A multi-source traffic data analysis method based on privacy computing fusion

Abstract

The present application belongs to the field of intelligent traffic data processing and privacy computing, and is used to solve the problem that the prior art cannot enable each participant to dynamically optimize the share transmission strategy according to the data fluctuation law and the network real-time pressure, specifically a multi-source traffic data analysis method based on privacy computing fusion, applied to a participant node, comprising: collecting local multi-source traffic data, and calculating the three-dimensional state coordinates of the current round according to the multi-source traffic data; mapping the three-dimensional state coordinates to a sub-cube in the preset three-dimensional feature space; performing differential compression on the secret sharing shares to be sent according to the compression ratio, and generating a compressed share package; sending the compressed share package together with the compression flag, the window type identifier and the current compression ratio to the server; the present application realizes accurate quantification of the dynamic evolution of the participant node transmission state; dynamically adjusts the compression ratio of the secret sharing share, so that the compression strength and the data mutation degree are adaptively matched.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation data processing and privacy computing technology, specifically a method for analyzing multi-source traffic data based on privacy computing fusion. Background Technology

[0002] With the development of intelligent transportation systems, traffic data from a single source can no longer meet the needs of precise control under complex road conditions, and multi-source data fusion has become an inevitable trend. However, data sharing between different data collectors (such as intersection monitoring, vehicle terminals, and mobile signaling) involves user privacy and data security. Therefore, privacy computing technologies (such as federated learning and secure multi-party computation) have been introduced into the field of multi-source traffic data analysis to achieve "data usable but not visible" fusion processing. In a typical implementation architecture, each participating node converts its local traffic data into a secret shared share and sends it to the server via the network for encrypted domain aggregation, thereby completing global state estimation without exposing the original data.

[0003] Existing methods typically assume stable communication links and sufficient bandwidth. Each participant uploads its share according to a fixed period and a fixed data format. The server uses a synchronous waiting mechanism to collect a specified number of shares before initiating fusion computation. This design works normally in laboratory environments or light-load scenarios. However, when the system is deployed in a real traffic environment, especially during morning and evening rush hours, a large number of intersection nodes simultaneously upload high-frequency traffic data, and the network bandwidth quickly approaches saturation. At this time, the fixed share size and fixed transmission strategy will cause significant delays in the data of some nodes due to buffer backlog, and the server's synchronous waiting mechanism will be further hampered by the slowest node, drastically increasing the overall fusion cycle. More importantly, traffic conditions themselves have dynamic characteristics—intersections may quickly evolve from free flow to congestion, and the data fluctuation amplitude will increase dramatically. However, existing transmission strategies cannot detect this trend, resulting in control commands being severely delayed due to data latency during the critical window period of congestion formation, missing the best intervention opportunity.

[0004] Therefore, the core deficiency of existing technologies lies in the lack of a refined control mechanism that can adaptively adjust data transmission behavior and balance resource competition and state sensitivity when communication resources are at critical saturation and the characteristics of traffic data change dynamically. How to enable each participant to dynamically optimize its share transmission strategy based on its own data fluctuation patterns and real-time network pressure without increasing additional bandwidth overhead has become an urgent technical problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-source traffic data analysis method based on privacy computing fusion, which solves the problem that existing technologies cannot enable each participant to dynamically optimize the share transmission strategy according to its own data fluctuation patterns and real-time network pressure. The technical problem to be solved by this invention is: how to provide a multi-source traffic data analysis method based on privacy computing fusion that enables each participant to dynamically optimize its share transmission strategy according to its own data fluctuation patterns and real-time network pressure.

[0006] The objective of this invention can be achieved through the following technical solutions: A multi-source traffic data analysis method based on privacy-preserving computation fusion, applied to participating nodes, includes: Step S1: Collect local multi-source traffic data and calculate the three-dimensional state coordinates of the current cycle based on the multi-source traffic data: the first coordinate component representing the degree of data fluctuation, the second coordinate component representing the degree of data urgency, and the third coordinate component representing the inherent attributes of the node. Step S2: Map the three-dimensional state coordinates to a sub-cube in the preset three-dimensional feature space, and determine the region identifier of the current round based on the mapping result; Step S3: Based on the current round and the regional identifier sequence of several historical rounds, calculate at least one trend feature. The trend feature includes the number of consecutive stay rounds, the speed of regional transitions, and the frequency of directional chain swings. Step S4: Based on trend characteristics and current region identifier, dynamically calculate the resource reservation ratio, dynamically adjust the compression ratio of the secret sharing share, and dynamically determine the anti-shake window size; Step S5: Perform differential compression on the secret shared share to be sent according to the compression ratio to generate a compressed share packet, and generate a compression flag to identify whether the current share is a differential share or a complete share, a window type flag to identify the window type to which the current share belongs, and allocate transmission bandwidth according to the resource reservation ratio. Step S6: Send the compressed share packet, along with the compression flag, window type identifier, and current compression ratio, to the server.

[0007] The present invention has the following beneficial effects: 1. By constructing a three-dimensional state coordinate system composed of data fluctuation level, data urgency level, and inherent node attributes, and extracting trend features such as the number of consecutive dwell rounds, the speed of regional transitions, and the frequency of directional chain swings, the system achieves precise quantification of the dynamic evolution of the transmission state of participating nodes. Based on this, the compression ratio of the secret sharing share is dynamically adjusted to adaptively match the compression intensity with the degree of data mutation: a high compression ratio is used to reduce the transmission load during the data stability period, and the compression ratio is automatically reduced to retain detailed features during the data turbulence period, avoiding the loss of key information or bandwidth waste caused by a fixed compression strategy. 2. Based on the region identifier and the number of consecutive dwell rounds, the resource reservation ratio is dynamically calculated and proactively adjusted according to the direction of region changes, realizing differentiated allocation of bandwidth resources under congestion critical state. Nodes in high-priority regions (regions III and IV) with longer dwell times receive higher transmission quotas, ensuring that urgent data at congested intersections can occupy bandwidth first. At the same time, the exponential smoothing mechanism avoids sudden oscillations in the reservation ratio, making the bandwidth allocation transition smoothly during load fluctuations, effectively alleviating head-of-line congestion and latency spikes when multiple nodes upload concurrently. 3. The size of the anti-shake window is dynamically determined by the directional chain sway frequency, so that the area switching criterion is adaptively correlated with the data noise intensity: when the sway frequency is high, the window is increased to suppress frequent policy switching caused by data jitter; when the sway frequency is low, the window is reduced to maintain a rapid response to changes in real traffic conditions. This mechanism eliminates the ping-pong effect of traditional fixed thresholds repeatedly jumping around the threshold, and significantly improves the stability of policy decisions. 4. A phase transition proximity index (the product of the number of consecutive dwell times and the first coordinate component) is introduced as a precursor criterion for congestion criticality. When the index exceeds a preset threshold, the downward adjustment of the compression ratio is automatically increased, enabling nodes to proactively reduce the compression ratio before congestion occurs in order to retain more data details. This predictive compensation mechanism effectively shortens the lag time from data upheaval to transmission strategy response, ensuring that the server obtains high-fidelity fusion results within the critical window of congestion evolution, thereby providing timely and accurate decision-making basis for traffic light control. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart of a method according to Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the adjustment of the phase transition proximity index and enhanced compression ratio in Embodiment 1 of the present invention. Detailed Implementation

[0010] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] In the field of intelligent traffic management, the fusion of multi-source traffic data is the foundation for achieving accurate road condition perception and signal control. Traditional methods rely on centralized data collection, but face the risk of privacy leakage when data sharing is involved among different data collectors. To address this, privacy-preserving computation technologies (such as secure multi-party computation and federated learning) have been introduced into traffic data analysis. Each participating node converts the traffic data collected locally into a secret shared share and sends it to the server for encrypted domain aggregation, thereby completing global fusion without exposing the original data. This architecture can operate effectively under ideal conditions of stable communication and load balancing.

[0012] However, real-world traffic environments exhibit significant dynamic and time-varying characteristics. During morning and evening rush hours, numerous intersection nodes simultaneously upload high-frequency data, rapidly pushing communication links to near saturation. Simultaneously, the data fluctuation at individual intersections varies drastically with traffic conditions (e.g., from free flow to congestion). Existing solutions typically employ fixed share transmission cycles and uniform compression strategies, failing to detect critical network congestion pressures or identify the dynamic evolution of node data characteristics. When the system is under a dual critical state of strained communication resources and intensified data fluctuations, fixed strategies can lead to the following cascading failures: On the one hand, all nodes send their shares with the same priority, and emergency data at critical intersections compete for bandwidth with redundant data at ordinary intersections, causing high-value data to be delayed due to buffer backlog. On the other hand, the baseline update of differential compression lags behind data mutations, causing share reconstruction errors to accumulate, and the fusion results obtained by the server lag behind the actual traffic conditions by several seconds to tens of seconds. More seriously, existing methods lack quantitative means for measuring the long-term trends of node states (such as regional dwell time, switching acceleration, and trajectory oscillation frequency), and cannot distinguish between data jitter and actual traffic phase changes, resulting in control commands missing the best intervention opportunity due to information distortion during the critical window period of congestion formation.

[0013] Therefore, the core deficiency of existing technologies lies in the lack of an adaptive transmission control mechanism that can simultaneously perceive network congestion pressure and the dynamic evolution trend of node data during the privacy-preserving computation fusion of multi-source traffic data. How to enable each participating node to autonomously adjust its share compression strategy, bandwidth occupancy priority, and state switching criteria based on its own state changes and real-time communication conditions, while protecting privacy, in order to ensure low-latency transmission and high-fidelity reconstruction of emergency data in critical states, has become an urgent technical problem to be solved.

[0014] Example 1: As Figure 1-2 As shown, a multi-source traffic data analysis method based on privacy-preserving computation fusion is applied to participating nodes, including: Step S1: Collect local multi-source traffic data and calculate the three-dimensional state coordinates of the current cycle based on the multi-source traffic data: the first coordinate component representing the degree of data fluctuation, the second coordinate component representing the degree of data urgency, and the third coordinate component representing the inherent attributes of the node. Participating nodes perform data collection and processing at fixed intervals (e.g., every 2 seconds is one round). In each round t (t is a positive integer, t=1 indicates the first round), the node acquires multi-source traffic data from locally connected sensors (such as geomagnetic coils, radar speedometers, and video analysis modules), including at least the average lane speed values ​​of the current round and the previous few rounds. For ease of description, let the average speed collected by the node in round t be... (Unit: km / h); In this embodiment, the preset speed limit is 120 km / h, which is used for subsequent normalization calculations.

[0015] Based on the collected velocity sequence, the node calculates the first coordinate component, which characterizes the degree of data fluctuation; specifically, the node extracts velocity values ​​from five consecutive rounds to form a sequence. (If t < 5, fill with existing rounds, replacing any insufficient values ​​with the most recent ones); First, count the number of sign flips in the direction of velocity change between adjacent rounds: for k from t-3 to t, calculate the direction sign. The sign function outputs +1 (increasing), -1 (decreasing), or 0 (unchanged); then the sign sequence is statistically analyzed. The number of times adjacent signs change from +1 to -1 or from -1 to +1 is denoted as f (range 0-3); secondly, the calculation speed is extremely poor. and normalized to Finally, the first coordinate component For example, if the velocity sequence for a certain round is [45,46,48,47,45], then the sign sequence is [+1,+1,-1,-1], the number of flips f=1 (+1→-1), and the range=48-45=3. =3 / 120=0.025, therefore =0.6×(1 / 3)+0.4×0.025=0.2+0.01=0.21, indicating that the data fluctuation is small; to prevent outliers from causing the normalization result to exceed the [0,1] interval, the system will calculate the... Compare with 1 and take the smaller value: .

[0016] The node then calculates the second coordinate component, which represents the urgency of the data. The node locally stores two fixed parameters: a value score (val) and a sensitivity score (sens). The value score is updated by the server every 5 minutes, with an initial default value of 0.5, reflecting the node's contribution to global congestion prediction. The sensitivity score is preset according to the data field type; in this embodiment, the sensitivity score for the speed field is 0.3, for the location field it is 0.9, and for the occupancy rate field it is 0.5. Furthermore, the node records the time-sensitivity round number (age) of the current round of data, i.e., the number of rounds elapsed from the data collection time to the current round. In the first round, age = 0, and age increases by 1 with each subsequent round. The time-sensitivity decay factor (decay = e) is also recorded. -0.1×age Where e is a natural constant, and when age is greater than the preset maximum number of time-limited rounds (50 in this embodiment), the node directly discards the data frame and no longer participates in subsequent processing; then the second coordinate component =(val×0.4+sens×0.6)×decay; For example, if val=0.8, sens=0.3, and age=0, then =(0.8×0.4+0.3×0.6)×1=0.32+0.18=0.50; If the same data is not sent after 5 rounds (age=5), then decay=e -0.5 ≈0.6065, =0.5×0.6065≈0.303.

[0017] The node then calculates its third coordinate component, which represents the node's inherent attributes. The node pre-stores a node type coefficient (type) and a time period coefficient (period). The type coefficient is preset based on the intersection's criticality: 1 for critical intersections (such as interchanges) and 0.3 for ordinary intersections. The time period coefficient is preset based on the current time period: 1 for peak hours (7:00-9:00, 17:00-19:00), 0.2 for off-peak hours, and 0 for nighttime (0:00-6:00). The third coordinate component... =type × 0.7 + period × 0.3; For example, key intersections during peak hours: =1×0.7+1×0.3=1.0; During off-peak hours at ordinary intersections: =0.3×0.7+0.2×0.3=0.21+0.06=0.27.

[0018] At this point, the node obtains the three-dimensional state coordinates of the current round. The coordinates are then used for region mapping in step S2. It is worth noting that the weighting coefficients (such as 0.6, 0.4, 0.7, 0.3) and the constant 0.1 in the attenuation factor involved in the above calculation process are all pre-calibrated based on traffic engineering experience in this embodiment. In actual applications, they can be adjusted according to the characteristics of the road network.

[0019] To fully support other technical features associated with step S1, this embodiment also needs to explain the following: In the calculation of the first coordinate component, the "preset number of rounds" is specifically 5 rounds, and the "preset speed limit" is 120km / h; In the calculation of the second coordinate component, the "time decay factor" adopts an exponential decay model with a decay rate λ=0.1; The specific assignment methods of the "node type coefficient" and "time period coefficient" in the calculation of the third coordinate component are as above; These specific values ​​and calculation rules are exemplary descriptions of this embodiment and do not constitute a limitation on the scope of protection.

[0020] The value score val is calculated by the server every 5 minutes based on the gradient change magnitude of the historical uploaded data of each participating node, and then distributed to each node. Specifically, the server calculates the standard deviation of the gradient vector magnitude of each node in the last 100 rounds, divides the standard deviation by the maximum value among all nodes, normalizes it to the [0,1] interval, and uses it as the new value score of the node. When a node is first started, the value score is set to 0.5 by default.

[0021] Through step S1 above, the participating nodes transform the original multi-source traffic data into three-dimensional state coordinates with unified dimensions, providing a basic input for subsequent trend perception and adaptive transmission control.

[0022] Step S2: Map the three-dimensional state coordinates to a sub-cube in the preset three-dimensional feature space, and determine the region identifier of the current round based on the mapping result; After completing the three-dimensional state coordinate calculation in step S1, the participating nodes execute step S2, mapping the three-dimensional state coordinates to a sub-cube in the preset three-dimensional feature space, and determining the region identifier of the current round based on the mapping result. In this embodiment, the three-dimensional feature space is composed of the first coordinate component (data fluctuation level), the second coordinate component (data urgency level), and the third coordinate component (node ​​inherent attributes) as the X-axis, Y-axis, and Z-axis, respectively, with each axis taking values ​​in the range [0,1]. To discretize the continuous space, the system pre-sets two grading thresholds in each dimension: the first preset grading threshold is denoted as... =0.33, the second preset threshold is denoted as =0.66, and The selection of these two thresholds is based on statistical quantile analysis of a large amount of historical traffic data: the empirical distribution of data fluctuation, urgency and inherent attributes is divided into three equal parts, so that the sample size of the three intervals of low ([0,0.33)), medium ([0.33,0.66)) and high ([0.66,1]) is roughly balanced, thereby avoiding mapping bias.

[0023] For the three-dimensional state coordinates of the current round For each coordinate component, the node performs the following judgment: if the component is less than... If the value is less than or equal to 0, it is classified as low-grade and encoded as 0; if the value is greater than or equal to 0, it is classified as low-grade and encoded as 0. and less than If the value is greater than or equal to 1, it is classified as medium-range and encoded as 1; if the value is greater than or equal to 1, it is classified as medium-range and encoded as 1. If the coordinates are (0.21, 0.50, 1.00), then x = 0.21 < 0.33 → code 0, y = 0.50 ∈ [0.33, 0.66) → code 1, z = 1.00 ≥ 0.66 → code 2. The combination of codes for the three dimensions forms a triple (0, 1, 2), which uniquely determines the sub-cube number to which the current round belongs. Specifically, the formula for calculating the sub-cube number is... ,Right now In the example above, C = 9 × 0 + 3 × 1 + 2 = 5, indicating that the cube falls into the sub-cube numbered 5. The above weight allocation is based on the ternary encoding principle: the level code for each coordinate component has three possible values: 0, 1, and 2; these three codes are considered as a single ternary number, where... Corresponding highest bit (weight 3) 2 =9), Corresponding to the second highest position (weight 3) 1 =3), Corresponding least significant bit (weight 3) 0 =1), thus linearly mapping the three-dimensional discrete space to a continuous integer from 0 to 26, ensuring that each sub-cube obtains a unique number; placing the first coordinate component with the highest degree of fluctuation in the highest position allows the number value to reflect the degree of data fluctuation first, making it easier to quickly distinguish subspaces with different fluctuation levels during subsequent region mapping; This weighting is not arbitrary, but based on the priority of each coordinate component's impact on the transmission strategy: the first coordinate component (data fluctuation level) directly determines the stability of differential compression and is the most sensitive indicator, therefore it is given the highest weight of 3. 2 =9; The second coordinate component (data urgency) affects bandwidth priority, with a weight of 3. 1 =3; The third coordinate component (an inherent attribute of the node) changes slowly, so it has the lowest weight of 3. 0 =1; Through this weighted encoding, the numerical value of the sub-cube number first reflects the level of fluctuation, and then the urgency and inherent attributes are superimposed, so that subsequent region mapping can be classified according to the level of fluctuation first.

[0024] It should be noted that the above-mentioned tiering thresholds and numbering calculation methods are only one feasible discretization scheme. Before actual deployment, the system will build an offline region mapping table. The core function of this mapping table is to map each sub-cube number to one of five non-overlapping regions. The five regions are Region I, Region II, Region III, Region IV, and Region V, which represent different transmission strategy priorities and resource allocation levels. The construction logic of the mapping table is as follows: First, based on the actual network congestion level and traffic state evolution patterns corresponding to each sub-cube in historical traffic data, experts... The system or clustering algorithm divides the 27 sub-cubes into five clusters. For example, sub-cubes with low volatility, low urgency, and low intrinsic attributes (such as (0,0,0)) typically correspond to Zone I (green pass-through zone), which can implement a high compression ratio strategy; sub-cubes with high volatility, high urgency, and high intrinsic attributes (such as (2,2,2)) correspond to Zone IV (red protection zone), which requires reserved bandwidth and prohibits compression. In this embodiment, the region mapping table is indexed by the sub-cube number (0-26) and is pre-stored in the local memory of the participating nodes. The specific mapping relationship is shown in Table 1. Table 1: Mapping Table of Sub-Cube Labels and Region Identifiers; The above range division is only an example; in actual deployment, it can be recalibrated according to specific road network characteristics and communication conditions. Taking the sub-cube number 5 corresponding to the coordinates (0.21, 0.50, 1.00) calculated in step S1 as an example, it can be found from the table that it belongs to Zone II (yellow buffer zone); the node identifies this zone as... And serve as the input for step S3.

[0025] Through the above process, step S2 discretizes the continuous three-dimensional state coordinates into a finite number of region identifiers, realizing the rapid classification of the current transmission status of participating nodes. This classification does not rely on complex online learning or external server interaction, but is based entirely on a locally pre-stored static mapping table, thus ensuring low latency and high stability. It is worth noting that the update frequency of the region identifiers is the same as the data collection round (once every 2 seconds), which can reflect the dynamic changes in traffic status in a timely manner.

[0026] Step S3: Based on the current round and the regional identifier sequence of several historical rounds, calculate at least one trend feature. The trend feature includes the number of consecutive stay rounds, the speed of regional transitions, and the frequency of directional chain swings. In step S2, the region identifier for the current round t is determined. Next, the participating nodes execute step S3, calculating three trend features based on the current round and the region identifier sequence of several historical rounds: the number of consecutive stay rounds, the speed of region transitions, and the frequency of directional chain swaying. These features are used for adaptive parameter adjustment in the subsequent step S4. The specific calculation process for each feature is described below: Calculation of consecutive stay rounds: Each node maintains a historical consecutive counter. The initial value is 0; during each round of execution of step S3, the node obtains the region identifier for the current round. Compared with the previous round of sub-regional identification (For the first round t=1, there is no) (At this point, directly set the number of consecutive stops to 1); compare whether the two are equal: if Then increment the counter by 1, that is... ;like If the counter fails, reset it to 1; then, use the current value of the counter as the number of consecutive stops. Output; for example, if the region identifier sequence of a node for five consecutive rounds is [Ⅱ,Ⅱ,Ⅱ,Ⅲ,Ⅲ], then the corresponding region identifier for each round is... The values ​​are 1, 2, 3, 1, 2 in sequence; this feature intuitively reflects the length of time a node stays in the current region, and the larger the value, the more stable the state.

[0027] Calculation of the abruptness of regional transitions: This feature is used to quantify the drasticness of node switching between different regions; First, the node maps five preset region identifiers (region I, region II, region III, region IV, and region V) to consecutive integer values: region I → 1, region II → 2, region III → 3, region IV → 4, region V → 5; This mapping relationship is pre-stored locally on the node.

[0028] Then, the node obtains the region identifier integer values ​​for three consecutive rounds, including the current round, denoted as . For the startup phase (t<3), where there are fewer than three historical rounds, the speed of the region transition is directly set to the preset default value (0 in this embodiment); for t≥3, the first absolute difference is calculated sequentially. Second absolute difference Next, calculate two intermediate values: the first difference. The second difference ; noticed Therefore, the two are equal after taking their absolute values; in this embodiment, the degree of abruptness of the regional transition is... Defined as and The average value, i.e. Simplified calculation is as follows .

[0029] For example: Suppose that the integer values ​​of three consecutive rounds of area identifiers are [2, 3, 5] (i.e., area II → area III → area V), then =|3-2|=1, =|5-3|=2, =|1-2|=1; If the sequence is [2,2,3], then =0, =1, =1; if the sequence is [2,4,2], then =2, =2, =0; The larger this characteristic value is, the greater the difference in the transition step size between two adjacent wheels, that is, the more obvious the acceleration of the area switching, which indicates that the traffic state may change abruptly. It is worth noting the varying degrees of speed of regional leaps. What is being measured is not the absolute step size of a single transition, but rather the change in step size between two adjacent transitions; therefore, although the [2,4,2] sequence has a relatively large single transition step size (both are 2), the two transition step sizes are equal, and the change is 0. =0 indicates that the transition behavior remains stable; while the [2,2,3] sequence, although having a smaller single step size, shows a positive change as the step size increases from 0 to 1, therefore... =1 indicates that the transition is accelerating; this feature is mainly used to identify precursors to a sudden change in traffic conditions, rather than to measure the absolute severity of the current state.

[0030] Calculation of the directional chain oscillation frequency: In each round's step S2, in addition to obtaining the region identifier, the node also records the directional chain triplet; the directional chain triplet records the sign of the change in the current round relative to the previous round in three dimensions; specifically, let the three-dimensional state coordinates of the previous round be... The current round is Then the sign of change for each dimension is defined as: ; Similarly defined and Triples This is the direction chain for the current round; for the first round (without a previous round), the direction chain triple is (0,0,0); the node maintains the direction chain triples for the three most recent rounds, denoted as... (If there are fewer than three rounds, fill with (0,0,0).

[0031] To calculate the oscillation frequency, the nodes are processed independently for the three dimensions of X, Y, and Z; taking the X dimension as an example, three consecutive changing symbols are extracted to form a symbol sequence. Each symbol takes the value of +1, -1, or 0. The number of times positive and negative signs change between adjacent symbols in the sequence is counted, including both +1→-1 and -1→+1. Changes involving 0 (such as +1→0 or 0→-1) are not counted. The counting method is as follows: sequentially check... and If a pair of symbols satisfies the condition that one is +1 and the other is -1, then the count is incremented by 1; this yields the number of transformations in dimension X. The value ranges from 0 to 2; similarly, we obtain... and The total number of transformations is obtained by summing the number of transformations in the three dimensions. The value ranges from 0 to 6.

[0032] Finally, the directional chain oscillation frequency Defined as the total number of transformations divided by a preset normalization factor; in this embodiment, the normalization factor is 3, i.e. The value ranges from 0 to 2; for example, if the symbol sequence of dimension X is [+1,-1,+1], then adjacent pairs (+1,-1) are counted once, (-1,+1) are counted once, for a total of 2 times; if dimension Y is [+1,0,-1]: (+1,0) is not counted, (0,-1) is not counted, for a total of 0 times; if dimension Z is [+1,+1,-1]: (+1,+1) is not counted, (+1,-1) is counted once, for a total of 1 time; then =2+0+1=3, =3 / 3=1.0; if none of the three dimensions have a sign inversion, then =0; if each dimension is reversed twice (e.g., [+1, -1, +1]), then =6 / 3=2.0; The higher the swaying frequency, the more frequently the three-dimensional state coordinates of the node are folded back and forth in space, which usually corresponds to the unstable jitter of the data source rather than the actual changes in traffic trends.

[0033] In step S3, the node obtains the number of consecutive stay rounds. The degree of speed of regional leap and directional chain oscillation frequency Three trend characteristics; these characteristics will be passed to step S4 for dynamically calculating the resource reservation ratio, adjusting the compression ratio, and determining the stabilization window size; it is worth noting that all calculations are based on locally stored historical window data and do not depend on information from servers or other nodes, thus ensuring the data independence of each participant under the privacy computing framework.

[0034] Before executing step S4, this embodiment also generates a phase transition proximity index based on the number of consecutive dwell rounds and the first coordinate component, which is used to identify the critical trend of traffic state evolving from free flow to congested flow in advance; specifically, the participating nodes obtain the number of consecutive dwell rounds calculated in step S3. (Value range 1 to 10) and the first coordinate component calculated in step S1 (Value range 0-1); Multiplying the two yields the phase transition proximity index. The physical meaning of this index is: when a node stays in the same area for a long time ( (relatively large) and its internal data fluctuates significantly. A large fluctuation indicates that although no area switch has occurred, the data is already fluctuating drastically, a typical precursor to impending congestion. For example, if a node is in Zone II (yellow buffer) for five consecutive rounds, and the data fluctuation index for each round is 0.20, 0.25, 0.30, 0.35, and 0.40 respectively, then in the fifth round... =5, =0.40, =2.0; if in the 6th round If it rises to 0.60, then =3.0.

[0035] The node will generate a phase transition proximity index With the preset exponential threshold A comparison is made; in this embodiment, based on statistical analysis of a large amount of historical traffic data, the threshold is set to... =5.0; when When the value is >5.0, it is determined to be in a "phase transition is about to occur" state, and the compression ratio needs to be further corrected downward in step S4 to retain more data details; when When the value is ≤5.0, it is considered to be in a normal state and no additional correction is required.

[0036] Step S4: Based on trend characteristics and current region identifier, dynamically calculate the resource reservation ratio, dynamically adjust the compression ratio of the secret sharing share, and dynamically determine the anti-shake window size; Step S3 completes the trend characteristics (number of consecutive stays). The degree of speed of regional leap directional chain oscillation frequency After the calculation, the participating nodes execute step S4, based on these trend characteristics and the current region identifier. The process involves dynamically calculating the resource reservation ratio, dynamically adjusting the compression ratio of the secret sharing share, and dynamically determining the anti-jitter window size. This step is the core of adaptive transmission control. The calculation of each parameter is independent of each other, but they ultimately work together to affect the share transmission behavior in step S5.

[0037] Dynamically adjust the compression ratio of the secret sharing share: The adjustment of the compression ratio depends on the speed of the regional transition. and the phase transition proximity index The node first maintains a historical sliding window, recording the agility or slowness of region transitions in the last 20 rounds (if there are fewer than 20 rounds in the initial phase, a default value of 0.5 is used to fill the gap); the median of all values ​​within this window is calculated and denoted as . The median is more resistant to transient noise than the mean.

[0038] The node obtains the speed of the region transition in the current round. And compare its relative size with the historical median; adjustment range Determined by the following formula: ; in =4 is the preset maximum adjustment range, and 0.1 in the denominator is used to prevent division by zero; this S-shaped function ensures that when hour, =2; when Much larger hour, Approaching 4; when much smaller hour, Approaching 0.

[0039] Next, determine the adjustment direction; the node calculates the change step size for the current round. (Mapping region identifiers I to V to 1 to 5), and the step size of the previous round of changes. ;like This indicates that the region handover is accelerating. At this point, the compression ratio should be reduced to retain more data details. Therefore, the adjustment direction is negative. Otherwise, adjust the direction to positive. .

[0040] Based on this, the node also needs to consider the phase transition proximity index. This index is obtained by multiplying the number of consecutive stops by the first coordinate component: The calculation has been completed after step S3 and before step S4; the node will With the preset exponential threshold =5 for comparison: If This indicates that nodes are staying in the same area for an extended period and data fluctuations are significant, suggesting an impending congestion state. In this case, the compression ratio needs to be further reduced to capture key changes; specifically, the change in compression ratio should be increased by a preset scaling factor. =0.5, which is the final change. ;otherwise Taking a specific numerical value as an example: Assume that the original calculation in step S4 yielded... (Indicates a need to reduce the compression ratio), currently =6.0>5.0, then =-2×1.5=-3.0, the final compression ratio is reduced even more; if =3.0, then maintain =-2 remains unchanged; through this mechanism, the system can automatically enhance data fidelity during the critical period of congestion, providing the server with more accurate information on the evolution of congestion, thereby improving the response effect of traffic light control.

[0041] It should be noted that the above-mentioned preset exponential threshold =5.0 and the scaling factor =0.5 are all calibration values ​​based on simulation experiments of typical urban intersections in this embodiment. In actual deployment, they can be calibrated according to road network characteristics and communication delay requirements. In addition, the calculation of the phase transition proximity index only involves multiplication operations, with extremely low computational overhead, and does not affect the real-time processing capability of the node.

[0042] Ultimately, the node update compression ratio ,in The compression ratio used in the previous round (initial value is 8); to ensure the compression ratio is within a reasonable range, the node will... The clamp is positioned between the preset minimum compression ratio of 2 and the maximum compression ratio of 12: And rounded down to the nearest integer; for example, suppose the current =8, =1.2, =0.6, then The exponent is -2 × 0.857 = -1.714, e -1.714 ≈0.18, =4×1 / (1+0.18)≈4×0.847=3.39, take the approximation 3.4; if ,but =-3.4, =4.6, after clamping, rounded to 5; if at this time =6>5, then Multiply by 1.5 to get -5.1. =2.9, clamping rounds to 3.

[0043] Dynamically determine the size of the debouncing window: The debouncing window is used in subsequent steps to determine the number of consecutive identical state rounds required for strategy switching; the node will use the oscillation frequency of the direction chain obtained in step S3. Mapped to a preset window value range [2, 8] (unit: rounds); the mapping relationship is linear: ;because The value range is 0 to 2, therefore The range is 2 to 8; the node rounds down the calculation result and uses it directly as the final debouncing window size; for example, When =0.3, =2 + 0.9 = 2.9, rounded up, we get 2 rounds; When =1.5, =2 + 4.5 = 6.5, rounded up to 6 rounds; this window size will be used in subsequent rounds to determine whether the region switching is real (e.g., if continuous switching is required). (The policy update is only triggered when the wheel is in a new region), thereby suppressing frequent policy changes caused by data jitter.

[0044] Dynamically calculate the resource reservation ratio: The resource reservation ratio determines the additional bandwidth quota that participating nodes can occupy when sending shares; the node first determines the target reservation ratio based on the current region identifier and the number of consecutive stay rounds. If the current region identifier belongs to the preset high-priority region set {region III, region IV}, and the number of consecutive stays is... If ≥3, then the target proportion is ;otherwise =0; This design allows nodes to obtain more reserved bandwidth the longer they stay in a high-priority area, until the maximum is reached (reaching 1 after 10 rounds).

[0045] To avoid bandwidth allocation oscillations caused by sudden changes in the reservation ratio, nodes update the current reservation ratio using an exponential smoothing method. : ; in This is the reserved proportion from the previous round, with α=0.8 as the smoothing coefficient (the closer to 1, the smoother the change); for example, the previous round =0.2, current =0.5, then =0.8×0.2+0.2×0.5=0.16+0.1=0.26, gradually moving closer to the target.

[0046] In addition, nodes proactively adjust the reservation ratio based on the direction of regional changes; nodes compare the change step size of the current round with that of the previous round: if >0 and >0 indicates that the regional changes are consistent and moving in the direction of increasing values ​​(e.g., from region II → region III → region IV), then it is predicted that the next round will continue to enter a higher priority region; if <0 and If the value is less than 0, the prediction will leave the current high-priority region; if the prediction enters a higher-priority region and the current value is less than 0, the prediction will leave the current high-priority region. When the value is greater than 0.5, the node will be the current one. Increase by 0.1 (but not exceeding 1); when the prediction leaves the high-priority area and When >0.5, Reduce by 0.1 (but not below 0); for example, assuming the current... =0.5, predicting it will enter zone IV and =0.8, then after the update =0.6; This forward-looking adjustment allows nodes to preempt bandwidth before congestion actually occurs, reducing response latency.

[0047] Image stabilization window This is used to determine the validity of region switching in subsequent rounds; specifically, after calculating the new region identifier in each round's step S2, the node maintains a counter to record the number of rounds in which it is continuously in the same region identifier; only when the number of consecutive rounds reaches a certain threshold will the node be considered valid. Only when the node confirms that the region switch is complete, and triggers the parameter update based on the new region in step S4; otherwise, it still uses the region identifier from the previous round for decision-making. This mechanism effectively suppresses frequent policy switching caused by data jitter.

[0048] In step S4, the node obtains the updated compression ratio. Shake stabilization window size and resource reservation ratio These parameters will be directly used for differential compression and buffer management in step S5, and for sending priority control in step S6. It is worth noting that all calculations are based on the trend characteristics observed locally by the nodes, without the need to interact with the server, thus ensuring the autonomy of each participant and low-latency response under the privacy computing framework.

[0049] Step S5: Perform differential compression on the secret shared share to be sent according to the compression ratio to generate a compressed share packet, and generate a compression flag to identify whether the current share is a differential share or a complete share, a window type flag to identify the window type to which the current share belongs, and allocate transmission bandwidth according to the resource reservation ratio. The compression ratio is completed in step S4. Resource reservation ratio and the size of the stabilization window After the calculation, the participating nodes execute step S5, which performs differential compression on the secret shared share to be sent, generates compressed share packets and corresponding control flags, and allocates transmission bandwidth according to the resource reservation ratio. This step is the specific execution layer of the adaptive transmission strategy, which transforms the preceding decisions into actual data packet processing behavior.

[0050] Generation and caching of secret sharing shares: In each round, participating nodes first convert locally collected multi-source traffic data (such as average speed, occupancy rate, etc.) into secret sharing shares; this embodiment adopts a (3,N) secret sharing scheme, where N is the total number of participating parties (N≥5), and the share is an integer ranging from 0 to 65535; the node will then generate and cache the original data of the current round. It was split into multiple shares, one of which This node is responsible for sending the data to the server; the node maintains a cache variable locally. This is used to store the actual shares sent in the previous round (whether full shares or differential shares); at the start of the first round, Initialize to 0.

[0051] Differential compression and compression flag generation: Nodes execute different differential compression logic depending on whether the current round is the starting round; Initial round (t=1): Since there is no previous round's share as a benchmark, the node cannot calculate the difference; at this time, the node directly uses the original share of the current round. As the share to be sent, and the compression flag will be used. Set to the first value (in this embodiment, the first value is 1, representing the full share); simultaneously, the node will Save to For use in the next round; Non-initial rounds (t≥2): Nodes acquire the current share. Compared with the share already sent in the previous round Calculate the difference and take the absolute value Subsequently, the node adjusts the compression ratio accordingly. Determine the maximum allowable deviation range for the transmission difference. Specifically, the node first obtains a preset basic threshold. =10, then calculate the compression ratio related terms. ×5, and take the larger of the two values ​​as the dynamic threshold: For example, if =8, then ×5=40, ;like =3, then .

[0052] Node comparison and Size: like ≤ This indicates that the difference between the current share and the previous share is small, and the difference can be used to save bandwidth; at this time, the node will use the difference. (rather than the original share) ) as the share to be sent, and the compression flag. Set to the second value (in this embodiment, the second value is 0, representing the difference share); note that the difference It may be negative. In this embodiment, an 8-bit signed integer (range -128 to 127) is used for encoding. Before sending, it needs to be clamped to this range (if it exceeds the range, the whole share will be sent instead). like > This indicates a significant change in the data; the difference suggests potential information loss or exceeding the encoding range. In this case, the node will return the original share. As the share to be sent, the compression flag is set to the first value (full share).

[0053] Regardless of the method used, after determining the share to be sent, the node updates it to... (Right now (so that it can be used as a benchmark in the next round).

[0054] For example: Assume the share sent in the previous round =100, current share =108, =8, then =40, =8 < 40, therefore send the difference. =+8, compression flag is 0; if =200, then =100>40, send the complete share 200, and set the compression flag to 1.

[0055] Window type identifier generation: To support differentiated processing on the server side, nodes need to generate a window type identifier for each share to be sent. This indicates whether the share should enter the server's fast window or a normal window; the window type identifier is generated based on a comprehensive priority, which is determined by the value score, sensitivity score, and time-sensitive rounds.

[0056] Specifically, the node first calculates the overall priority pri of the current share: ; Where val is the value score updated periodically by the server (initially 0.5 by default), sens is the sensitivity score (0.3 for speed, 0.9 for location, and 0.5 for occupancy), and age is the current data's time-lapse round (the definition of age is the same as in step S1); for example, if val=0.8, sens=0.3, and age=0, then pri=0.8×0.3+0.3×0.7=0.24+0.21=0.45; if the same data has gone through 2 rounds (age=2), then the decay factor e -0.2 ≈0.8187, pri≈0.45×0.8187≈0.368.

[0057] The node maps the overall priority to the probability of the fast window. The meaning of this formula is: when pri ≤ 0.5, =0 (will definitely use the normal window); when pri≥1.0, =1 (always uses the fast window); when pri is between 0.5 and 1.0, the probability increases linearly; in the example above where pri = 0.45, =0, the window type identifier is set to normal window; if pri=0.75, then =(0.75-0.5)×2=0.5, the node generates a random number r uniformly distributed in [0,1]. If r<0.5, the window type identifier is set to fast window (represented by 1 in this embodiment), otherwise it is set to normal window (represented by 0). The random number can be generated using standard pseudo-random number generators such as linear congruential method.

[0058] Transmission bandwidth allocation based on resource reservation ratio: Resource reservation ratio (Value range 0-1) determines the additional bandwidth quota that a node can use when sending the current round's share packet; in this embodiment, the node uses a leaky bucket mechanism for bandwidth control: each round assigns the node a sending quota. ,in The baseline quota is 500 bytes (in this example, it is taken as 500 bytes); the length of the frame to be sent is (A complete share frame is approximately 22 bytes, and a difference share frame is approximately 8 bytes); if the total length of the queue that has not been sent in the current round is added If the value does not exceed Q, send immediately; otherwise, postpone the frame to the next round. The higher the value, the larger the amount of data that can be sent per round, thus enabling differentiated bandwidth allocation based on the reserved ratio.

[0059] In addition, the node also based on Adjust the scheduling priority of the send queue: when When the value is greater than 0.5, the node marks the share packet as high priority and sends it in the local queue before other ordinary data packets.

[0060] The bandwidth allocation mechanism described above directly affects the actual transmission capacity of a node; the longer a node stays in a high-priority area, the higher its transmission capacity. Larger nodes have more opportunities to send data, thus ensuring timely transmission of critical data even during congestion.

[0061] Assembly and output of share packets: Finally, the node will send the share to be sent (original value or difference) and the compression flag. Window type identifier and the current compression ratio Pack them into a single data packet; the packaging format is as follows: The first byte is the control field, where the highest bit (bit7) is the compression flag, 1 indicates the full share, and 0 indicates the difference share; the second highest bit (bit6) is the window type identifier, 1 indicates a fast window, and 0 indicates a normal window; the remaining 6 bits (bit5 to bit0) are reserved as 0; the second byte is the compression ratio field, storing the current compression ratio. The integer value (range 2 to 12); the following bytes carry the share data: if the compression flag is 1 (complete share), the length of this field is 16 bytes, storing the binary representation of the original secret shared share (big-endian, range 0 to 65535); if the compression flag is 0 (difference share), the length of this field is 2 bytes, storing the 8-bit signed difference binary two's complement representation (range -128 to 127); this data packet will be sent to the server as the output of step S6 through the MPC secure channel.

[0062] Through step S5, the node transforms the higher-level adaptive decisions (compression ratio, reservation ratio, and dithering window) into specific share compression, flag generation, and bandwidth allocation behaviors, thus realizing the closed-loop implementation of the transmission strategy. All processing is based on local computation, without leaking the original data information, which meets the requirements of the privacy computing framework.

[0063] Step S6: Send the compressed share packet, along with the compression flag, window type identifier, and current compression ratio, to the server.

[0064] After completing the compression of the share packet, flag generation, and bandwidth allocation preparation in step S5, the participating node executes step S6, sending the compressed share packet along with the compression flag, window type identifier, and current compression ratio to the server. This step is the final execution stage of the adaptive transmission strategy, ensuring that the data can be reliably and efficiently delivered to the server for secret sharing reconstruction according to the results of the previous decisions.

[0065] Data packet format for transmission: The node first assembles the fields to be sent into a complete transmission frame according to a predefined protocol format; the frame structure is defined as follows: Frame header (4 bytes): contains round sequence number (2 bytes, value range 1 to 65535), node identifier (1 byte, supports up to 256 participants) and version number (1 byte, for protocol compatibility); Control field (1 byte): The highest bit (bit7) is the compression flag, 1 indicates the full share and 0 indicates the difference share; the second highest bit (bit6) is the window type identifier, 1 indicates the fast window and 0 indicates the normal window; the remaining 6 bits (bit5 to bit0) are reserved as 0; Compression ratio field (1 byte): Stores the current compression ratio. The integer value, ranging from 2 to 12, is stored directly in binary form; Share data field (variable length): If the compression flag is 1 (complete share), the field length is 16 bytes, storing the binary representation of the original secret shared share (big-endian, range 0 to 65535, with the high 14 bits padded with 0); if the compression flag is 0 (difference share), the field length is 2 bytes, storing the 8-bit signed difference binary two's complement representation (range -128 to 127). Cases exceeding the range have been processed in step S5 to send the complete share, so the difference share here must be within the encoding range.

[0066] For example, a node in round t=100, node ID=5, compression ratio =8, the share to be sent is the difference +8, the window type is fast window, then the assembled sending frame is: frame header round number 0x0064 (100), node ID=0x05, version number 0x01; control field bit7=0 (difference), bit6=1 (fast window), the rest are 0, resulting in 0x40; compression ratio field 0x08; share data field is 0x0008 (2 bytes represent +8); total length is 4+1+1+2=8 bytes.

[0067] Selection of transmission channel and secure transmission: An MPC secure channel is established between the node and the server. This channel is encrypted based on the TLS 1.3 protocol to ensure that the share data is not eavesdropped or tampered with during transmission. The node selects different Quality of Service (QoS) priorities according to the window type identifier: If the window type identifier is a fast window (1), the node marks the sent frame as high priority (e.g., setting the DSCP field in the IP header to EF (accelerated forwarding)); if it is a normal window (0), it is marked as best-effort (BE). The network switch or router provides lower queuing latency for high priority data according to the DSCP mark. This design enables urgent shares (high overall priority) to reach the server faster, thereby meeting the needs of real-time traffic control.

[0068] Local state update after transmission: After successfully transmitting a frame, the node performs the following local state update operations: Cache the shares already sent in the previous round. Update to the original share sent in the current round (Note: regardless of whether a full share or a difference share is sent, Always store the original secret share of the current round (Instead of the difference, to ensure the accuracy of the benchmark for the next round of difference calculation). Current compression ratio Updated to This is used for iterative calculation in the next round, step S4; The round counter is incremented by t=t+1 to prepare for the next round of data collection and processing; If the transmission fails (e.g., due to insufficient tokens, the node does not update). and The node records a transmission failure event. When three consecutive failures occur, the node triggers a local alarm and falls back to conservative mode (fixed compression ratio of 8, fixed anti-jitter window of 3 rounds, resource reservation ratio of 0) until synchronization with the server is re-established. The retransmission queue adopts the first-in-first-out principle and has a maximum length of 100 frames. When the queue exceeds this length, the oldest frame is discarded and the frame loss event is recorded.

[0069] In step S6, the node reliably sends the share packets, which have been adaptively compressed and prioritized, to the server. At the same time, it uses a token bucket controlled by the resource reservation ratio to achieve differentiated bandwidth allocation. The entire process does not rely on server feedback and is driven entirely by the node's local decision-making, ensuring the autonomy of each participant and low-latency response under the privacy computing framework.

[0070] Interaction with the server (supplementary explanation): Although this method focuses on the participating node side, in order to fully understand the data flow, the corresponding behavior on the server side is briefly described: After receiving the transmission frame, the server first parses the control field to obtain the compression flag and window type identifier, and distributes the frame to the corresponding window queue (fast window or normal window) according to the window type identifier; then it restores the original share according to the compression flag and decompression logic; finally, it performs secret sharing reconstruction; the server will also send an acknowledgment message (ACK) to the node according to the reception status. The node can adjust the retransmission behavior according to the ACK, but ACK processing is not a necessary part of this step.

[0071] Based on Embodiment 1, the beneficial effects of the present invention can be summarized as follows: By constructing a three-dimensional state coordinate system composed of data fluctuation level, data urgency level, and inherent node attributes, and discretizing the continuous space into finite region identifiers, the system can quantify the transmission state of participating nodes in real time without relying on server feedback; furthermore, by extracting trend features such as the number of consecutive stay rounds, the degree of speed of region transitions, and the swaying frequency of the directional chain based on the region identifier sequence, nodes are able to identify their own state change trends: the number of consecutive stay rounds reflects regional stability, the degree of speed of transitions reveals the acceleration of state switching, and the swaying frequency distinguishes between real traffic evolution and data noise jitter. Based on this, the compression ratio, resource reservation ratio, and anti-jitter window size of the secret sharing share are dynamically adjusted to achieve adaptive matching between compression intensity and data mutation degree, as well as bandwidth pre-allocation of high-priority data under congestion critical state. Finally, through the coordination of differential compression and window type identification, the transmission load is reduced while ensuring low-latency delivery of key shares. Thus, under the dual critical conditions of communication resource shortage and drastic traffic changes, each participating node can autonomously complete the smooth degradation and recovery of transmission strategy, avoiding data congestion, reconstruction error accumulation, and control command lag, and significantly improving the real-time performance and reliability of multi-source traffic data under the privacy computing fusion framework.

[0072] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0073] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0074] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for analyzing multi-source traffic data based on privacy-preserving computation fusion, characterized in that, Applied to participating nodes, including: Step S1: Collect local multi-source traffic data and calculate the three-dimensional state coordinates of the current cycle based on the multi-source traffic data: a first coordinate component representing the degree of data fluctuation, a second coordinate component representing the degree of data urgency, and a third coordinate component representing the inherent attributes of the node. Step S2: Map the three-dimensional state coordinates to a sub-cube in a preset three-dimensional feature space, and determine the region identifier of the current round based on the mapping result; Step S3: Based on the current round and the regional identifier sequence of several historical rounds, calculate at least one trend feature, which includes the number of consecutive stay rounds, the degree of regional transition speed, and the directional chain swing frequency; Step S4: Based on the trend characteristics and the current region identifier, dynamically calculate the resource reservation ratio, dynamically adjust the compression ratio of the secret sharing share, and dynamically determine the anti-shake window size; Step S5: Perform differential compression on the secret shared share to be sent according to the compression ratio to generate a compressed share packet, and generate a compression flag to identify whether the current share is a differential share or a complete share, a window type flag to identify the window type to which the current share belongs, and allocate transmission bandwidth according to the resource reservation ratio. Step S6: Send the compressed share packet, along with the compression flag, window type identifier, and current compression ratio, to the server.

2. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, Calculating the three-dimensional state coordinates of the current round specifically includes: First coordinate component calculation: Extract speed values ​​of a consecutive preset number of rounds from the multi-source traffic data to form a speed sequence; count the number of sign flips in the direction of speed change of adjacent rounds in the speed sequence, and calculate the normalized value after dividing the range of the maximum and minimum values ​​in the speed sequence by a preset speed upper limit; perform a weighted summation of the number of sign flips and the normalized value to obtain the first coordinate component representing the degree of data fluctuation. Second coordinate component calculation: Obtain the locally stored value score and sensitivity score, and calculate the time-effect decay factor based on the time-effect round number of the current round of data. Multiply the weighted sum of the value score and sensitivity score by the time-effect decay factor to obtain the second coordinate component representing the urgency of the data. Calculation of the third coordinate component: Obtain the node type coefficient and the time period coefficient, and perform a weighted summation of the node type coefficient and the time period coefficient to obtain the third coordinate component representing the inherent attributes of the node.

3. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, Step S2 specifically includes: Each coordinate component in the three-dimensional state coordinates is compared with a first preset grading threshold and a second preset grading threshold. The first preset grading threshold is less than the second preset grading threshold. When the coordinate component is less than the first preset grading threshold, it is determined to be low grade. When the coordinate component is between the first preset grading threshold and the second preset grading threshold, it is determined to be medium grade. When the coordinate component is greater than the second preset grading threshold, it is determined to be high grade. Based on the gear combination of the three coordinate components, determine the sub-cube number to which the current round belongs; Using the sub-cube number as an index, query the pre-stored region mapping table. Each sub-cube number in the region mapping table corresponds to a preset region identifier, which is selected from one of multiple non-overlapping regions.

4. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, The number of consecutive stops is calculated as follows: Get the region identifier of the current round and the region identifier of the previous round; If the region identifier of the current round is the same as that of the previous round, the historical consecutive counter is incremented by 1; otherwise, the historical consecutive counter is reset to 1. The current value of the historical continuous counter is output as the number of consecutive stops.

5. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, The method for calculating the degree of urgency of regional transitions is as follows: Map multiple preset region identifiers to consecutive integer values; Obtain the region identifier integer values ​​for three consecutive rounds, including the current round, and calculate the first absolute difference between the first and second rounds, and the second absolute difference between the second and third rounds in sequence; Calculate the first difference between the first absolute difference and the second absolute difference, and the second difference between the second absolute difference and the first absolute difference. Take the absolute values ​​of the first difference and the second difference and then calculate the average value to obtain the degree of abruptness of the regional transition. When a participating node is in the startup phase and has less than three historical rounds, the speed of the regional transition is set to a preset default value.

6. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, The method for calculating the oscillation frequency of the directional chain is as follows: Get the direction chain triplet for three consecutive rounds including the current round. Each direction chain triplet records the sign of the change in the current round relative to the previous round in three dimensions. The sign of the change in each dimension can be positive, negative or zero. For each dimension, extract three consecutive changing symbols to form a symbol sequence, and count the number of times positive and negative signs change between adjacent symbols in the symbol sequence, excluding changes involving zero. The total number of transformations is obtained by summing the number of transformations in the three dimensions; the total number of transformations is then divided by a preset normalization factor to obtain the directional chain oscillation frequency.

7. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, The compression ratio for dynamically adjusting the secret sharing share specifically includes: Obtain the degree of regional transition speed in the current round and several historical rounds, and determine the adjustment range of the compression ratio based on the comparison between the median of the historical regional transition speed and the current regional transition speed. Obtain the change step size of the current round and the change step size of the previous round, and determine the adjustment direction based on the comparison results; The current compression ratio is updated according to the adjustment direction and adjustment range, and the updated compression ratio is limited to a preset compression ratio range.

8. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 6, characterized in that, Dynamically determining the size of the stabilization window specifically includes: The directional chain swing frequency is mapped to a preset window value range, and the size of the anti-shake window increases as the directional chain swing frequency increases in the mapping relationship. The mapped value is used as the final debouncing window size.

9. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, The dynamic computing resource reservation ratio specifically includes: The target reservation ratio is determined based on whether the current area identifier belongs to the preset high-priority area set and the number of consecutive stay rounds; The previous resource reservation ratio is approximated to the target reservation ratio using an exponential smoothing method to obtain the current resource reservation ratio; Based on the prediction results of regional change direction, the current resource reservation ratio is adjusted in a forward-looking manner.

10. The method for analyzing multi-source traffic data based on privacy-preserving computation fusion according to claim 1, characterized in that, The specific process of performing differential compression and generating compression flags includes: Determine if the current round is the starting round; if so, use the complete share as the share to be sent and set the compression flag to the first value. If it is not the starting round, calculate the difference between the current share and the share sent in the previous round, and determine the maximum allowable deviation range of the sending difference based on the current compression ratio; If the difference is within the maximum deviation range, then the difference is used as the share to be sent and the compression flag is set to the second value; otherwise, the complete share is used as the share to be sent and the compression flag is set to the first value.

Images