Multi-source time-series data analysis processing system for intelligent meeting

Abstract

The application relates to the technical field of electric digital data processing, and discloses a multi-source time sequence data analysis processing system for intelligent meetings, which comprises the following steps: a data acquisition module acquires multi-path heterogeneous meeting data streams; a mapping and correlation module establishes time sequence correlation weights between logical anchor points, and constructs a multi-dimensional time sequence mapping space; an evolution trend analysis module calculates a step frequency and a phase bias base, and generates a logical evolution feature matrix; and an adaptive calibration module monitors data transmission states, generates virtual evolution feature quantities according to the logical evolution feature matrix when signal gaps are encountered, maintains the logical constraint strength of the multi-dimensional time sequence mapping space, adjusts the confidence weight according to the alignment feature cost parameters, and suppresses fault data. The application utilizes a feedback adjustment mechanism, effectively makes up for logical breaks caused by signal loss, realizes essential remodeling, and ensures the physical authenticity and logical stability of analysis results.

Description

Technical Field

[0001] This invention relates to a multi-source time-series data analysis and processing system for intelligent conferencing, belonging to the field of electronic digital data processing technology. Background Technology

[0002] Current intelligent conferencing systems involve the collaborative processing of heterogeneous time-series data such as audio streams, video streams, and control command streams. To achieve synchronous presentation of multi-source information, a timestamp alignment mechanism based on a global clock is adopted. The sending end encapsulates physical timestamps for various raw data packets, and the receiving end performs linear alignment of each signal according to the timestamp. This processing method can maintain basic consistency under stable channel conditions. However, due to the non-steady-state fluctuations of the network transmission environment, multiple data streams face differentiated jitter and packet loss during transmission. Under the condition of long gap loss caused by congestion, the physical alignment benchmark is broken. Although the missing components can be formally supplemented by increasing the buffer length or performing linear interpolation, due to the lack of modeling of the evolutionary inertia of business logic, this processing method cannot restore the implicit logical causal relationship between heterogeneous data streams, which is prone to producing false prosperity of semantic logic. When key interaction signals are misaligned on the logical axis, relying solely on the physical timestamp alignment method will produce instantaneous changes at the semantic level, resulting in damage to the continuity of the conference intent stream.

[0003] In addressing alignment deficiencies, some technologies focus on high-level semantic logical connections. For example, Chinese invention patent CN111782800B discloses an intelligent meeting analysis method for event tracing. This method automatically associates meeting content and traces tasks by constructing a knowledge graph. This solution anchors the static association of high-level semantic information such as meeting text and speech transcription records, which is a post-hoc logical processing method. It is difficult to cope with the dynamic transient changes of underlying physical signals in real-time interactive scenarios. When faced with sudden long-interval signal loss, due to the lack of real-time modeling of the data flow logic step rhythm and phase evolution inertia, it cannot provide virtual support with logical self-sustaining force during physical signal interruption. This causes a semantic leap when the signal recovers. Existing technologies cannot quantify and evaluate the degree of intrusion of calibration actions on the original sampling structure, and it is difficult to adaptively suppress fault data caused by hardware detuning, thus limiting the fidelity and stability of multi-source time-series data processing under complex working conditions.

[0004] Therefore, the technical problem to be solved by this invention is how to construct a collaborative processing mechanism with logical self-sustaining ability and multi-center intention arbitration capability in the face of data discontinuity loss and high-concurrency intention entanglement, so as to realize the logical reconstruction of heterogeneous data streams in virtual space. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A multi-source time-series data analysis and processing system for intelligent conferencing, comprising:

[0006] The data acquisition module acquires multiple heterogeneous conference data streams;

[0007] The mapping and association module establishes temporal association weights between the logical anchor points of various conference data streams, and constructs a multi-dimensional temporal alignment space;

[0008] The evolution trend analysis module calculates the logical step frequency and phase offset base phase of each conference data stream on the time axis, and generates a logical evolution feature matrix that characterizes the evolution trend of the data stream.

[0009] The adaptive calibration module is configured to monitor the transmission status of each conference data stream. When a signal gap is detected in the target data stream, it maintains the temporal correlation weights corresponding to the target data stream. Based on the logical evolution feature matrix, it generates virtual evolution feature quantities corresponding to the signal gaps in the multi-dimensional temporal alignment space to maintain the logical constraint strength of the multi-dimensional temporal alignment space during signal loss. It calculates the cumulative logical displacement of each conference data stream in real time when performing feature alignment, generates alignment feature cost parameters that characterize the degree of intrusion of the calibration action on the original sampling structure, and compares the alignment feature cost parameters with a preset fidelity threshold. Based on the comparison results, it assigns confidence weights to each conference data stream and adjusts the correction contribution of the temporal correlation weights to the multi-dimensional temporal alignment space according to the confidence weights to suppress faulty data caused by source device failure or phase drift.

[0010] Preferably, the evolution trend analysis module analyzes the displacement residual distribution of the logical anchor point within a preset period window and uses the least squares regression algorithm to fit the evolution slope of the displacement residual on the time axis; the adaptive calibration module generates a frequency correction command based on the evolution slope and sends the frequency correction command to the data acquisition module to eliminate the linear cumulative clock deviation of the conference data stream by adjusting the original sampling frequency.

[0011] Preferably, the temporal correlation weights established by the mapping correlation module are used to define the causal coupling strength between different modal data streams. The evolution trend analysis module extracts historical temporal features and compensates for the instantaneous loss of temporal correlation weights by predicting the causal coupling strength during signal gaps.

[0012] Preferably, the adaptive calibration module quantifies the alignment feature cost parameter based on the adjustment amount of the calibration operation on the original sampled structure. Alignment feature cost parameters satisfy: ,in, For the alignment feature cost parameter, The total number of data streams participating in the alignment. For the first Preset weighting factors for each data stream. For the first The logical displacement deviation of the data stream within the current processing cycle.

[0013] Preferably, the adaptive calibration module performs hierarchical processing on each conference data stream according to the confidence weight. When the confidence weight is lower than the preset logical isolation threshold, the logical association between the conference data stream and the multidimensional time alignment space is automatically severed.

[0014] Preferably, the data acquisition module includes multiple distributed terminals, each of which is used to synchronously acquire audio signals, video signals, and collaborative document update instructions, and encapsulate the acquisition results into data packets with original timestamps and send them to the mapping and association module.

[0015] Preferably, when generating frequency correction instructions, the adaptive calibration module also calculates the network jitter parameters of the distributed terminals and injects the network jitter parameters as a regularization constraint into the generation logic of the frequency correction instructions to distinguish between hardware clock drift and logic shifts caused by network latency fluctuations.

[0016] Preferably, the multidimensional temporal alignment space displays the correlation status of multi-source data through a real-time updated topological mapping graph. Each node in the topological mapping graph represents a logical anchor point, and the edge weights between nodes are dynamically modulated by the logical evolution feature matrix.

[0017] Preferably, when calculating the logical evolution feature matrix, the evolution trend analysis module identifies the dominant feature mode of each conference data stream during the interaction process by comparing the feature evolution rate of each conference data stream, and dynamically improves the weight centrality of the dominant feature mode in the multi-dimensional temporal alignment space.

[0018] Preferably, the system also includes an instruction issuing module, which is connected to the adaptive calibration module. The instruction issuing module is used to feed back real-time synchronization parameters to each data source terminal based on the calibration results of the multi-dimensional time-series alignment space, so as to realize the logical coordination between the acquisition frequency of each terminal and the system processing frequency.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. In multi-source time-series data analysis and processing, the inertial computing module is used to extract the logical step slope of the data stream on the virtual time axis and generate a logical evolution tensor that represents the inherent rhythm of the business logic. During the period of missing input features, the logic unit maintains the tension distribution of the virtual anchor point in the virtual alignment space based on this tensor. The processing unit calculates the residual between the measured pose after the real signal is recovered and the logical displacement of the virtual anchor point, and uses this as a damping factor to dynamically adjust the convergence rate of the time-series scaling. This collaborative mechanism enables the system to have logical self-sustaining capability when facing data gap loss conditions. By guiding the scaling amount to smoothly regress to the real state, it eliminates the semantic step jump generated by the traditional linear prediction technology at the moment of signal recovery, and ensures the logical continuity in the back-end causal reconstruction process.

[0021] 2. The logical unit establishes mutually orthogonal sub-logical planes for data streams with different identities, realizing the physical isolation of multi-subject causal chains. The processing unit dynamically adjusts the temporal pull of each sub-plane on the corresponding data stream by calculating the entropy concentration weight of the topological association key in each sub-logical plane. This asymmetric weight allocation method based on entropy concentration enables the highly consistent main intent stream to obtain higher alignment stiffness and suppresses stray pull interference from the high-entropy noise plane. It solves the semantic entanglement bottleneck caused by multi-center concurrent interaction in ultra-large-scale collaborative scenarios from the underlying data processing procedure, and realizes accurate layering and automatic focusing of complex meeting intent streams without human intervention in speaking permissions.

[0022] 3. The processing unit calculates the cumulative logic displacement of each data stream during the elastic scaling process in real time, and generates corresponding topology elastic power consumption parameters. These parameters are used to quantify and evaluate the degree of intrusion of the calibration action on the original sampling structure. When the elastic power consumption generated by a specific source exceeds the preset fidelity threshold, the logic unit automatically reduces its corresponding topology association weight, enabling the system to have real-time perception of alignment costs. It can automatically identify and suppress faulty sources caused by hardware failure or non-steady phase drift based on calibration costs, avoid global logic collapse caused by forced correction, and ensure the physical reliability of the output analysis graph. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall data processing logic and closed-loop feedback process of the system of the present invention;

[0024] Figure 2 This is a diagram showing the internal parameter definitions and detailed interaction structure of the system functional modules of the present invention. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0026] This invention provides a multi-source time-series data analysis and processing system for intelligent conferencing, comprising a data acquisition module, a mapping and correlation module, an evolution trend analysis module, and an adaptive calibration module. The data acquisition module acquires multiple heterogeneous conferencing data streams and transmits the data packets to the mapping and correlation module. The mapping and correlation module establishes time-series correlation weights between logical anchor points, constructing a multi-dimensional time-series alignment space. The evolution trend analysis module calculates the logical step frequency and phase offset base phase of the data streams, generating a logical evolution feature matrix. The adaptive calibration module monitors the transmission status, maintains the logical constraint strength based on the logical evolution feature matrix during signal gaps, and adjusts the calibration according to the alignment feature cost parameters. To address the issue of maintaining semantic logic continuity under extreme congestion, the mapping and association module establishes temporal association weights between the logical anchor points of each conference data stream. Here, a logical anchor point is defined as a discrete time node with semantic features in the data stream, such as an energy transition point in the audio stream or a transaction trigger point in a collaborative instruction. For the identification of energy transition points in the audio stream, the system calculates the average power of the signal within a five-millisecond moving average window. When the average power value of the current window exceeds twice the average power of the background noise within the previous 100ms, and the duration of this state exceeds three sampling periods, the processor marks the starting point of this time as a logical anchor point to filter out false triggers caused by environmental white noise. The system generates temporal association weights that characterize logical consistency by calculating the causal coupling strength of different data streams at the anchor points, thereby mapping the heterogeneous data streams to a unified multi-dimensional temporal alignment space.

[0027] To address the risk of logical anchor point failure caused by long data gaps, the evolution trend analysis module executes an evolution law modeling procedure, extracting the step characteristics of each conference data stream on the virtual time axis in real time to generate a logical evolution feature matrix. When the system detects a signal gap in the target data stream, the adaptive calibration module generates a virtual evolution feature quantity corresponding to the signal gap in the multi-dimensional temporal alignment space based on the logical evolution feature matrix. This virtual feature quantity acts as a virtual anchor point during the physical signal loss, maintaining the logical constraint strength of the alignment space and preventing disordered drift of the data stream during signal interruption. At the moment of signal recovery, to avoid semantic step jumps caused by traditional linear compensation techniques, the adaptive calibration module executes a damped convergence procedure. The processing unit calculates the logical displacement residual between the measured pose of the recovery point and the virtual evolution feature quantity, and uses it as the control parameter for adjusting the step size to guide the temporal stretching quantity to smoothly regress to the real state. The specific damping adjustment logic is as follows: divide the calculated logical displacement residual value by ten to obtain the single-step compensation increment. In the subsequent ten consecutive sampling cycles, each processing cycle moves the current alignment position towards the real position by the length of the single-step compensation increment. Through this step-by-step method of amortizing sudden deviations to multiple subsequent cycles, the stretching or compression rate of the time axis is kept within the variation range of 3ms per sampling cycle, ensuring the logical continuity of the conference intent flow in the overall dimension under discontinuous packet loss conditions.

[0028] The adaptive calibration module is also used to quantitatively evaluate the degree of intrusion of calibration actions on the original data structure. The system calculates the cumulative logical displacement of each conference data stream in real time during feature alignment and generates alignment feature cost parameters. Alignment feature cost parameters Satisfying the formula: ,in, Alignment feature cost parameters; The total number of data streams participating in the alignment; For the first The preset weighting factor of the data stream reflects the service priority of the information source; For the first The logical shift deviation of the data stream within the current processing cycle, and the system comparison parameters. Based on a preset fidelity threshold, a confidence weight is assigned to each conference data stream, and the contribution is adjusted accordingly. When the confidence weight falls below a preset logical isolation threshold, the system automatically disconnects the data stream from the alignment space, thereby suppressing faulty data caused by source device failure or linearly accumulated clock skew. Addressing the challenge of multi-gravitational-center competition in ultra-large-scale collaborative scenarios, the system executes an intent arbitration procedure. The mapping and association module establishes mutually orthogonal sub-logical planes for data streams with different identities, achieving isolation of the main causal chain. The adaptive calibration module calculates the entropy concentration of the topological association keys within each sub-logical plane. The lower the entropy value, the more meaningful the plane is. Figure 1 The higher the consistency, the more the system depends on the entropy concentration. The timing pull weight is dynamically adjusted to achieve higher alignment stiffness based on the highly consistent main intent flow and to suppress stray interference from the high-noise plane. The adaptive calibration module statistically analyzes the trigger frequency of the association keys within the sub-logic bit plane during the sampling period, calculates the normalized probability distribution of the association keys, and solves the entropy concentration of the sub-logic bit plane according to the normalized probability distribution. According to the mapping relationship , entropy concentration Converted to time-series pull weights, where As time-series pull weight, Entropy concentration, To preset the sensitivity damping coefficient, adjust the high-entropy noise suppression ratio; align the feature cost parameters. Satisfy the formula ,in For the alignment feature cost parameter, The total number of data streams, For the first Road data stream weighting factor For the first The logical displacement deviation of the data stream within the current processing cycle is determined by selecting a synchronous reference audio stream as the calibration source. The cumulative logical displacement sequence statistical variance of the reference audio stream within the observation window is calculated, and the statistical variance is calibrated to 1.2 times as the global fidelity threshold, which serves as a quantitative indicator for determining the confidence of the conference data stream.

[0029] The evolution trend analysis module also integrates a clock self-calibration function. This module analyzes the displacement residual distribution of the logic anchor point within a preset period window, uses a least squares regression algorithm to fit the evolution slope of the residual on the time axis, and the adaptive calibration module generates a frequency correction command based on this slope and sends it to the corresponding data source terminal. After receiving the command, the terminal adjusts the original sampling frequency to eliminate the linear cumulative clock deviation caused by the aging of the terminal crystal oscillator, reducing the computational overhead of the system when maintaining a long-period alignment state. The evolution trend analysis module selects a 20ms sliding window to sample the conference data stream and calculates the signal envelope autocorrelation function within the window to obtain the logic step frequency. Logic step frequency The input second-order phase-locked loop circuit tracks the phase residual distribution between adjacent logic anchor points and calculates the phase bias base phase of the conference data stream. Logic step frequency With phase bias base phase The logic evolution feature matrix is ​​generated by encapsulating it into a two-dimensional feature vector; the adaptive calibration module generates a frequency correction instruction carrying a 16-bit digital compensation value; the data acquisition module loads the digital compensation value into the frequency division control register of the distributed terminal fractional frequency synthesizer; the frequency division control register is adjusted to adjust the sampling clock frequency by feeding back the frequency division ratio; and the hardware source end cancels the linear cumulative clock deviation caused by crystal oscillator aging.

[0030] Example 1: In a specific application scenario of the present invention, the system is deployed in a remote conferencing environment containing 12 distributed collaborative terminals. Due to cross-regional network link fluctuations, asymmetric delay disturbances occur in the audio stream, video stream, and collaborative instruction stream. The initial delay offset between the audio stream and the video stream is 450ms. Furthermore, at the 120s mark of system operation, a sudden signal interruption lasting 350ms occurs in the uplink channel. The mapping and correlation module extracts the energy transition characteristics in the audio stream and the transaction triggering instructions in the collaborative instructions as logical anchor points. Since the evolution trend analysis module within the preceding period has calculated that the logical step frequency of the audio stream is 100Hz and the phase offset base phase remains stable, the mapping and correlation module... Based on the temporal correlation weights, a causal coupling benchmark between modes is established in the multidimensional temporal alignment space. During the audio signal interruption, the adaptive calibration module generates virtual evolutionary feature quantities corresponding to the signal gaps in the multidimensional temporal alignment space with a step size of 10ms according to the logical evolutionary feature matrix. The topological constraint of the multidimensional temporal alignment space is maintained by the virtual evolutionary feature quantities. When the real audio signal is received again at 120.35s, the adaptive calibration module calculates that the logical displacement residual between the measured pose of the recovery point and the current virtual evolutionary feature quantity is 25ms. The system injects this logical displacement residual as a damping factor into the temporal scaling logic, so that the elastic scaling amount of the audio stream returns to the real physical state at a preset convergence rate.

[0031] Simultaneously, the adaptive calibration module calculates in real time the cumulative logical displacement of each conference data stream during feature alignment within the current processing cycle, generating alignment feature cost parameters. Alignment feature cost parameters satisfy: ,in, Alignment feature cost parameters; The total number of data streams participating in the alignment is 3 in this embodiment; For the first The preset weighting factors for the data streams, including the audio stream weight. The video stream weight is 0.6. The document instruction flow weight is 0.3. It is 0.1; For the first The system determines the logical displacement deviation of the data stream within the current processing cycle by comparing it with the alignment feature cost parameters. Based on a preset fidelity threshold, a reliability weight is assigned to each conference data stream, and the contribution of the corresponding source to the global alignment space is adjusted according to this weight. During continuous operation after signal recovery, the evolution trend analysis module analyzes the displacement residual distribution of the logical anchor point within a preset period window, uses the least squares regression algorithm to fit the evolution slope of the residual on the time axis, and identifies it as a linear cumulative clock deviation caused by the aging of the terminal device crystal oscillator. The adaptive calibration module generates a frequency correction command based on this slope and sends it to the corresponding data source terminal, so that the terminal adjusts its original sampling frequency based on the frequency offset parameter. This feedforward compensation procedure transforms the logical alignment pressure of the back end into hardware sampling coordination at the source end. After a 350ms signal interruption, the conference intent map reconstructed by the back end maintains a logical consistency state, realizing the maintenance of logical consistency of heterogeneous conference data streams under non-steady-state network conditions.

[0032] Example 2: In the multi-source heterogeneous conference data stream alignment stability verification experiment, the system was deployed in a distributed simulation environment containing 10 independent network nodes to verify the system's logical self-sustaining capability under discontinuous data loss conditions. The test data came from real conference records collected by the physical experimental platform, with a time resolution of 1ms and a raw audio sampling frequency of no less than 48kHz. To simulate electromagnetic and channel disturbances in real high-frequency collaboration, Gaussian white noise with a signal-to-noise ratio of 20dB was actively superimposed in the test signal source, and the sampling period was... The settings depend on the trade-off between the real-time performance of data acquisition and the system processing load, due to the spectral bandwidth of the monitored signal. Within the standard audio frequency band, sampling period The value is set to 10ms, thus maintaining alignment accuracy while keeping the dynamic overhead of the processing unit below the power consumption baseline. The experimental group uses the aforementioned specific implementation method with the logic evolution feature matrix and damped closed-loop correction procedure, while the control group uses a physical timestamp linear extrapolation alignment method. By adjusting the core variable, namely the signal gap duration... Multiple experimental groups with gradients were set up within the range of 50ms to 1000ms. The logic displacement residuals generated at the signal recovery points were recorded, and alignment feature cost parameters were generated based on the cumulative logic displacement during feature alignment of each data stream. See Table 1.

[0033] Table 1: Comparison of Alignment Performance under Different Signal Gap Durations

[0034]

[0035] Table 1 shows that in the sample group of this invention, when the signal gap duration... When fluctuating within the range of 50ms to 500ms, the alignment feature cost parameter provides virtual anchor points for predicted pose in the multidimensional temporal alignment space because the virtual evolutionary features provide virtual anchor points for predicted pose. The growth is gradual and remains within the fidelity threshold of 0.30, when the signal interval duration is... When the time is increased to 800ms, the cumulative residual of the historical evolution trend increases, and the alignment feature cost parameter... Rising to 0.42, the adaptive calibration module injects the residual as a damping factor into the elastic calibration engine, guiding the time-series scaling to regress to the true physical state. This avoids the semantic step jump phenomenon observed in the control group under the same operating conditions, demonstrating the system's stability when facing extreme flow interruptions. After 1000ms, the growth rate of key performance indicators decreases because the information entropy loss exceeds the compensation boundary of the fidelity threshold. This result proves that the present invention's scheme, by extracting the logic step frequency and phase bias base phase, makes the data processing effect exhibit a monotonically correlated gradient law with the degree of signal loss. The above experimental results confirm that the present invention's scheme utilizes the alignment feature cost parameter. The quantitative feedback enables automatic identification and suppression of fault sources, aligning the feature cost parameters. satisfy: ,in, Alignment feature cost parameters; The total number of data streams participating in the alignment is 3 in this experiment; For the first The preset weighting factors for the data streams, including the audio stream weight. The video stream weight is 0.6. The document instruction flow weight is 0.3. It is 0.1; For the first The logical displacement deviation of the data stream within the current processing cycle is used by the system to maintain the essential logical consistency of multi-source heterogeneous time-series data in an unsteady network environment.

[0036] Example 3: This example combines Figures 1 to 2 This section describes a multi-source time-series data analysis and processing system for intelligent conferencing, such as... Figure 1 As shown, the multi-source time-series data analysis and processing system for intelligent conferencing takes multiple heterogeneous conference data streams as input. The data acquisition module acquires the multiple heterogeneous conference data streams and identifies logical anchor points. The separated time-series data streams are transmitted to the evolution trend analysis module, while the identified logical anchor points are transmitted to the mapping and association module. The mapping and association module establishes time-series association weights and constructs a multi-dimensional time-series mapping space based on the input data. It outputs the constructed multi-dimensional time-series alignment space to the adaptive calibration module. The evolution trend analysis module calculates the step frequency and phase offset to generate a logical evolution feature matrix. The adaptive calibration module receives the logical evolution feature matrix and generates virtual evolution feature quantities in the signal gaps to fill the gaps. At the same time, it calculates the alignment feature cost parameters and adjusts the confidence weights. Through a feedback adjustment mechanism, it suppresses faults based on the confidence weights and finally outputs analysis results with physical authenticity and logical stability.

[0037] like Figure 2As shown, the conference data stream entity includes the original timestamp, logical anchor point, modal type (including audio, video, and command), and signal payload. The data generated by this entity is input into the data acquisition module, which maintains the distributed terminal list, original sampling frequency, and network jitter parameters. It is responsible for acquiring multiple heterogeneous data streams, encapsulating data packets, and adjusting the original sampling frequency according to frequency correction commands. Externally input heterogeneous data streams are transmitted to the mapping and association module, which maintains the temporal association weights, multi-dimensional temporal alignment space, topology mapping graph, and sub-logical plane set. It is responsible for establishing temporal association weights based on logical anchor points, constructing a multi-dimensional temporal mapping space, and establishing orthogonal sub-logical planes. The evolution trend analysis module receives the temporal features extracted from the data acquisition module, which internally includes a logical evolution feature matrix. The logic step frequency and phase offset base phase are calculated, and the logic evolution feature matrix is ​​generated. The evolution slope is fitted using the least squares method, and the dominant characteristic mode is identified. The feature matrix is ​​provided to the adaptive calibration module. The adaptive calibration module manages the alignment feature cost parameter P, confidence weight, virtual evolution feature quantity and fidelity threshold. It performs operations such as monitoring the transmission status, generating virtual evolution feature quantity in the signal gap, calculating the alignment feature cost parameter and adjusting the confidence weight, and generating frequency correction instructions. The correction instructions are sent to the instruction issuing module and fed back to the mapping association module to adjust the association weight to maintain the constraint. The instruction issuing module manages the real-time synchronization parameters and adjusts the hardware parameters of the data acquisition module by feeding back the real-time synchronization parameters and issuing frequency correction instructions.

[0038] Example 4: In a large-scale collaborative conference scenario involving 50 heterogeneous communication terminals, the terminal devices experience an intrinsic sampling deviation of 50 ppm due to crystal oscillator aging. To determine the quantized value of the global fidelity threshold, the system adopts a standardized calibration procedure. The initial state of the procedure is defined as follows: the adaptive calibration module selects a reference audio stream with a sampling frequency of 48 kHz and a synchronization reference as the calibration source; the processor records the cumulative logic displacement of the reference audio stream within a preset observation window; the statistical variance of the obtained displacement sequence is calculated; and the corresponding reference displacement power consumption is multiplied by a coefficient of 1.2, thereby calibrating the global fidelity threshold to 0.30, which serves as a quantified indicator for evaluating the confidence of heterogeneous data streams. The generation process of the logic evolution feature matrix is ​​implemented through the following algorithm path: the processor samples the conference data stream with a window width of 20 ms, calculates the autocorrelation function of the signal envelope within the current window to extract the logic step frequency. The obtained frequency parameters are then injected into a second-order phase-locked loop.

[0039] The phase bias base phase is calculated by tracking the phase residual distribution between adjacent logic anchor points. The specific matrix construction process is as follows: A two-row, ten-column circular shift storage array is established as the logical evolution feature matrix. The first row stores the logical step frequency values ​​of the most recent ten sampling periods in chronological order. The second row synchronously stores the corresponding phase bias base phase values. Whenever new frequency and phase data are acquired, the entire matrix is ​​shifted one column to the left, the oldest data in the tenth column is removed, and the newly acquired feature parameters are written into the first column to achieve dynamic feature capture of the evolution trend. The processor encapsulates the frequency vector and phase vector into a logical evolution feature matrix. When the signal gap duration exceeds 100ms, state extrapolation is performed based on the logical evolution feature matrix to generate virtual evolution feature quantities in the multi-dimensional temporal alignment space. The topological constraint strength of the multi-dimensional temporal alignment space is maintained through virtual evolution feature quantities. To handle alignment conflicts caused by multi-center intention entanglement, the processor calculates the entropy concentration for the topological association key distribution in each sub-logic plane. The specific steps are as follows: statistically analyze the triggering frequency of the associated keys within a specific sub-logic plane, calculate its normalized probability distribution, and solve for the entropy concentration of the sub-logic plane based on the probability distribution. ,in, The lower the value, the more meaningful the sub-logic plane. Figure 1 The higher the consistency, the higher the entropy concentration of a certain sub-logic plane. When the logical isolation threshold is below 0.15, the bit plane is determined to have a concentrated logical intent center, the timing pull weight of the conference data stream is increased, and the spurious data interference of the high entropy bit plane is suppressed. The processing results show that the system uses the global fidelity threshold and the solution logic of the logical evolution feature matrix to maintain the logical self-sustaining state of heterogeneous timing data during the physical channel interruption, and the residual displacement deviation of each data stream converges to within the engineering allowable range of less than 15ms.

[0040] Example 5: In a distributed conferencing deployment scenario involving multiple asymmetric network nodes, the processor executes a pre-calibration procedure to establish an operational baseline before initiating the alignment process. It controls the data acquisition module to send a synchronization pulse sequence with a duration of 5 seconds and a sampling interval of 100ms to the distributed terminals, and records the loop response time of the data stream feedback to calculate the network delay variance. and terminal clock bias parameters ,in, For network latency variance, For the terminal timing offset parameters, the processor processes the sampling frequency deviation trajectory using a linear fitting algorithm and calculates the initial fidelity threshold for each independent source node corresponding to the initial pose. ,in, As the initial fidelity threshold, this initial state definition procedure reduces the initial offset caused by the difference in hardware crystal oscillator accuracy in the allocation of timing correlation weights by using measured delay data, and provides a starting anchor point with physical coordinate attributes for the multi-dimensional timing alignment space.

[0041] When the system runs continuously for more than 4 hours and faces the cumulative phase drift caused by thermal drift of the terminal hardware, the adaptive calibration module monitors the time-varying rate of entropy concentration in each sub-logic plane. Execute the dynamic refactoring procedure, in which, The entropy concentration of the sub-logic plane. For the duration of the meeting, if the processor detects that the rate of change of entropy concentration of a specific sub-logic plane exceeds the preset steady-state drift threshold for three consecutive sampling periods, and the cumulative logic displacement residual reaches 1.5 times the initial calibration baseline, the processing unit triggers the logic plane realignment action. While maintaining the current virtual evolution feature quantity, the system performs position remapping on the logic anchor point through phase recapture logic.

[0042] Example 6: In a conference environment deployment scenario with heterogeneous sensor components, the processor executes an adaptive weight calibration procedure during the initialization phase. The initial state is defined as the data acquisition module acquiring the noise floor signal for 5 seconds under no-load conditions. The intrinsic signal-to-noise ratio of the source device is evaluated by calculating the power spectral density of the noise floor signal sequence. The initial weighting factor of each conference data stream is calculated using a linear mapping algorithm. ,in, For channel index, For the first Weighting factors for conference data streams, weighting factors The calibration procedure is positively correlated with the corresponding measured signal-to-noise ratio. It compensates for the initial logic offset caused by the difference in hardware performance in the allocation of timing correlation weights by using measured data.

[0043] The processor executes a gradient interference injection procedure to determine the logic isolation threshold. The procedure includes injecting a spurious intent sequence with known causal characteristics into the data acquisition module and increasing the external interference intensity from 10dB to 40dB in 5dB increments. The processor calculates the entropy concentration of each sub-logic plane under different interference gradients in real time. The correlation between the accuracy of intent recognition and the accuracy of intent recognition, where, To characterize the degree of orderliness in intent distribution, the processor defines the upper limit of entropy concentration corresponding to an intent recognition accuracy of 95% or higher as the logical isolation threshold. In this application example, a value of 0.15 was selected based on testing. This threshold is used when the entropy concentration of a specific sub-logic plane is monitored during actual operation. When the value remains above the threshold, the processing unit triggers feedforward compensation logic to reduce the pull contribution of the bit plane to the multidimensional timing alignment space, ultimately keeping the residual logic displacement deviation of the multi-source heterogeneous data stream within the engineering threshold range of 12.5ms under complex disturbance environment.

[0044] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-source time-series data analysis and processing system for intelligent conferencing, characterized in that, include: The data acquisition module acquires multiple heterogeneous conference data streams. The data acquisition module includes multiple distributed terminals, each of which is used to synchronously acquire audio signals, video signals, and collaborative document update instructions, and encapsulates the acquisition results into data packets with original timestamps and sends them to the mapping and association module. The mapping and association module establishes temporal association weights between the logical anchor points of various conference data streams, and constructs a multi-dimensional temporal alignment space; The evolution trend analysis module calculates the logical step frequency and phase offset base phase of each conference data stream on the time axis. Specifically, the logical step frequency is obtained by calculating the autocorrelation function of the signal envelope within the calculation window. Logic step frequency The input second-order phase-locked loop circuit tracks the phase residual distribution between adjacent logic anchor points and calculates the phase bias base phase of the conference data stream. Logic step frequency With phase bias base phase Encapsulate the data into two-dimensional feature vectors to generate a logical evolution feature matrix; The adaptive calibration module is configured to monitor the transmission status of each conference data stream. When a signal gap is detected in the target data stream, it maintains the temporal correlation weights corresponding to the target data stream. Based on the logical evolution feature matrix, it performs state extrapolation to generate virtual evolution feature quantities in the multi-dimensional temporal alignment space to maintain the logical constraint strength of the multi-dimensional temporal alignment space during signal loss. It calculates the cumulative logical displacement of each conference data stream in real time during feature alignment, generates alignment feature cost parameters that characterize the degree of intrusion of the calibration action on the original sampling structure, and compares the alignment feature cost parameters with a preset fidelity threshold. Based on the comparison results, it assigns confidence weights to each conference data stream and adjusts the correction contribution of the temporal correlation weights to the multi-dimensional temporal alignment space according to the confidence weights to suppress faulty data caused by source device failure or phase drift. Furthermore, the adaptive calibration module quantifies the alignment feature cost parameters based on the adjustment amount of the original sampling structure during the calibration operation. Alignment feature cost parameters satisfy: ,in, For the alignment feature cost parameter, The total number of data streams participating in the alignment. For the first Preset weighting factors for each data stream. For the first The logical displacement deviation of the data stream within the current processing cycle.

2. The multi-source time-series data analysis and processing system for intelligent conferencing according to claim 1, characterized in that, The evolution trend analysis module analyzes the distribution of displacement residuals of logical anchor points within a preset period window and uses the least squares regression algorithm to fit the evolution slope of displacement residuals on the time axis. The adaptive calibration module generates a frequency correction command based on the evolution slope and sends the frequency correction command to the data acquisition module to eliminate the linear cumulative clock deviation of the conference data stream by adjusting the original sampling frequency.

3. The multi-source time-series data analysis and processing system for intelligent conferencing according to claim 1, characterized in that, The temporal correlation weights established by the mapping correlation module are used to define the causal coupling strength between different modal data streams. The evolution trend analysis module extracts historical temporal features and compensates for the instantaneous loss of temporal correlation weights by predicting the causal coupling strength during signal gaps.

4. The multi-source time-series data analysis and processing system for intelligent conferencing according to claim 1, characterized in that, The adaptive calibration module performs hierarchical processing on each conference data stream according to the confidence weight. When the confidence weight is lower than the preset logical isolation threshold, it automatically cuts off the logical association between the conference data stream and the multi-dimensional time alignment space.

5. A multi-source time-series data analysis and processing system for intelligent conferencing according to claim 2, characterized in that, When generating frequency correction instructions, the adaptive calibration module also calculates the network jitter parameters of the distributed terminals and injects the network jitter parameters as a regularization constraint into the generation logic of the frequency correction instructions to distinguish between logic shifts caused by hardware clock drift and network latency fluctuations.

6. The multi-source time-series data analysis and processing system for intelligent conferencing according to claim 1, characterized in that, The multidimensional temporal alignment space displays the correlation status of multi-source data through a real-time updated topological mapping graph. Each node in the topological mapping graph represents a logical anchor point, and the edge weights between nodes are dynamically modulated by the logical evolution feature matrix.

7. The multi-source time-series data analysis and processing system for intelligent conferencing according to claim 1, characterized in that, When calculating the logical evolution feature matrix, the evolution trend analysis module identifies the dominant feature mode of each conference data stream during the interaction process by comparing the feature evolution rate of each conference data stream, and dynamically improves the weight centrality of the dominant feature mode in the multi-dimensional temporal alignment space.

8. The multi-source time-series data analysis and processing system for intelligent conferencing according to claim 1, characterized in that, The system also includes a command issuance module, which is connected to the adaptive calibration module and is used to feed back real-time synchronization parameters to each data source terminal based on the calibration results of the multi-dimensional time-series alignment space.

Images