Session recovery method and apparatus after call drop

Abstract

The application discloses a session recovery method and device after call disconnection, and relates to the technical field of real-time communication. The method comprises the following steps: obtaining a first compressed session state feature vector before a first time and a second compressed session state feature vector after a second time of a target session; determining a third compressed session state feature vector of at least one disconnection recovery time according to the first time, the second time, the first compressed session state feature vector and the second compressed session state feature vector; fusing a target center feature vector corresponding to each third compressed session state feature vector, and performing inverse normalization processing on a target feature vector obtained by the fusion to obtain a recovered multi-dimensional session state parameter; and controlling a telephone soft switching platform to perform continuity recovery on the target session according to the recovered multi-dimensional session state parameter, so that the technical problems of high memory overhead, low recovery precision and low recovery efficiency of related call disconnection recovery technologies are solved.

Description

Technical Field

[0001] This application relates to the field of real-time communication technology, and more specifically, to a method and apparatus for resuming a session after a call is disconnected. Background Technology

[0002] In real-time communication scenarios such as enterprise IP PBX (IP Private Branch Exchange) systems, customer service call centers, and remote video conferencing systems, the complexity of the network environment is a major cause of call interruptions. Specifically, user terminals are highly susceptible to momentary network outages when switching Wi-Fi networks, experiencing LAN fluctuations, or encountering unstable mobile network signals. This brief interruption of the physical link directly causes the RTP (Real-time Transport Protocol) media stream transmission to stall, and without an effective session recovery mechanism, the communication connection will be completely broken. Therefore, how to quickly restore the session without interrupting the user's subjective experience has become a challenge in ensuring business continuity.

[0003] To address the issue of call recovery after a disconnection, existing technical solutions mainly fall into two categories, but both have significant technical drawbacks:

[0004] One approach is "full cache recovery technology." Its core principle is to pre-cache the complete RTP packet sequence and detailed session media parameter history while the session is connected. Once a call disconnection is detected, the cached data is directly replayed to rebuild the session. However, this approach has significant drawbacks: due to the need to store massive amounts of media data, its memory overhead is extremely high; in high-load call systems with tens of thousands of concurrent connections, full caching is highly susceptible to server memory overflow risks, significantly increasing hardware deployment and expansion costs. Furthermore, the high latency of data read / write operations makes it difficult to meet the millisecond-level real-time recovery requirements, resulting in low recovery efficiency.

[0005] The second approach is "coarse-grained state recovery technology." Its core principle is to cache only a small amount of static session metadata (such as encoding type, initial RTP packet sequence number, etc.). Upon detecting a call disconnection, it jumps directly to the latest cached state to continue transmission, attempting to achieve recovery with low overhead. However, this approach ignores the dynamic continuity of session states and the temporal correlation of media streams. Due to the lack of fine-grained derivation of state evolution during the disconnection, the recovered media stream often exhibits severe audio-visual misalignment, playback stuttering, and audio-visual asynchrony. This low-precision recovery method cannot achieve true seamless transition, leading to a significant decline in user experience, particularly negatively impacting customer service connection rates and the continuity of meeting processes.

[0006] There is currently no effective solution to the above problems. Summary of the Invention

[0007] This application provides a method and apparatus for resuming a session after a call is disconnected, in order to at least solve the technical problems of high memory overhead, low recovery accuracy, and low recovery efficiency in related call disconnection recovery technologies.

[0008] According to one aspect of the embodiments of this application, a method for restoring a session after a call disconnection is provided, comprising: obtaining a first compressed session state feature vector of a target session before a first moment and a second compressed session state feature vector after a second moment in a telephone softswitch platform, wherein the first moment is the moment when the target session disconnects and the second moment is the moment when the target session resumes; determining at least one disconnection recovery moment and its corresponding weight value based on the first moment and the second moment, and determining a third compressed session state feature vector for each disconnection recovery moment based on the first compressed session state feature vector and the second compressed session state feature vector; determining a target center feature vector corresponding to each third compressed session state feature vector from a preset mapping table, fusing the target center feature vectors according to the weight value to obtain a target feature vector, and performing inverse normalization processing on the target feature vector to obtain a recovery multidimensional session state parameter, wherein the mapping table stores the center feature vectors corresponding to each of the multiple compressed session state feature vectors; and controlling the telephone softswitch platform to continuously restore the target session according to the recovery multidimensional session state parameter.

[0009] According to another aspect of the embodiments of this application, a session recovery device after a call disconnection is also provided, comprising: an acquisition module, configured to acquire a first compressed session state feature vector of a target session before a first moment and a second compressed session state feature vector after a second moment in a telephone softswitch platform, wherein the first moment is the moment when the target session experiences a call disconnection and the second moment is the moment when the target session resumes operation; a first determination module, configured to determine at least one disconnection recovery moment and its corresponding weight value based on the first moment and the second moment, and to determine a third compressed session state feature vector for each disconnection recovery moment based on the first compressed session state feature vector and the second compressed session state feature vector; a second determination module, configured to determine a target center feature vector corresponding to each third compressed session state feature vector from a preset mapping table, fuse each target center feature vector according to the weight value to obtain a target feature vector, and perform inverse normalization processing on the target feature vector to obtain recovery multidimensional session state parameters, wherein the mapping table stores the center feature vectors corresponding to each of the multiple compressed session state feature vectors; and a recovery module, configured to control the telephone softswitch platform to continuously recover the target session based on the recovery multidimensional session state parameters.

[0010] In an exemplary embodiment, the above-described apparatus is further configured to obtain a first compressed session state feature vector of a target session in a telephone softswitch platform prior to a first moment by means of the following method: collecting and obtaining session state monitoring indicators of the target session at preset time intervals; at each moment, determining whether the target session has experienced a call disconnection at the current moment based on the session state monitoring indicators at the current moment; and if the target session experiences a call disconnection at the first moment, obtaining the first compressed session state feature vector of the target session prior to the first moment from a preset memory queue, wherein the memory queue retains compressed session state feature vectors corresponding to multiple moments when the target session has not experienced a call disconnection.

[0011] In an exemplary embodiment, the session state monitoring indicators include at least: the cumulative duration of no real-time transport protocol data packets received before the corresponding time, the round-trip transmission delay and packet loss rate of sending network probe messages to the network peer at the corresponding time, the session state variables at the corresponding time, and the control signaling events reported by the telephone softswitch platform at the corresponding time; the above-mentioned apparatus is further configured to determine whether the target session has experienced a call disconnection at the current time based on the session state monitoring indicators by the following methods: triggering a media stream interruption alarm when the cumulative duration is not less than a preset duration threshold; triggering a network link anomaly alarm when the round-trip transmission delay is not less than a preset delay threshold or the packet loss rate is not less than a preset packet loss rate threshold; triggering a session abnormal interruption alarm when the session state variables and control signaling events are inconsistent; determining that the target session has experienced a call disconnection when the target session triggers at least two of the media stream interruption alarm, network link anomaly alarm, and session abnormal interruption alarm simultaneously; and determining that the target session has not experienced a call disconnection when the target session triggers only any one of the media stream interruption alarm, network link anomaly alarm, and session abnormal interruption alarm.

[0012] In an exemplary embodiment, before acquiring the first compressed session state feature vector of the target session before the first moment, the above-mentioned apparatus is further configured to control the telephone softswitch platform to collect multi-dimensional session state parameters of the target session at preset time intervals before the first moment. The multi-dimensional session state parameters include at least: media transmission sequence parameters, media encoding feature parameters, network link quality parameters, and call control state parameters, and the call control state parameters include at least: session universal unique identifier and initiation time. For the multi-dimensional session state parameters at each moment, the multi-dimensional session state parameters are preprocessed, and the processed multi-dimensional session state parameters are combined based on a preset session correlation coefficient to obtain a session state feature vector. The session state feature vector is then processed using a pre-trained mapping function to obtain the corresponding compressed session state feature vector. The preprocessing includes: abnormal data detection, abnormal data correction, and standardization processing. The compressed session state feature vectors are stored sequentially in a memory queue according to the time sequence, and when the storage capacity of the memory queue reaches a preset capacity threshold, the earliest compressed session state feature vector in the memory queue is removed according to the first-in-first-out principle.

[0013] In an exemplary embodiment, the above-described apparatus is further configured to map the session state feature vector using a pre-trained mapping function to obtain a corresponding compressed session state feature vector, including: linearly projecting the session state feature vector using a pre-trained projection matrix to obtain a one-dimensional session state feature sequence, wherein the projection matrix is ​​trained based on historical session data from multiple historical sessions, and during the training process, the projection matrix is ​​iteratively updated using a gradient descent algorithm with the objective function of minimizing storage overhead and session recovery error as the optimization objective until a preset convergence condition is met, wherein the historical session data includes at least: historical compressed session state feature vectors, recovery result identifiers, and recovered call quality indicators; equidistantly quantizing the one-dimensional session state feature sequence according to a preset spatial dimension of the compressed session state feature space to obtain multiple discrete session state values ​​equal to the spatial dimension, and forming a compressed session state feature vector from the multiple discrete session state values, wherein the spatial dimension of the compressed session state feature space is obtained by analyzing the value range of the session state feature vector of the telephone softswitch platform using the elbow rule.

[0014] In an exemplary embodiment, the above-described apparatus is further configured to obtain a second compressed session state feature vector of a target session in a telephone softswitch platform after a second time point by means of the following method: obtaining new multidimensional session state parameters of a new session in the telephone softswitch platform after a second time point, wherein the new session is a session re-initiated after the second time point, and the new multidimensional session state parameters include at least: media transmission sequence parameters, media encoding feature parameters, network link quality parameters, and call control state parameters, and the call control state parameters include at least: a session universally unique identifier and an initiation time; if the session universally unique identifier in the new multidimensional session state parameters is consistent with the session universally unique identifier of the target session, and the time interval between the initiation time of the new multidimensional session state parameters and the first time point is less than a preset duration threshold, determining that the new session and the target session are associated sessions; preprocessing the new multidimensional session state parameters, and combining the processed new multidimensional session state parameters based on a preset session association coefficient to obtain a second session state feature vector, and processing the second session state feature vector using a pre-trained mapping function to obtain a second compressed session state feature vector, wherein the preprocessing includes: abnormal data detection, abnormal data correction, and standardization processing.

[0015] In an exemplary embodiment, the above-described apparatus is further configured to determine at least one disconnection recovery time and its corresponding weight value based on a first time and a second time by the following method: determining the difference between the second time and the first time to obtain the disconnection duration; determining the number of median points based on the disconnection duration and a preset median point granularity adaptive strategy, and dividing the disconnection duration based on the number of median points to obtain at least one disconnection recovery time; and determining the weight value of each of the at least one disconnection recovery time based on a weight allocation rule based on the principle of symmetric attenuation.

[0016] In an exemplary embodiment, the above-described apparatus is further configured to determine a third compressed session state feature vector at each disconnection recovery time by means of a first compressed session state feature vector and a second compressed session state feature vector, including: determining an initial compressed session state feature vector at each disconnection recovery time based on the first compressed session state feature vector, the second compressed session state feature vector, and the number of median points; for each disconnection recovery time, determining whether the initial compressed session state feature vector at the disconnection recovery time exceeds a preset compressed session state feature space; if the initial compressed session state feature vector at the disconnection recovery time exceeds the preset compressed session state feature space, correcting the initial compressed session state feature vector at the disconnection recovery time according to the spatial boundary of the compressed session state feature space to obtain a third compressed session state feature vector at the disconnection recovery time; if the initial compressed session state feature vector at the disconnection recovery time does not exceed the compressed session state feature space, using the initial compressed session state feature vector as the third compressed session state feature vector at the disconnection recovery time.

[0017] In an exemplary embodiment, the above-described apparatus is further configured to construct a mapping table by means of the following method: randomly combining the dimensional values ​​of multiple spatial dimensions of a preset compressed session state feature space to obtain multiple compressed session state feature vectors; for each compressed session state feature vector, determining multiple historical session state feature vectors mapped to the compressed session state feature vector from a preset database, performing clustering processing on the multiple historical session state feature vectors to obtain corresponding central feature vectors, wherein the preset database stores multiple historical session state feature vectors and compressed session state feature vectors corresponding to each historical session state feature vector; and summarizing the central feature vectors corresponding to each of the multiple compressed session state feature vectors to obtain a mapping table.

[0018] In an exemplary embodiment, after restoring the target session based on the restored multidimensional session state parameters, the apparatus is further configured to reacquire new session state monitoring indicators for the target session. These session state monitoring indicators include at least: the cumulative duration of no received Real-Time Transport Protocol (RTP) data packets, and the round-trip transmission delay and packet loss rate for sending network probe messages to the network peer. If the cumulative duration is less than a preset duration threshold, the round-trip transmission delay is less than a preset delay threshold, and the packet loss rate is less than a preset packet loss rate threshold, the restored multidimensional session state parameters are determined to have passed verification. If the cumulative duration is not less than the preset duration threshold, or the round-trip transmission delay is not less than the preset delay threshold, or the packet loss rate is not less than the preset packet loss rate threshold, the restored multidimensional session state parameters are determined to have failed verification.

[0019] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein the computer-readable storage medium, when executed by a processor, performs the steps in any of the above method embodiments.

[0020] According to another aspect of the embodiments of this application, a computer program product or computer program is also provided, which includes a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, causing the computer device to perform the steps in any of the method embodiments described above.

[0021] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the steps of any of the above method embodiments through the computer program.

[0022] In this embodiment, a first compressed session state feature vector before a first moment and a second compressed session state feature vector after a second moment are obtained from the target session in the telephone softswitch platform. The first moment is the moment when the target session experiences a call disconnection, and the second moment is the moment when the target session resumes operation. At least one disconnection recovery moment and its corresponding weight value are determined based on the first and second moments. A third compressed session state feature vector for each disconnection recovery moment is determined based on the first and second compressed session state feature vectors. The target center feature vector corresponding to each third compressed session state feature vector is determined from a preset mapping table. The target center feature vectors are fused according to the weight values ​​to obtain a target feature vector. The target feature vector is then denormalized to obtain recovery multidimensional session state parameters. The mapping table stores the center feature vectors corresponding to multiple compressed session state feature vectors. The telephone softswitch platform is controlled to continuously restore the target session based on the recovery multidimensional session state parameters. The above technical solution achieves the effects of accurately connecting breakpoints without full caching, greatly reducing memory usage, and significantly shortening recovery time. It achieves the goal of ensuring high concurrency, low overhead, and seamless recovery of calls in complex network environments, thereby solving the technical problems of high memory consumption, low recovery accuracy, and low recovery efficiency in related call disconnection recovery technologies. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0024] Figure 1 This is a flowchart illustrating an optional method for resuming a session after a call is disconnected, according to an embodiment of this application.

[0025] Figure 2 This is a schematic diagram of an optional process for processing multidimensional session state parameters according to an embodiment of this application;

[0026] Figure 3 This is a schematic diagram of an optional process for determining the disconnection recovery time and weight value according to an embodiment of this application;

[0027] Figure 4 This is a schematic diagram of an optional process for determining the third compressed session state feature vector according to an embodiment of this application;

[0028] Figure 5 This is a structural block diagram of an optional call recovery device according to an embodiment of this application;

[0029] Figure 6 This is a structural block diagram of a computer terminal according to an embodiment of the present application for implementing a session recovery method after a call is dropped. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0031] It should be noted that the terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] To better understand the embodiments of this application, the following is a translation and explanation of some nouns or terms that appear in the description of the embodiments of this application:

[0033] Linear function normalization: By applying a linear transformation to the original data, the result is mapped to the range [0,1], achieving proportional scaling of the original data. The normalization formula is as follows: In the formula, X represents the original data. These represent the maximum and minimum values ​​of the data, respectively.

[0034] According to an embodiment of this application, a session recovery method after a call is dropped in a call system (which can be deployed on a server of a telephone softswitch platform or an external computing node) is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than that shown here.

[0035] Figure 1 This is a flowchart illustrating a method for resuming a call after it has been disconnected, according to an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0036] Step S102: Obtain the first compressed session state feature vector of the target session before the first moment and the second compressed session state feature vector after the second moment in the telephone softswitch platform.

[0037] In this embodiment, the aforementioned telephone softswitch platform (i.e., FreeSwitch) can provide voice calls, instant messaging, and multimedia routing functions. It supports multiple communication protocols and can be deployed as a switch engine, multimedia gateway, or server.

[0038] In this embodiment, the target session is a real-time audio and video call session established on a telephone softswitch platform, where the media stream is interrupted due to a network outage but has not yet been fully released (or can be quickly re-established). Therefore, the core engine of the telephone softswitch platform can control the lifecycle, session state variables, and media stream management of the real-time audio and video call session.

[0039] In this embodiment of the application, the first moment is the specific moment when the target session disconnects, and the first compressed session state feature vector is the last valid low-dimensional discrete compressed state value (i.e., the anchored state before disconnection) selected from the original session after preprocessing and compression. The second moment is the specific moment when the target session resumes operation, and the second compressed session state feature vector is the first valid low-dimensional discrete compressed state value (i.e., the anchored state after recovery) collected from the newly resumed session after the second moment.

[0040] In this embodiment of the application, the aforementioned call disconnection is an event in a real-time audio and video call session based on a telephone softswitch platform where media stream transmission is interrupted due to network interruption (such as Wi-Fi switching or LAN fluctuations), but the call session logic has not been fully released or can be quickly reconnected.

[0041] Step S104: Determine at least one disconnection recovery time and its corresponding weight value based on the first time and the second time, and determine the third compressed session state feature vector for each disconnection recovery time based on the first compressed session state feature vector and the second compressed session state feature vector.

[0042] In this embodiment of the application, the above-mentioned disconnection recovery time is an intermediate time derived from the first time and the second time and used to connect the disconnection point. Therefore, these times are not the real system time, but logical interpolation points in the state space.

[0043] In this embodiment of the application, the aforementioned third compressed session state feature vector is an intermediate compressed session state feature vector derived from the first compressed session state feature vector and the second compressed session state feature vector through compressed state transition operation. These vectors represent various intermediate discrete states in the process of smoothly transitioning from the anchored state before the disconnection to the anchored state after the disconnection is restored.

[0044] Step S106: Determine the target center feature vector corresponding to each third compressed session state feature vector from the preset mapping table, fuse each target center feature vector according to the weight value to obtain the target feature vector, and perform inverse normalization processing on the target feature vector to obtain the recovered multidimensional session state parameters.

[0045] In this embodiment, the mapping table is established through offline training and stores the central feature vectors corresponding to each of the possible compressed session state feature vectors. Specifically, each index (i.e., compressed session state feature vector) in the table corresponds to a central feature vector, and the central feature vector represents the statistical average or cluster center value of the original parameters under the compressed session state feature vector.

[0046] In the embodiments of this application, the above-mentioned inverse normalization process is the process of restoring the feature parameter values ​​in the target feature vector to the actual physical dimension values.

[0047] In the embodiments of this application, the above-mentioned recovered multidimensional session state parameters are a set of session state data containing complete original physical dimensions obtained after inverse normalization processing, such as media transmission sequence parameters (i.e., real-time transmission protocol data packet sending / receiving sequence number, timestamp, etc.), media encoding characteristic parameters (i.e., encoding type, bit rate, etc.), and network link quality parameters (i.e., jitter value, packet loss rate, etc.).

[0048] Step S108: Control the telephone softswitch platform to continuously restore the target session based on the recovery multi-dimensional session state parameters.

[0049] In this embodiment, the aforementioned continuity recovery utilizes derived recovery multidimensional session state parameters, which are directly injected into the media stream being recovered via the control interface of the telephone softswitch platform (such as event socket commands). This ensures that the sequence number and timestamp of the recovered Real-time Transport Protocol (RTP) data packets are logically continuous with the last RTP data packet before the disconnection, preventing audio and video playback from stuttering, muting, or out of order. The entire process does not require a complete release and re-establishment of the Session Initiation Protocol (SIP) call flow, thereby significantly shortening the call recovery time.

[0050] Based on the scheme defined in steps S102 to S108 above, it can be understood that in this embodiment, at least one intermediate disconnection recovery time is determined based on the disconnection time and disconnection recovery time of the target session. Combining the compressed session state feature vectors before and after the disconnection, the intermediate compressed session state feature vectors of each intermediate disconnection recovery time are derived by linear interpolation and inversely mapped back to the recovery multi-dimensional session state parameters, so as to continuously restore the target session. This achieves the technical effects of accurately connecting breakpoints without full caching, greatly reducing memory usage, and significantly shortening recovery time. It achieves the goal of ensuring high concurrency, low overhead, and seamless recovery of calls in complex network environments, thereby solving the technical problems of high memory overhead, low recovery accuracy, and low recovery efficiency in related call disconnection recovery technologies.

[0051] The following describes each step of the session recovery method after call disconnection provided in this application embodiment, with reference to the specific implementation process.

[0052] To achieve rapid and seamless recovery after a call disconnection, the primary challenge is to accurately capture and standardize multi-dimensional session state parameters before and after the disconnection as a recovery benchmark. However, directly collected raw session state parameters suffer from issues such as inconsistent dimensions, interference from abnormal data, and difficulties in efficiently compressing and storing high-dimensional features. To address this, this application proposes the following method for collecting and processing multi-dimensional session state parameters, thereby providing high-quality, low-overhead, and standardized data input for subsequent connection derivation, ensuring the accuracy and real-time performance of the recovery process.

[0053] As an optional implementation, the call system can be implemented through, for example... Figure 2 The process shown involves collecting and processing multi-dimensional session state parameters of the target session:

[0054] Step S11: Before the first moment, control the telephone softswitch platform to collect multi-dimensional session state parameters of the target session at preset time intervals.

[0055] The above steps can be understood as follows: The calling system establishes a persistent TCP network connection with the telephone softswitch platform through the Event Socket Library (ESL) protocol, and registers with the telephone softswitch platform to listen for specific signaling and media events. When a session state change (such as call setup or connection) or media stream activity (such as Real-Time Transport Protocol (RTP) data packet transmission and reception events) occurs in the telephone softswitch platform, the telephone softswitch platform will actively push these events to the calling system through this long connection, thereby triggering the subsequent parameter acquisition process.

[0056] Specifically, the parameter acquisition process is as follows: When the target session is detected to have entered the connected state (i.e., before the first moment), the calling system sends "uuid_getvar" commands to the telephone softswitch platform in batches at preset time intervals to collect multi-dimensional session state parameters of the target session. The preset time interval can be set according to actual application requirements and system overhead, such as 20ms / time. The collected multi-dimensional session state parameters are multi-dimensional parameters, including:

[0057] (1) Media transmission sequence parameters, such as the RTP data packet sequence number of the sending end. The RTP packet sequence number at the receiving end RTP timestamp ts (in milliseconds);

[0058] (2) Media coding characteristic parameters, such as media coding type codec, coding bit rate br, sampling rate sr, etc.;

[0059] (3) Network link quality parameters, such as jitter (in milliseconds) and packet loss rate (in %).

[0060] (4) Call control status parameters, such as call duration (Unit: seconds), Session Universal Unique Identifier .

[0061] For example, the collected multidimensional session state parameters can be .

[0062] In addition, multiple sets of multidimensional session state parameters collected can be stored sequentially in a local database according to the order of collection time. The local database can be a circular buffer, so if the number of stored parameters exceeds the maximum storage capacity of the circular buffer, the excess will automatically overwrite the earliest multidimensional session state parameters collected at the time of collection, and a validity identifier will be marked for each set of session state data (if RTP packets are not received for 3 consecutive times during collection, the session state parameters at the corresponding collection time will be marked as invalid).

[0063] Step S12: For the multidimensional session state parameters at each time step, preprocess the multidimensional session state parameters, combine the processed multidimensional session state parameters based on the preset session correlation coefficient to obtain the session state feature vector, and use the pre-trained mapping function to process the session state feature vector to obtain the corresponding compressed session state feature vector.

[0064] The above steps can be broken down into the following execution steps:

[0065] Step 1: First, preprocess the collected multidimensional session state parameters. This preprocessing includes, but is not limited to, anomaly detection, anomaly correction, and standardization.

[0066] (1) Abnormal data detection is a process of monitoring the collected multi-dimensional session state parameters through a dual mechanism of threshold judgment and trend analysis to identify “dirty data” or “noisy data” that do not conform to normal network fluctuation patterns or physical logic. For example, in normal real-time audio and video calls, jitter is usually controlled within 20ms-30ms. Once the jitter exceeds 50ms, it means that the unevenness of data packet arrival time has increased significantly, which can easily lead to overflow or underload of the receiver's jitter buffer. Therefore, jitter values ​​exceeding 50ms can be marked as abnormal data. In normal real-time audio and video calls, a packet loss rate exceeding 15% is usually considered an unacceptable low-quality call or even a precursor to call interruption. Therefore, a packet loss rate exceeding 15% can be marked as abnormal data. Since the sequence number of RTP data packets is monotonically increasing, usually by 1 each time, the normal range for sequence number jumps within a sampling period of 20mA to 50ms is between 1 and 4. Once the RTP data packet sequence number jump exceeds 1000, it indicates that there is a serious data gap or session state tear before and after the currently collected data point. Therefore, RTP data packet sequence number jumps exceeding 1000 can be marked as abnormal data.

[0067] (2) Abnormal data correction is to not directly discard the data when the multidimensional session state parameters have abnormal values ​​that do not conform to the normal network fluctuation pattern. Instead, the theoretical value that the abnormal data should have is calculated by using the previous valid data and the corresponding time, the next valid data and the corresponding time, and the linear interpolation algorithm. The original abnormal value is replaced with this theoretical value, thereby ensuring the continuity and smoothness of the session parameter time series.

[0068] For example, in five consecutive multidimensional session state parameters, if the jitter in the third multidimensional session state parameter is 62ms (exceeding the threshold of 50ms), and the second and fourth multidimensional session state parameters are d(1698300000080, 25ms) and d(1698300000120, 28ms) respectively, then the correction value is... .

[0069] (3) Standardization processing includes: numerical mapping for discrete parameters, normalization processing for continuous parameters, and processing for proportional parameters. Specifically: For discrete parameters, numerical mapping involves converting non-numerical or categorical parameters with fixed value sets (such as media encoding types: G.711a, G.729, OPUS, etc.) into corresponding integer codes (e.g., G.711a corresponds to an integer code of 1, G.729 corresponds to an integer code of 3, and OPUS corresponds to an integer code of 4) through predefined rules, thus transforming discrete variables in text or symbolic form into numerical forms that can be processed by computers; For continuous parameters, normalization involves linearly mapping continuous numerical parameters with vastly different value ranges (such as RTP packet sequence numbers, send / receive timestamps, bit rates, jitter values, etc.) to a unified [0,1] interval through linear function normalization (such as the Min-Max normalization formula), thereby eliminating dimensional and numerical magnitude differences between different parameter dimensions and preventing certain parameters with large numerical ranges from dominating the feature vector construction or state compression process; For proportional parameters (such as packet loss rate), the original value is usually directly retained.

[0070] Step 2: Combine the processed multidimensional session state parameters based on the preset session correlation coefficient to obtain the session state feature vector. The preset session correlation coefficient is obtained based on the analytic hierarchy process (AHP).

[0071] Specifically, the correlation between multi-dimensional session state parameters and session continuity, media synchronization, and network quality is categorized into three levels: high, medium, and low. The importance of each pair of parameters is determined through expert scoring or historical data analysis. Based on the resulting judgment matrix, a preset session correlation coefficient for each session state parameter is calculated. For example, if the total weight coefficient for high-level parameters is set to 0.6, and high-level parameters only contain... and If so, then and The preset session correlation coefficients for each parameter are 0.3. If the total weight coefficient for the mid-level parameters is 0.3, and the mid-level parameters only include the media encoding type (codec) and timestamp (ts), then the preset session correlation coefficients for the media encoding type (codec) and timestamp (ts) are 0.15. If the total weight coefficient for the low-level parameters is 0.1, and the low-level parameters include jitter, packet loss rate (plr), bitrate (br), and sampling rate (sr), then the preset session correlation coefficients for jitter, packet loss rate (plr), bitrate (br), and sampling rate (sr) can be 0.03, 0.03, 0.02, and 0.02, respectively. Therefore, the session state feature vector can be written as: In the formula Indicates the sequence number of the preprocessed RTP data packets sent by the sender. Indicates the sequence number of the preprocessed RTP data packets received. Indicates the preprocessed media encoding type, Represents the preprocessed RTP timestamp, Indicates the preprocessed encoding bit rate, Indicates the sampling rate after preprocessing. Indicates the jitter value after preprocessing. This indicates the packet loss rate after preprocessing.

[0072] It should be noted that the sum of the preset session correlation coefficients for all dimensions of session state parameters is 1.

[0073] Step 3: Use the mapping function to process the session state feature vector to obtain the corresponding compressed session state feature vector.

[0074] The mapping function described above is used to compress high-dimensional feature vectors into low-dimensional discrete state values, thereby reducing storage overhead while preserving the state transition rules.

[0075] Alternatively, the above processing solution can be implemented through the following steps:

[0076] First, the session state feature vector is linearly projected using a pre-trained projection matrix to obtain a one-dimensional session state feature sequence.

[0077] The projection matrix is ​​trained based on the historical session state feature vectors of multiple historical sessions. During the training process, the gradient descent algorithm is used to iteratively update the projection matrix with the optimization objective of minimizing the combined function of storage overhead and session recovery error, until the preset convergence condition is met.

[0078] Specifically, if projection matrix optimization is initiated at a fixed time each day, historical session data from multiple historical sessions over the past 24 hours (i.e., historical compressed session state feature vectors, recovery success / failure indicators, and post-recovery call quality metrics) can be used as the training set. Simultaneously, the objective function is defined as follows: In the formula This represents the session recovery error (i.e., the deviation rate between the sequence number of the recovered RTP packets and the actual sequence number of the RTP packets). Indicates the storage overhead of compressed state. Represents the weighting coefficient (e.g.) ),and Then, in each iteration, the projection matrix obtained in the previous iteration is updated using gradient descent: In the formula This represents the learning rate (e.g., 0.001). This represents the projection matrix obtained from the previous iteration. This represents the new projection matrix obtained in the current iteration. Repeat the above iterative optimization process until the preset convergence condition is met (e.g., reaching the maximum number of iterations) to obtain the final optimized projection matrix, such as... .

[0079] Therefore, the multidimensional session state parameters can be represented using the pre-trained projection matrix described above. Projection as a one-dimensional session state feature sequence .

[0080] Then, the one-dimensional session state feature sequence is quantized at equal intervals according to the preset spatial dimension of the compressed session state feature space to obtain multiple discrete session state values ​​with the same spatial dimension, and the multiple discrete session state values ​​are combined to form a compressed session state feature vector.

[0081] The spatial dimension of the compressed session state feature space is obtained by analyzing the value range of the session state feature vector of the telephone softswitch platform using the elbow rule. For example, if the determined spatial dimension is 128, then the compressed session state feature vector... The aforementioned isometric quantization is the one-dimensional session state feature sequence obtained by linear projection. The range of values The space is divided into multiple discrete intervals with the same spatial dimension, and each discrete interval corresponds to a discrete state value. ,Right now In the formula, K represents the number of spatial dimensions. Represents a one-dimensional session state feature sequence The minimum value in, Represents a one-dimensional session state feature sequence The maximum value in, Represents a one-dimensional session state feature sequence Any value in, This indicates the rounding up operation.

[0082] Step S13: Store the compressed session state feature vectors into the memory queue in chronological order. When the storage capacity of the memory queue reaches the preset capacity threshold, remove the earliest compressed session state feature vector from the memory queue according to the first-in-first-out principle.

[0083] The above steps can be understood as follows: The call system first stores the compressed session state feature vectors generated in chronological order into a memory queue. Since the memory queue has limited capacity, when the queue reaches a preset threshold, the call system will automatically remove the oldest (i.e., oldest) compressed session state feature vector from the queue, following a first-in, first-out (FIFO) principle. This ensures that the latest session state snapshot is always retained in memory, providing real-time and compact data support for subsequent disconnection detection and state anchoring, while effectively preventing the risk of overflow caused by unlimited memory growth.

[0084] It should be noted that for long conversations (e.g., more than 300 seconds), the call system can extract a compressed session state feature vector at certain time intervals and store it on the disk to avoid memory overflow.

[0085] It should also be noted that the compression granularity setting can be determined by detecting the state change rate of the session, where the state change rate... In the formula denoted as the discrete state value at time t (its value range is [0, K-1]). This represents the discrete state value at time ti (its value range is also [0, K-1]), and n represents the statistical window size. For example, if n=10, then the analysis is the average of the sum of the absolute differences between the compressed session state feature vector at time t and the compressed session state feature vectors of the most recent 10 historical times. If so, it indicates a rapid change in session state, in which case the compression period can be shortened; if If so, it indicates that the session state is stable, and the compression period can be extended. Specifically, when extending the compression period, to ensure the continuity of state transitions, linear interpolation can be used to process the feature vectors of adjacent compressed session states to supplement intermediate states. In the formula, k represents interpolation compensation, and T represents the length of the extended compression period.

[0086] The above solution achieves lightweight representation and efficient management of session state parameters. It ensures that the latest high-value compressed session state feature vectors are always kept in the memory queue through the first-in-first-out principle to support millisecond-level disconnection state anchoring and rapid recovery. Furthermore, the "tiered storage" strategy sinks the compressed session state feature vectors of long sessions to the disk, which significantly reduces memory overhead in high-concurrency scenarios. While ensuring high accuracy and low latency in disconnection recovery, it greatly improves the resource utilization and concurrency carrying capacity of the call system.

[0087] As an optional approach, the first compressed session state feature vector of the target session in the telephone softswitch platform prior to the first moment is obtained, including:

[0088] Step S21: Collect and acquire session status monitoring indicators of the target session according to a preset time interval.

[0089] In this embodiment of the application, the aforementioned session state monitoring indicators are a key dataset used for multi-dimensional and accurate determination of network outage events, and include at least:

[0090] RTP timeout detection metrics: This involves monitoring the "RTPTimeout" event on the telephone softswitch platform to obtain the cumulative duration during which the target session has not received Real-Time Transport Protocol data packets before the corresponding (collection) time, which is used to identify media stream continuity interruptions.

[0091] Network connectivity testing metrics: The telephone softswitch platform sends network probe messages (such as paging commands) to the network peer (such as a media server) through ESL to obtain the round-trip transmission delay and packet loss rate of sending network probe messages (such as paging commands) to the network peer at the corresponding (collection) time, which is used to evaluate the transmission quality of the underlying network.

[0092] Session state verification metrics: These are the session state variables of the target session at the corresponding (collection) time, which are obtained by querying the telephone softswitch platform, as well as the control signaling events reported by the telephone softswitch platform at the corresponding (collection) time, and are used to confirm the abnormal state of the call control layer.

[0093] Step S22: At each time point, determine whether the target session has experienced a call disconnection at the current time based on the session status monitoring indicators at that time. Specifically:

[0094] If the cumulative duration is not less than a preset duration threshold (such as 100ms), a media stream interruption alarm will be triggered.

[0095] If the round-trip transmission delay is not less than a preset delay threshold (e.g., 300ms) or the packet loss rate is not less than a preset packet loss rate threshold (e.g., 20%), a network link anomaly alarm will be triggered.

[0096] If the session state variable and the control signaling event are inconsistent (e.g., the session state variable is "Down" but no "Channel Handup" event is reported), a session abnormal interruption alarm will be triggered.

[0097] If at least two of the following alarms are triggered simultaneously in the target session: media stream interruption alarm, network link anomaly alarm, and session abnormal interruption alarm, it is determined that the target session has experienced a call disconnection, and the subsequent recovery process is triggered.

[0098] If the target session only triggers any one of the following alarms: media stream interruption alarm, network link abnormality alarm, or session abnormality interruption alarm, it is determined that the target session has not experienced a call drop-off. This situation is usually judged as "momentary jitter" and will not trigger the subsequent recovery process.

[0099] Step S23: If the target session experiences a call disconnection at the first moment, retrieve the first compressed session state feature vector of the target session at the moment preceding the first moment from the preset memory queue.

[0100] In other words, compressed session state feature vectors with times prior to the first instant are selected from the memory queue. Compressed session state feature vectors with an "invalidity flag" of "invalid" are removed from the selected data. Then, the compressed session state feature vector with the largest time (i.e., the latest time) is selected from the removed data as the first compressed session state feature vector (i.e., the anchored state before disconnection). Additionally, the first compressed session state feature vector can be... First moment The multidimensional session state parameters corresponding to the first compressed session state feature vector and the associated session universally unique identifier Package and store the data in an "offline backup pool" (such as a Redis database, with an expiration time set to 10 minutes to avoid long-term storage occupation), and generate a unique index key. .

[0101] As an optional approach, the second compressed session state feature vector of the target session in the telephone softswitch platform after the second time point is obtained, including:

[0102] Step S31: Obtain the new multidimensional session state parameters of the new session in the telephone softswitch platform after the second time.

[0103] In this embodiment of the application, the new session is a session re-initiated by the target object after the second moment (i.e. the target session is disconnected). The new multidimensional session state parameters also include at least: media transmission sequence parameters, media encoding feature parameters, network link quality parameters, and call control state parameters. Since these four types of parameters have been specifically described above, they will not be repeated here.

[0104] Specifically, the above steps can be understood as follows: when the target session is interrupted due to a network outage, the target (such as customer service or customer) can dial again through a softphone terminal or re-establish the connection through signaling. The softswitch platform assigns a new session universal unique identifier and establishes a new SIP session channel.

[0105] Step S32: If the session universal unique identifier in the new multidimensional session state parameters is consistent with the session universal unique identifier of the target session, and the time interval between the initiation time and the first time of the new multidimensional session state parameters is less than the preset duration threshold, then the new session and the target session are determined to be associated sessions.

[0106] Specifically, by comparing the new multidimensional session state parameters with the original multidimensional session state parameters of the target session, if the session universal unique identifier in the new multidimensional session state parameters is consistent with the session universal unique identifier of the target session, and the time interval between the initiation time and the first time of the new multidimensional session state parameters is less than a preset duration threshold (such as 10 minutes), then the new session is associated with the target session.

[0107] Step S33: Preprocess the new multidimensional session state parameters, and combine the processed new multidimensional session state parameters based on the preset session correlation coefficient to obtain the second session state feature vector. Then, process the second session state feature vector using the pre-trained mapping function to obtain the second compressed session state feature vector. The preprocessing includes: abnormal data detection, abnormal data correction and standardization.

[0108] In this embodiment of the application, the process of obtaining the second compressed session state feature vector can refer to step S12 above. Since the process has been described in detail in step S12 above, it will not be described in detail here.

[0109] It should be noted that, in order to verify the rationality of the second compressed session state feature vector, the second compressed session state feature vector can be compared with the compressed session state feature vector of the same type of historical session. If the deviation result is greater than the preset deviation threshold (such as 30%), the average value of multiple second compressed session state feature vectors will be used as the final anchored state after recovery.

[0110] The above technical solution can accurately anchor the session "breakpoint" state at the instant before the network interruption occurs and the session "breakpoint" state at the instant after the network interruption is restored, avoiding anchoring deviations caused by data lag or state jumps, and providing high-fidelity start and end coordinates for subsequent derivation of the connection median based on the anchored state.

[0111] As an optional solution, the call system can be configured as follows: Figure 3 The process shown determines at least one disconnection recovery time and its corresponding weight value based on the first and second time points, including:

[0112] Step S41: Determine the difference between the second time point and the first time point to obtain the duration of the disconnection.

[0113] In this embodiment of the application, the aforementioned disconnection duration is the second moment. Subtract the first moment The difference, i.e., the duration of the disconnection. .

[0114] Step S42: Based on the disconnection duration and the preset median point granularity adaptive strategy, determine the number of median points, and divide the disconnection duration according to the number of median points to obtain at least one disconnection recovery time.

[0115] The aforementioned median point granularity adaptive strategy is based on the duration of the disconnection. The rule for dynamically adjusting the number of median points (m) derived from the connection state. For example:

[0116] If the disconnection duration If the median number is not greater than the first duration threshold (e.g., 2s), then the median number is determined as the first number (e.g., m=1).

[0117] If the disconnection duration If the median value is greater than the first duration threshold (e.g., 2s) but not greater than the second duration threshold (e.g., 10s), the number of median points is determined as the second number (e.g., m=3), where the second number is greater than the first number.

[0118] If the disconnection duration If the duration exceeds the second threshold (e.g., 10s), the number of median points is determined as the third number (e.g., m=5), where the third number is greater than the second number.

[0119] Furthermore, in the first moment (i.e., the anchoring time before the line breaks) and the second time. Using (the anchoring time after line restoration) as the endpoint, and based on the determined number of median points m, the time interval of the line break is divided. It is evenly divided into m+1 subintervals.

[0120] Step S43: Determine the weight values ​​for each of the at least one disconnection recovery time using a weight allocation rule based on the principle of symmetric attenuation.

[0121] In this embodiment, the weight allocation rule of the aforementioned symmetrical attenuation principle is to determine the number of median points m, and design the weight vector w as a distribution that is symmetrical about the center and decreases towards both ends, so that the median point in the middle has the highest weight and the median point at both ends has the lowest weight. For example, if m=3, then the weight values ​​of the three line break recovery times are 0.2, 0.6, and 0.2 respectively; if m=5, then the weight values ​​of the five line break recovery times are 0.1, 0.2, 0.4, 0.2, and 0.1 respectively.

[0122] This application's embodiments achieve precise matching of connection state density under different disconnection scenarios by dynamically and adaptively adjusting the number of median points based on the disconnection duration. This avoids computational redundancy in short disconnection scenarios and ensures the smoothness of state derivation in long disconnection scenarios. Simultaneously, a weight allocation mechanism based on the principle of symmetrical attenuation is introduced, assigning higher fusion weights to intermediate states. This effectively simulates the physical laws governing the natural transition of call states over time, significantly improving the continuity and restoration accuracy of the conversation states before and after disconnection. This greatly reduces audio and video stuttering, out-of-order playback, and echo phenomena in the initial recovery phase, enhancing the user's perceived seamless experience.

[0123] As an optional solution, the call system can be configured as follows: Figure 4 The process shown determines the third compressed session state feature vector at each disconnection recovery time based on the first compressed session state feature vector and the second compressed session state feature vector, including:

[0124] Step S51: Determine the initial compressed session state feature vector at each disconnection recovery time based on the first compressed session state feature vector, the second compressed session state feature vector, and the number of median points.

[0125] Specifically, the above scheme uses the first compressed session state feature vector and the second compressed session state feature vector as the computational objects, and uses different linear interpolation or piecewise interpolation algorithms to calculate the intermediate state according to the determined number of median points m.

[0126] For example: if the number of median points is 1, then the initial compressed session state feature vector at this disconnection recovery time... In the formula, This represents the feature vector of the first compressed session state. This represents the feature vector of the second compressed session state. If the number of median points is 3, then the initial compressed session state feature vectors at the three disconnection recovery times are as follows: , , .

[0127] Step S52: For each disconnection recovery time, determine whether the initial compressed session state feature vector at the disconnection recovery time exceeds the preset compressed session state feature space. Specifically, if the initial compressed session state feature vector at the disconnection recovery time exceeds the preset compressed session state feature space, proceed to step S53; if the initial compressed session state feature vector at the disconnection recovery time does not exceed the compressed session state feature space, proceed to step S54.

[0128] Step S53: Based on the spatial boundary of the compressed session state feature space, the initial compressed session state feature vector at the time of disconnection recovery is corrected to obtain the third compressed session state feature vector at the time of disconnection recovery.

[0129] Step S54: Use the initial compressed session state feature vector as the third compressed session state feature vector at the time of disconnection recovery.

[0130] Steps S52-S53 above can be understood as follows: the multidimensional session state parameters are compressed and mapped to a fixed, finite compressed session state feature space. In other words, any valid compressed session state feature vector must strictly fall within this interval. However, the initial compressed session state feature vectors calculated using linear interpolation or piecewise interpolation in the preceding steps may mathematically exceed the preset valid range due to floating-point operation errors, boundary conditions, or extreme network fluctuations. Therefore, if the call system detects these initial compressed session state feature vectors... After exceeding the boundary, it can be corrected based on the spatial boundary values ​​of the compressed session state feature space (i.e., the minimum value 0 and the maximum value K-1):

[0131] If the calculated value is less than 0, then correct it to 0;

[0132] If the calculated value is greater than K-1, then correct it to K-1;

[0133] If the calculated value is between [0, K-1], it remains unchanged.

[0134] Through the above correction process, the third compressed session state feature vector at the time of disconnection recovery can be obtained.

[0135] This application's embodiments effectively solve the problem of discrete state values ​​going out of bounds due to linear interpolation or state transition operations by introducing an out-of-bounds correction mechanism. This ensures that all derived compressed session state feature vectors corresponding to disconnection recovery times are strictly within the legal range, significantly improving the robustness of the state connection algorithm and the stability of system operation.

[0136] Furthermore, the target center feature vector corresponding to each third compressed session state feature vector is determined from the preset mapping table.

[0137] The mapping table can be constructed through the following steps:

[0138] Step 1: Randomly combine the dimensional values ​​of multiple spatial dimensions of the preset compressed session state feature space to obtain multiple compressed session state feature vectors.

[0139] The above steps can be understood as follows: by traversing all possible values ​​of the compressed session state feature space (i.e., randomly combining or fully enumerating the dimensional values ​​of multiple spatial dimensions), multiple compressed session state feature vectors representing the discrete compressed session state feature space are generated to form a sample vector set.

[0140] Step 2: For each compressed session state feature vector, determine multiple historical session state feature vectors mapped to the compressed session state feature vector from the preset database, and perform clustering processing on the multiple historical session state feature vectors to obtain the corresponding central feature vector.

[0141] The above steps can be understood as follows: Due to information loss during the compression process, a discrete compressed state may correspond to multiple different original high-dimensional feature vectors. To find the most representative restored value, historical data accumulated during the offline training phase is used, specifically, multiple historical session state feature vectors mapped to compressed session state feature vectors are determined from a pre-defined database. This database stores multiple historical session state feature vectors and their corresponding compressed session state feature vectors. Next, these historical session state feature vectors belonging to the same compressed session state feature vector are clustered. The purpose of clustering is to find the "centroid" or "typical representative" of this data set in the feature space, i.e., the central feature vector. This central feature vector best represents the common features of all original session state feature vectors covered by the compressed session state feature vector, thus serving as the optimal restored value for the compressed session state feature vector.

[0142] Step 3: Summarize the central feature vectors corresponding to the multiple compressed session state feature vectors to obtain a mapping table.

[0143] The above steps can be understood as follows: summarizing the "central feature vector" calculated for each discrete compressed state and establishing a one-to-one correspondence (e.g., compressed session state feature vectors of 0 -> central feature vector A, compressed session state feature vectors of 1 -> central feature vector B). This summarized structure is the mapping table (or inverse mapping lookup table).

[0144] Therefore, during the online recovery phase, the call system only needs to rely on the derived third compressed session state feature vector. The corresponding target center feature vector can be obtained by directly looking up the table. .

[0145] Furthermore, the determined target center feature vector is then analyzed based on the weight values. Weighted fusion is performed to obtain the target feature vector, i.e., the target feature vector. In the formula This represents the third compressed session state feature vector corresponding to the i-th disconnection recovery time. The weight values ​​are then calculated. Next, the target feature vector is inversely normalized (i.e., the inverse normalization operation) to restore the original multidimensional session state parameters, thus obtaining the recovered multidimensional session state parameters. .

[0146] Finally, the call system can control the telephone softswitch platform based on the recovery multidimensional session state parameters. To restore continuity to the target session. Specifically:

[0147] (1) Session association and variable setting:

[0148] The calling system sends the uuid_setvar command to the telephone softswitch platform via ESL to restore the pushed multidimensional session state parameters. This is written to the session variables corresponding to the UUID, specifically including: the RTP packet sequence number of the sender, the RTP packet sequence number of the receiver, the RTP timestamp, the media encoding type, and the jitter buffer size (e.g., ...). )wait.

[0149] (2) Media channel reset:

[0150] After the parameter settings are completed, a `uuid_media_reset` command is sent to the telephone softswitch platform. This command forcibly restarts the media channel for the call's UUID, making the aforementioned variable settings take effect immediately and triggering a new SDP (Session Description Protocol) Offer / Answer negotiation process. Next, a message containing... If the SDP Offer with the same parameters is received, and the SDP Answer from the peer device matches the negotiated parameters, the media channel establishes a connection according to the restored multidimensional session state parameters. If the SDP Answer from the peer device does not match the negotiated parameters, the optimal media encoding type supported by both parties is automatically selected (preferably the one with the most compatible parameters). Compatible media encoding types.

[0151] (3) RTP stream connection configuration:

[0152] The system invokes an internal interface of the telephone softswitch platform or executes a specific command to configure the starting receiver point for the RTP stream. The system sets the starting sequence number of the RTP receiver for resuming the session to... The start timestamp is set to This step ensures that newly received RTP packets can be correctly sorted and decoded by the telephone softswitch platform, achieving a seamless and smooth transition with the media stream before the disconnection, and avoiding audio out-of-order, packet loss, or stuttering.

[0153] In addition, after the session is restored, in order to objectively assess the quality of the restored session, the call system can perform verification through the following steps:

[0154] First, reacquire new session state monitoring metrics for the target session. These metrics include at least the cumulative duration of no received Real-Time Transport Protocol (RTP) data packets (reflecting network outages), and the round-trip transmission delay (reflecting network connectivity) and packet loss rate (reflecting network stability) for sending network probe messages to the network peer.

[0155] If the cumulative duration is less than the preset duration threshold (e.g., 100ms), the round-trip transmission delay is less than the preset delay threshold, and the packet loss rate is less than the preset packet loss rate threshold (e.g., 5%s), the multidimensional session state parameters are determined to have passed the verification.

[0156] If the cumulative duration is not less than the preset duration threshold, or the round-trip transmission delay is not less than the preset delay threshold, or the packet loss rate is not less than the preset packet loss rate threshold, it is determined that the recovery of the multidimensional session state parameters has failed the verification.

[0157] It should be noted that if the verification fails, a secondary derivation process is triggered, which involves re-acquiring multiple second compressed session state feature vectors after the second time step and repeating the above connection state derivation and recovery steps. If the recovery multidimensional session state parameters obtained from the secondary derivation still fail the verification, the user is given a corresponding prompt message, such as "It is recommended to call back."

[0158] In the above embodiments of this application, a lightweight session recovery mechanism based on "dynamic programming state compression" and "median point dynamic connection" is constructed. By compression mapping and preset session correlation coefficients, high-dimensional multi-dimensional session state parameters are compressed into low-dimensional compressed session state feature vectors. The number of median points and Gaussian distribution weights are determined adaptively by disconnection duration. The disconnection connection state is accurately derived without the need for full caching of RTP streams, which significantly reduces system memory overhead and improves concurrent processing capabilities. At the same time, through multi-dimensional disconnection detection and "recovery-verification" mechanism, fast and seamless recovery and high recovery success rate are achieved, solving the pain points of poor user experience, high resource consumption and insufficient dynamic adaptability in traditional solutions.

[0159] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0160] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / random access memory (RAM), magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0161] According to another aspect of the embodiments of this application, a session recovery apparatus for implementing the session recovery method after call disconnection provided in the above embodiments is also provided, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0162] Figure 5 This is a structural block diagram of an optional call recovery device according to an embodiment of this application, such as... Figure 5As shown, the session recovery device after a call is disconnected includes:

[0163] The acquisition module 52 is used to acquire the first compressed session state feature vector of the target session before the first moment and the second compressed session state feature vector after the second moment in the telephone softswitch platform, wherein the first moment is the moment when the target session disconnects and the second moment is the moment when the target session resumes.

[0164] The first determining module 54 is used to determine at least one disconnection recovery time and its corresponding weight value based on the first time and the second time, and to determine the third compressed session state feature vector for each disconnection recovery time based on the first compressed session state feature vector and the second compressed session state feature vector.

[0165] The second determining module 56 is used to determine the target center feature vector corresponding to each third compressed session state feature vector from the preset mapping table, fuse each target center feature vector according to the weight value to obtain the target feature vector, and perform inverse normalization processing on the target feature vector to obtain the recovered multidimensional session state parameters. The mapping table stores the center feature vectors corresponding to each of the multiple compressed session state feature vectors.

[0166] The recovery module 58 is used to control the telephone softswitch platform to continuously recover the target session based on the recovery multidimensional session state parameters.

[0167] In an exemplary embodiment, the above-described apparatus is further configured to obtain a first compressed session state feature vector of a target session in a telephone softswitch platform prior to a first moment by means of the following method: collecting and obtaining session state monitoring indicators of the target session at preset time intervals; at each moment, determining whether the target session has experienced a call disconnection at the current moment based on the session state monitoring indicators at the current moment; and if the target session experiences a call disconnection at the first moment, obtaining the first compressed session state feature vector of the target session prior to the first moment from a preset memory queue, wherein the memory queue retains compressed session state feature vectors corresponding to multiple moments when the target session has not experienced a call disconnection.

[0168] In an exemplary embodiment, the session state monitoring indicators include at least: the cumulative duration of no real-time transport protocol data packets received before the corresponding time, the round-trip transmission delay and packet loss rate of sending network probe messages to the network peer at the corresponding time, the session state variables at the corresponding time, and the control signaling events reported by the telephone softswitch platform at the corresponding time; the above-mentioned apparatus is further configured to determine whether the target session has experienced a call disconnection at the current time based on the session state monitoring indicators by the following methods: triggering a media stream interruption alarm when the cumulative duration is not less than a preset duration threshold; triggering a network link anomaly alarm when the round-trip transmission delay is not less than a preset delay threshold or the packet loss rate is not less than a preset packet loss rate threshold; triggering a session abnormal interruption alarm when the session state variables and control signaling events are inconsistent; determining that the target session has experienced a call disconnection when the target session triggers at least two of the media stream interruption alarm, network link anomaly alarm, and session abnormal interruption alarm simultaneously; and determining that the target session has not experienced a call disconnection when the target session triggers only any one of the media stream interruption alarm, network link anomaly alarm, and session abnormal interruption alarm.

[0169] In an exemplary embodiment, before acquiring the first compressed session state feature vector of the target session before the first moment, the above-mentioned apparatus is further configured to control the telephone softswitch platform to collect multi-dimensional session state parameters of the target session at preset time intervals before the first moment. The multi-dimensional session state parameters include at least: media transmission sequence parameters, media encoding feature parameters, network link quality parameters, and call control state parameters, and the call control state parameters include at least: session universal unique identifier and initiation time. For the multi-dimensional session state parameters at each moment, the multi-dimensional session state parameters are preprocessed, and the processed multi-dimensional session state parameters are combined based on a preset session correlation coefficient to obtain a session state feature vector. The session state feature vector is then processed using a pre-trained mapping function to obtain the corresponding compressed session state feature vector. The preprocessing includes: abnormal data detection, abnormal data correction, and standardization processing. The compressed session state feature vectors are stored sequentially in a memory queue according to the time sequence, and when the storage capacity of the memory queue reaches a preset capacity threshold, the earliest compressed session state feature vector in the memory queue is removed according to the first-in-first-out principle.

[0170] In an exemplary embodiment, the above-described apparatus is further configured to map the session state feature vector using a pre-trained mapping function to obtain a corresponding compressed session state feature vector, including: linearly projecting the session state feature vector using a pre-trained projection matrix to obtain a one-dimensional session state feature sequence, wherein the projection matrix is ​​trained based on historical session data from multiple historical sessions, and during the training process, the projection matrix is ​​iteratively updated using a gradient descent algorithm with the objective function of minimizing storage overhead and session recovery error as the optimization objective until a preset convergence condition is met, wherein the historical session data includes at least: historical compressed session state feature vectors, recovery result identifiers, and recovered call quality indicators; equidistantly quantizing the one-dimensional session state feature sequence according to a preset spatial dimension of the compressed session state feature space to obtain multiple discrete session state values ​​equal to the spatial dimension, and forming a compressed session state feature vector from the multiple discrete session state values, wherein the spatial dimension of the compressed session state feature space is obtained by analyzing the value range of the session state feature vector of the telephone softswitch platform using the elbow rule.

[0171] In an exemplary embodiment, the above-described apparatus is further configured to obtain a second compressed session state feature vector of a target session in a telephone softswitch platform after a second time point by means of the following method: obtaining new multidimensional session state parameters of a new session in the telephone softswitch platform after a second time point, wherein the new session is a session re-initiated after the second time point, and the new multidimensional session state parameters include at least: media transmission sequence parameters, media encoding feature parameters, network link quality parameters, and call control state parameters, and the call control state parameters include at least: a session universally unique identifier and an initiation time; if the session universally unique identifier in the new multidimensional session state parameters is consistent with the session universally unique identifier of the target session, and the time interval between the initiation time of the new multidimensional session state parameters and the first time point is less than a preset duration threshold, determining that the new session and the target session are associated sessions; preprocessing the new multidimensional session state parameters, and combining the processed new multidimensional session state parameters based on a preset session association coefficient to obtain a second session state feature vector, and processing the second session state feature vector using a pre-trained mapping function to obtain a second compressed session state feature vector, wherein the preprocessing includes: abnormal data detection, abnormal data correction, and standardization processing.

[0172] In an exemplary embodiment, the above-described apparatus is further configured to determine at least one disconnection recovery time and its corresponding weight value based on a first time and a second time by the following method: determining the difference between the second time and the first time to obtain the disconnection duration; determining the number of median points based on the disconnection duration and a preset median point granularity adaptive strategy, and dividing the disconnection duration based on the number of median points to obtain at least one disconnection recovery time; and determining the weight value of each of the at least one disconnection recovery time based on a weight allocation rule based on the principle of symmetric attenuation.

[0173] In an exemplary embodiment, the above-described apparatus is further configured to determine a third compressed session state feature vector at each disconnection recovery time by means of a first compressed session state feature vector and a second compressed session state feature vector, including: determining an initial compressed session state feature vector at each disconnection recovery time based on the first compressed session state feature vector, the second compressed session state feature vector, and the number of median points; for each disconnection recovery time, determining whether the initial compressed session state feature vector at the disconnection recovery time exceeds a preset compressed session state feature space; if the initial compressed session state feature vector at the disconnection recovery time exceeds the preset compressed session state feature space, correcting the initial compressed session state feature vector at the disconnection recovery time according to the spatial boundary of the compressed session state feature space to obtain a third compressed session state feature vector at the disconnection recovery time; if the initial compressed session state feature vector at the disconnection recovery time does not exceed the compressed session state feature space, using the initial compressed session state feature vector as the third compressed session state feature vector at the disconnection recovery time.

[0174] In an exemplary embodiment, the above-described apparatus is further configured to construct a mapping table by means of the following method: randomly combining the dimensional values ​​of multiple spatial dimensions of a preset compressed session state feature space to obtain multiple compressed session state feature vectors; for each compressed session state feature vector, determining multiple historical session state feature vectors mapped to the compressed session state feature vector from a preset database, performing clustering processing on the multiple historical session state feature vectors to obtain corresponding central feature vectors, wherein the preset database stores multiple historical session state feature vectors and compressed session state feature vectors corresponding to each historical session state feature vector; and summarizing the central feature vectors corresponding to each of the multiple compressed session state feature vectors to obtain a mapping table.

[0175] In an exemplary embodiment, after restoring the target session based on the restored multidimensional session state parameters, the apparatus is further configured to reacquire new session state monitoring indicators for the target session. These session state monitoring indicators include at least: the cumulative duration of no received Real-Time Transport Protocol (RTP) data packets, and the round-trip transmission delay and packet loss rate for sending network probe messages to the network peer. If the cumulative duration is less than a preset duration threshold, the round-trip transmission delay is less than a preset delay threshold, and the packet loss rate is less than a preset packet loss rate threshold, the restored multidimensional session state parameters are determined to have passed verification. If the cumulative duration is not less than the preset duration threshold, or the round-trip transmission delay is not less than the preset delay threshold, or the packet loss rate is not less than the preset packet loss rate threshold, the restored multidimensional session state parameters are determined to have failed verification.

[0176] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0177] According to another aspect of the embodiments of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein the program executes the steps in any of the above method embodiments when it is run.

[0178] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, ROMs, RAMs, portable hard drives, magnetic disks, or optical disks.

[0179] According to another aspect of the embodiments of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor is configured to perform the steps of any of the method embodiments described above via the computer program. In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0180] Specific examples in this embodiment can be found in the examples described in the above embodiments and exemplary implementations, and will not be repeated here.

[0181] According to another aspect of the embodiments of this application, a computer program product is also provided, the computer program product including a computer program / instructions containing program code for performing the method shown in the flowchart.

[0182] Specifically, the computer program executes the following steps during runtime: It acquires a first compressed session state feature vector of the target session before a first moment and a second compressed session state feature vector after a second moment from the telephone softswitch platform, where the first moment is the moment the target session experiences a call disconnection, and the second moment is the moment the target session resumes operation after the call disconnection; it determines at least one disconnection recovery moment and its corresponding weight value based on the first and second moments, and determines a third compressed session state feature vector for each disconnection recovery moment based on the first and second compressed session state feature vectors; it determines the target center feature vector corresponding to each third compressed session state feature vector from a preset mapping table, fuses the target center feature vectors according to the weight values ​​to obtain a target feature vector, and performs inverse normalization on the target feature vector to obtain recovery multidimensional session state parameters, where the mapping table stores the center feature vectors corresponding to each of the multiple compressed session state feature vectors; and it controls the telephone softswitch platform to continuously restore the target session based on the recovery multidimensional session state parameters.

[0183] As an alternative implementation, the above-mentioned electronic device may exist in the form of a mobile terminal, a computer terminal, or a similar computing device. Figure 6 A hardware block diagram of a computer terminal for implementing a session resumption method after a call is dropped is shown. Figure 6 As shown, the computer terminal 60 may include a processor 602, a memory 604 for storing data, and a transmission device 606 for communication functions. The processor 602 may include one or more processors. For ease of explanation, Figure 6 The diagram illustrates multiple processors, including a first processor 602a, a second processor 602b, ..., an nth processor 602n. Processor 602 may be, but is not limited to, a processing device such as a microprocessor (MCU) or a field-programmable gate array (FPGA).

[0184] In addition, the computer terminal 60 may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 6 The structure shown is for illustrative purposes only and does not limit the structure of the computer terminal described above. For example, computer terminal 60 may also include... Figure 6 The more or fewer components shown, or having the same Figure 6 The different configurations shown.

[0185] Processor 602 may include one or more processing units. For example, processor 602 may include processing devices such as central processing unit (CPU), graphics processing unit (GPU), digital signal processing (DSP) chip, microcontroller unit (MCU), field-programmable gate array (FPGA), neural network processing unit (NPU), tensor processing unit (TPU), artificial intelligence (AI) type processor, etc. Different processing units may be independent components or integrated into one or more processors.

[0186] It should be noted that the aforementioned one or more processors 602 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 60. As described in the embodiments of this application, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0187] The memory 604 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the call recovery method after disconnection in this embodiment. The processor 602 executes various functional applications and data processing by running the software programs and modules stored in the memory 604, thereby realizing the aforementioned call recovery method after disconnection. The memory 604 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 604 may further include memory remotely located relative to the processor 602, and these remote memories can be connected to the computer terminal 60 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0188] The transmission device 606 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 60. In one example, the transmission device 606 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 606 may be a Radio Frequency (RF) module, used for data interaction with external networks or devices via wired and / or wireless network connections.

[0189] The display can be, for example, a liquid crystal display (LCD) or a touchscreen display (also known as a "touchscreen" or "touch display screen"). This LCD allows the user to interact with the user interface of the mobile terminal. Furthermore, the computer terminal 60 can be connected to a cursor control device and a keyboard via an input / output interface. The cursor control device (e.g., a mouse, trackball, or touchpad) is primarily used to transmit directional information and command selections to the terminal and control the movement of the cursor on the display; the keyboard is primarily used to receive letters, numbers, and other control commands input by the user, thereby enabling interaction between the user and the computer terminal 60.

[0190] It should be noted here that, in some optional embodiments, the above... Figure 6 The computer terminal shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 6 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0191] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0192] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for resuming a call after it has been disconnected, characterized in that, include: Obtain the first compressed session state feature vector of the target session before the first moment and the second compressed session state feature vector after the second moment in the telephone softswitch platform, wherein the first moment is the moment when the target session drops the call and the second moment is the moment when the target session resumes the call. At least one disconnection recovery time and its corresponding weight value are determined based on the first time and the second time, and a third compressed session state feature vector is determined for each disconnection recovery time based on the first compressed session state feature vector and the second compressed session state feature vector. The target center feature vector corresponding to each of the third compressed session state feature vectors is determined from the preset mapping table. The target center feature vectors are fused according to the weight value to obtain the target feature vector. The target feature vector is then denormalized to obtain the recovered multidimensional session state parameters. The mapping table stores the center feature vectors corresponding to each of the multiple compressed session state feature vectors. The telephone softswitch platform is controlled to continuously restore the target session based on the restored multidimensional session state parameters.

2. The method according to claim 1, characterized in that, Obtain the first compressed session state feature vector of the target session in the telephone softswitch platform before the first moment, including: Session status monitoring indicators of the target session are collected at preset time intervals; At each moment, the target session is determined to have experienced a call disconnection based on the session status monitoring indicators at the current moment. In the event that the target session is disconnected at the first moment, the first compressed session state feature vector of the target session at the moment preceding the first moment is obtained from a preset memory queue, wherein the memory queue retains compressed session state feature vectors corresponding to multiple moments when the target session is not disconnected.

3. The method according to claim 2, characterized in that, The session state monitoring indicators include at least: the cumulative duration of not receiving real-time transmission protocol data packets before the corresponding time, the round-trip transmission delay and packet loss rate of sending network probe messages to the network peer at the corresponding time, the session state variables at the corresponding time, and the control signaling events reported by the telephone softswitch platform at the corresponding time. Determining whether the target session has experienced a call drop at the current moment based on the session state monitoring metrics includes: If the cumulative duration is not less than a preset duration threshold, a media stream interruption alarm is triggered; If the round-trip transmission delay is not less than a preset delay threshold or the packet loss rate is not less than a preset packet loss rate threshold, a network link anomaly alarm is triggered. If the session state variable is inconsistent with the control signaling event, a session abnormal interruption alarm is triggered; If at least two of the following are triggered simultaneously in the target session: the media stream interruption alarm, the network link anomaly alarm, and the session anomaly interruption alarm, it is determined that the target session has experienced a call disconnection. If the target session triggers only one of the media stream interruption alarm, the network link abnormality alarm, and the session abnormality interruption alarm, it is determined that the target session has not experienced a call disconnection.

4. The method according to claim 2, characterized in that, Before obtaining the first compressed session state feature vector of the target session prior to the first moment, the method further includes: Before the first moment, the telephone softswitch platform is controlled to collect multi-dimensional session state parameters of the target session at preset time intervals. The multi-dimensional session state parameters include at least: media transmission sequence parameters, media encoding feature parameters, network link quality parameters, and call control state parameters. The call control state parameters include at least: session universal unique identifier and initiation time. For the multidimensional session state parameters at each time step, the multidimensional session state parameters are preprocessed. Based on the preset session correlation coefficient, the processed multidimensional session state parameters are combined to obtain a session state feature vector. The session state feature vector is then processed using a pre-trained mapping function to obtain a corresponding compressed session state feature vector. The preprocessing includes: abnormal data detection, abnormal data correction, and standardization. The compressed session state feature vectors are stored sequentially into a memory queue according to their time sequence. When the storage capacity of the memory queue reaches a preset capacity threshold, the earliest compressed session state feature vector in the memory queue is removed according to the first-in-first-out principle.

5. The method according to claim 4, characterized in that, The session state feature vector is mapped using a pre-trained mapping function to obtain the corresponding compressed session state feature vector, including: The session state feature vector is linearly projected using a pre-trained projection matrix to obtain a one-dimensional session state feature sequence. The projection matrix is ​​trained based on historical session data from multiple historical sessions. During training, the objective function of minimizing storage overhead and session recovery error is used as the optimization objective. The gradient descent algorithm is used to iteratively update the projection matrix until a preset convergence condition is met. The historical session data includes at least: historical compressed session state feature vectors, recovery result identifiers, and call quality indicators after recovery. The one-dimensional session state feature sequence is quantized at equal intervals according to the preset spatial dimension of the compressed session state feature space to obtain multiple discrete session state values ​​with the same spatial dimension. The multiple discrete session state values ​​are then used to form the compressed session state feature vector. The spatial dimension of the compressed session state feature space is obtained by analyzing the value range of the session state feature vector of the telephone softswitch platform using the elbow rule.

6. The method according to claim 1, characterized in that, Obtain the second compressed session state feature vector of the target session in the telephone softswitch platform after the second time point, including: Obtain new multidimensional session state parameters of the new session in the telephone softswitch platform after the second time point, wherein the new session is a session re-initiated after the second time point, and the new multidimensional session state parameters include at least: media transmission sequence parameters, media encoding feature parameters, network link quality parameters, and call control state parameters, and the call control state parameters include at least: session universal unique identifier and initiation time; If the session universal unique identifier in the new multidimensional session state parameter is consistent with the session universal unique identifier of the target session, and the time interval between the initiation time of the new multidimensional session state parameter and the first time is less than a preset duration threshold, then the new session and the target session are determined to be associated sessions. The new multidimensional session state parameters are preprocessed, and the processed new multidimensional session state parameters are combined based on a preset session correlation coefficient to obtain a second session state feature vector. The second session state feature vector is then processed using a pre-trained mapping function to obtain a second compressed session state feature vector. The preprocessing includes: abnormal data detection, abnormal data correction, and standardization processing.

7. The method according to claim 1, characterized in that, Determining at least one disconnection recovery time and its corresponding weight value based on the first time and the second time includes: The difference between the second time point and the first time point is determined to obtain the duration of the disconnection. Based on the disconnection duration and a preset median point granularity adaptive strategy, the number of median points is determined, and the disconnection duration is divided according to the number of median points to obtain the at least one disconnection recovery time. The weighting rule based on the principle of symmetric attenuation determines the respective weight values ​​of the at least one disconnection recovery time.

8. The method according to claim 7, characterized in that, Based on the first compressed session state feature vector and the second compressed session state feature vector, a third compressed session state feature vector is determined for each of the disconnection recovery times, including: Based on the first compressed session state feature vector, the second compressed session state feature vector, and the number of median points, the initial compressed session state feature vector for each disconnection recovery time is determined. For each disconnection recovery time, determine whether the initial compressed session state feature vector at the disconnection recovery time exceeds the preset compressed session state feature space; If the initial compressed session state feature vector at the time of connection recovery exceeds the preset compressed session state feature space, the initial compressed session state feature vector at the time of connection recovery is corrected according to the spatial boundary of the compressed session state feature space to obtain the third compressed session state feature vector at the time of connection recovery. If the initial compressed session state feature vector at the time of disconnection recovery does not exceed the compressed session state feature space, the initial compressed session state feature vector is used as the third compressed session state feature vector at the time of disconnection recovery.

9. The method according to claim 1, characterized in that, The mapping table is constructed in the following ways: The dimension values ​​of multiple spatial dimensions of the preset compressed session state feature space are randomly combined to obtain multiple compressed session state feature vectors. For each compressed session state feature vector, multiple historical session state feature vectors mapped to the compressed session state feature vector are determined from a preset database. The multiple historical session state feature vectors are clustered to obtain the corresponding central feature vector. The preset database stores multiple historical session state feature vectors and the compressed session state feature vector corresponding to each historical session state feature vector. The mapping table is obtained by summing the central feature vectors corresponding to the multiple compressed session state feature vectors.

10. The method according to claim 1, characterized in that, After restoring the target session based on the restored multidimensional session state parameters, the method further includes: Reacquire new session state monitoring metrics for the target session, wherein the session state monitoring metrics include at least: the cumulative duration of not receiving real-time transport protocol data packets, and the round-trip transmission delay and packet loss rate for sending network probe messages to the network peer; If the cumulative duration is less than a preset duration threshold, the round-trip transmission delay is less than a preset delay threshold, and the packet loss rate is less than a preset packet loss rate threshold, then the recovery multidimensional session state parameters are determined to pass the verification. If the cumulative duration is not less than a preset duration threshold, or the round-trip transmission delay is not less than a preset delay threshold, or the packet loss rate is not less than a preset packet loss rate threshold, then the recovery multidimensional session state parameter is determined to have failed the verification.

11. A session recovery device after a call is dropped, characterized in that, include: The acquisition module is used to acquire the first compressed session state feature vector of the target session before the first moment and the second compressed session state feature vector after the second moment in the telephone softswitch platform, wherein the first moment is the moment when the target session drops the call and the second moment is the moment when the target session resumes the call. The first determining module is used to determine at least one disconnection recovery time and its corresponding weight value based on the first time and the second time, and to determine the third compressed session state feature vector for each disconnection recovery time based on the first compressed session state feature vector and the second compressed session state feature vector. The second determining module is used to determine the target center feature vector corresponding to each of the third compressed session state feature vectors from a preset mapping table, fuse each of the target center feature vectors according to the weight value to obtain a target feature vector, and perform inverse normalization processing on the target feature vector to obtain the recovered multidimensional session state parameters. The mapping table stores the center feature vectors corresponding to each of the multiple compressed session state feature vectors. The recovery module is used to control the telephone softswitch platform to continuously recover the target session based on the recovery multidimensional session state parameters.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the device containing the computer-readable storage medium executes the session recovery method after call disconnection as described in any one of claims 1 to 10 by running the computer program.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the session recovery method after call disconnection as described in any one of claims 1 to 10 through the computer program.

Images