is adjusted and updated. Specifically, when the combined redundancy difference is positive, the system increases the timing window length. This allows the difference and prediction algorithms to capture patterns over longer time spans, thereby improving compression efficiency. When the overall redundancy difference is negative, the system reduces the time-series window length, allowing for more frequent data updates and improved information freshness. Simultaneously, a gradual adjustment strategy is employed to prevent excessive window length oscillations or rapid adjustments. The magnitude of each adjustment is proportional to the size of the overall redundancy difference; the larger the deviation, the larger the adjustment step, and vice versa. Upper and lower limits for the window length are set to prevent adjustments from exceeding reasonable ranges. Finally, the updated window length is sent to the corresponding transmission end, ensuring improved data transmission and compression efficiency while minimizing data redundancy.

[0048] In the above text, refer to Figure 1 This paper describes in detail a low-redundancy compressed queuing and calling data transmission method based on embodiments of the present invention. Next, we will refer to... Figure 2 This invention describes a low-redundancy compressed queuing and calling data transmission device based on an embodiment of the present invention.

[0049] The low-redundancy compression queuing and calling data transmission device according to embodiments of the present invention is used to solve the technical problem in the prior art where a unified compression strategy cannot adapt to the differences in real-time waiting time perception at different transmission ends, resulting in an inability to dynamically balance data redundancy and freshness. It achieves the technical effect of improving data transmission and compression efficiency while minimizing data redundancy. Figure 2 As shown, the low-redundancy compressed queuing and calling data transmission device includes: a time series acquisition module 10, a time feature extraction module 20, a differential data calculation module 30, a data compression processing module 40, and a redundancy difference analysis module 50.

[0050] The time series acquisition module 10 is used to acquire multiple transmission endpoints connected to the cloud side, and to acquire the single service time series and queue waiting time series of each of the multiple transmission endpoints; the time feature extraction module 20 is used to extract the time features of the single service time series and the queue waiting time series, and input the time features into the endpoint adaptive time series window module to determine the time series window length of each transmission endpoint; the differential data calculation module 30 is used to construct the endpoint state vector of each transmission endpoint based on the time series window length, and to calculate the state vector differential data of adjacent time slices; the data compression processing module 40 is used to output prediction residual data through a lightweight prediction model, and to compress and output compressed encoded data based on the state vector differential data and the prediction residual data; the redundancy difference analysis module 50 is used by the cloud side to perform redundancy difference analysis on the multiple compressed encoded data corresponding to the multiple transmission endpoints according to a preset redundancy, and to update the time series window length based on the multiple redundancy differences.

[0051] The specific configuration of the time feature extraction module 20 will be described in detail below. The time feature extraction module 20 further includes: constructing single-service time series samples and queue waiting time series samples; slicing the single-service time series samples and queue waiting time series samples to construct a basic training dataset; extracting time feature training samples and corresponding time window length label samples from the basic training dataset; constructing a lightweight neural network; training the time feature training samples and the time window length label samples based on the lightweight neural network to obtain a trained end-side adaptive time window module; inputting the time features into the end-side adaptive time window module to calculate the time window length for each transmission end, wherein the time features include average service time, average waiting time, fluctuation amount, and queue change frequency.

[0052] The specific configuration of the time feature extraction module 20 will be described in detail below. The time feature extraction module 20 further includes: obtaining a candidate time-series window set; performing simulated compression on the basic training dataset based on the candidate time-series window set to obtain a compressed simulated dataset; calculating a delay penalty index for the compressed simulated dataset using a delay penalty function; performing a minimization search on the candidate time-series window set according to the delay penalty index; outputting a preferred time-series window set for each basic training dataset; and performing labeling processing on the preferred time-series window set to construct time-series window length label samples.

[0053] The specific configuration of the differential data calculation module 30 will be described in detail below. The differential data calculation module 30 further includes: dynamically slicing the time axis based on the time window length to output continuous time slice intervals; extracting traffic volume indicators, efficiency indicators, queuing status indicators, and trend indicators from the continuous time slice intervals to output end-side state vectors; and comparing the end-side state vectors of adjacent time slices within the continuous time slice intervals field-by-field to extract differential state vector data based on differential masks and differential values.

[0054] The specific configuration of the data compression processing module 40 will be described in detail below. The data compression processing module 40 further includes: inputting the end-side state vector of the previous time slice in the continuous time slice interval into the lightweight prediction model to obtain the predicted end-side state vector of the continuous time slice, wherein the lightweight prediction model is a Markov probability prediction model based on state transition rules; performing a field-by-field comparison between the predicted end-side state vector and the actual end-side state vector to obtain prediction error data; and extracting the residual mask and residual values ​​from the prediction error data to output prediction residual data.

[0055] The specific configuration of the data compression processing module 40 will be described in detail below. The data compression processing module 40 further includes: fusing the state vector difference data and the prediction residual data to construct a fused data structure; performing bit-level compression encoding on the difference mask and residual mask of the fused data structure; performing variable-length compression encoding on the difference values ​​and residual values ​​of the fused data structure; and outputting compressed encoded data.

[0056] The specific configuration of the redundancy difference analysis module 50 will be described in detail below. The redundancy difference analysis module 50 further includes: obtaining a preset target redundancy on the cloud side; performing redundancy analysis on multiple compressed encoded data corresponding to the multiple transmission ends to obtain multiple redundancies; performing redundancy difference analysis based on the preset target redundancy and the multiple redundancies corresponding to the multiple transmission ends to obtain individual redundancy differences and global redundancy differences; obtaining multiple redundancy differences corresponding to the multiple transmission ends based on the individual redundancy differences and the global redundancy differences; and updating the timing window length based on each redundancy difference.

[0057] The low-redundancy compressed queuing and calling data transmission device provided in this embodiment of the invention can execute the low-redundancy compressed queuing and calling data transmission method provided in this embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A low-redundancy compressed data transmission method for queuing and calling systems in China, characterized in that: The method includes: Collect data from multiple transmission endpoints connected to the cloud, and collect the single service time series and queue waiting time series for each of the multiple transmission endpoints. Extract the time features of the single service time series and the queue waiting time series, and input the time features into the adaptive timing window module on the transmission end side to determine the timing window length for each transmission end side; Based on the time window length, construct the end-side state vector of each transmission end, and calculate the state vector difference data of adjacent time slices; The prediction residual data is output through a lightweight prediction model, and compressed data is output based on the state vector difference data and the prediction residual data. The cloud side performs redundancy difference analysis on multiple compressed encoded data corresponding to the multiple transmission end sides according to a preset redundancy, and updates the time window length according to the multiple redundancy differences.

2. The low-redundancy compressed queuing and calling data transmission method for domestically developed information technology as described in claim 1, characterized in that, The method for determining the timing window length for each transmission end by the adaptive timing window module at the input end of the time feature includes: Construct single-service time series samples and queue waiting time series samples, slice the single-service time series samples and queue waiting time series samples, and construct a basic training dataset; Extract the temporal feature training samples and the corresponding temporal window length label samples from the basic training dataset; A lightweight neural network is constructed, and the time feature training samples and the time window length label samples are trained based on the lightweight neural network to obtain a trained end-side adaptive time window module; The time features are input into the end-side adaptive timing window module to calculate the timing window length for each transmission end. The time features include average service time, average waiting time, fluctuation amount, and queue change frequency.

3. The low-redundancy compressed queuing and calling data transmission method for information technology innovation as described in claim 2, characterized in that, Methods for constructing time-series window length label samples include: Obtain a candidate time series window set, and perform simulated compression on the basic training dataset based on the candidate time series window set to obtain a compressed simulated dataset; The delay penalty index of the compressed simulation dataset is calculated using a delay penalty function; Minimize the candidate time window set according to the delay penalty index, output the preferred time window set for each basic training dataset, and perform labeling processing on the preferred time window set to construct time window length label samples.

4. The low-redundancy compressed queuing and calling data transmission method for domestically developed information technology as described in claim 1, characterized in that, Based on the time window length, an end-side state vector is constructed for each transmission end, and the state vector difference data between adjacent time slices is calculated. The method includes: The time axis is dynamically sliced ​​based on the time window length, and continuous time slice intervals are output. Extract the traffic volume indicators, efficiency indicators, queuing status indicators, and trend indicators of the continuous time slice interval, and output the end-side state vector; The end-side state vectors of adjacent time slices in the continuous time slice interval are compared field by field to extract state vector differential data based on differential mask and differential value.

5. The low-redundancy compressed queuing and calling data transmission method for information technology innovation as described in claim 4, characterized in that, Methods for outputting prediction residual data using lightweight prediction models include: The end-side state vector of the previous time slice in the continuous time slice interval is input into the lightweight prediction model to obtain the predicted end-side state vector of the continuous time slice. The lightweight prediction model is a Markov probability prediction model based on state transition rules. The prediction error data is obtained by comparing the predicted terminal state vector with the actual terminal state vector field by field. Based on the prediction error data, extract the residual mask and residual values, and output the prediction residual data.

6. The low-redundancy compressed queuing and calling data transmission method for domestically developed information technology as described in claim 5, characterized in that, The method involves compressing the state vector difference data and the prediction residual data to output compressed coded data, including: The state vector difference data and the prediction residual data are fused to construct a fused data structure; Bit-level compression encoding is performed on the differential mask and residual mask of the fused data structure, and variable-length compression encoding is performed on the differential value and residual value of the fused data structure to output compressed encoded data.

7. The low-redundancy compressed queuing and calling data transmission method for domestically developed information technology as described in claim 1, characterized in that, The cloud side performs redundancy difference analysis on multiple compressed encoded data corresponding to the multiple transmission ends based on a preset redundancy, and the method includes: Obtain the preset target redundancy on the cloud side, and perform redundancy calculations on the multiple compressed encoded data corresponding to the multiple transmission end sides to obtain multiple redundancy values; Redundancy difference analysis is performed based on the preset target redundancy and the multiple redundancies corresponding to the multiple transmission end sides to obtain individual redundancy difference and global redundancy difference; Based on the individual redundancy difference and the global redundancy difference, multiple redundancy differences corresponding to the multiple transmission end sides are obtained, and the timing window length is updated according to each redundancy difference.

8. A low-redundancy compressed queuing and calling data transmission device for domestically developed information technology, characterized in that, The apparatus is used to implement the low-redundancy compressed queuing and calling data transmission method according to any one of claims 1 to 7, and the apparatus comprises: The time series acquisition module is used to acquire data from multiple transmission endpoints connected to the cloud side, and to acquire the single service time series and queue waiting time series of each of the multiple transmission endpoints. The time feature extraction module is used to extract the time features of the single service time series and the queue waiting time series, and input the time features into the adaptive timing window module on the input end side to determine the timing window length of each transmission end side; The differential data calculation module is used to construct the end-side state vector of each transmission end based on the time window length, and calculate the state vector differential data of adjacent time slices; The data compression processing module is used to output prediction residual data through a lightweight prediction model, and to compress and output compressed encoded data based on the state vector difference data and the prediction residual data. The redundancy difference analysis module is used to perform redundancy difference analysis on multiple compressed encoded data corresponding to multiple transmission ends on the cloud side according to a preset redundancy, and update the time window length according to multiple redundancy differences.

Images