Toll station cloud warehouse management scheduling method, system and equipment based on tide prediction

Abstract

The present application relates to the technical field of highway traffic operation management, and particularly relates to a toll station cloud warehouse management scheduling method, system and equipment based on tidal prediction. The present application accurately captures traffic flow fluctuations by using a tidal prediction model, and converts traffic flow data into a quantitative congestion level by using a congestion prediction model, thereby realizing scientific mapping from traffic flow to a schedulable state. In combination with the real-time state and prediction level of the cloud warehouse, a corresponding resource allocation strategy is generated to ensure accurate resource deployment and globally optimal configuration. Finally, through a closed-loop feedback mechanism, the strategy is dynamically adjusted according to the processing result to form a self-adaptive and optimized scheduling closed loop. Thus, the present method realizes early prediction of the congestion risk of the toll station and dynamic and accurate scheduling of the cloud warehouse resources, significantly improves the traffic efficiency and resource utilization, and effectively solves the core problems of insufficient prediction accuracy, disconnection between evaluation and scheduling, and response lag in the prior art.

Description

Technical Field

[0001] This invention relates to the field of highway traffic operation management technology, and in particular to a toll station cloud warehouse management and scheduling method, system and equipment based on tidal prediction. Background Technology

[0002] As bottleneck nodes in the road network, highway toll stations directly impact the overall network's operational efficiency. To address the ever-increasing traffic demand, the industry has developed various toll station management and traffic flow prediction technologies, but many areas still require optimization.

[0003] Currently, various technical solutions have been developed for the management of highway toll stations. In terms of traffic flow prediction, existing technologies such as CN116386020B improve prediction accuracy through multi-source data fusion, but this solution does not fully combine traffic flow tidal characteristics and path features, and the prediction accuracy during peak hours is still insufficient.

[0004] In terms of monitoring and emergency response, existing technologies such as CN118824017B analyze lane pressure and generate emergency plans through digital twin technology, but do not integrate cloud warehouse resource scheduling mechanisms, making it impossible to quickly allocate resources across regions when special situations occur; CN114937312B optimizes the toll collection process, but does not design a dynamic congestion response mechanism for tidal traffic flow; CN116597649B realizes road network traffic projection, but similarly does not link it to resource scheduling, and cannot provide decision support for congestion scenarios.

[0005] In summary, existing technologies generally suffer from the following shortcomings: First, traffic flow prediction lacks adaptation to tidal characteristics, resulting in insufficient accuracy during peak hours; second, congestion assessment and resource scheduling mechanisms are disconnected, failing to establish a linkage between "congestion level" and "scheduling strategy"; third, the cloud warehouse management system is not integrated, making it impossible to quickly allocate resources across regions in special circumstances; and fourth, the architecture design is not optimized for real-time performance, making it difficult to meet minute-level scheduling requirements. The core reason for this lies in the fact that existing technologies have not formed a closed-loop logic for the entire process of "prediction-assessment-scheduling."

[0006] Therefore, there is a need for a toll station cloud warehouse management and scheduling method, system, and equipment that can overcome the problems of insufficient prediction accuracy and the disconnect between congestion assessment and scheduling. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of insufficient prediction accuracy and the disconnect between congestion assessment and scheduling in the existing technology, and to provide a toll station cloud warehouse management and scheduling method, system and equipment based on tidal prediction.

[0008] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0009] A toll station cloud warehouse management and scheduling method based on tidal prediction, characterized by the following steps: S1: Obtain real-time toll station data and perform preprocessing; S2: Input the real-time toll station data into the pre-trained traffic flow tidal prediction model to generate the predicted traffic flow data for the corresponding toll station after a set time. S3: Input the predicted traffic flow data into the pre-trained congestion prediction model to generate the predicted congestion level for the corresponding toll station after a set time. S4: Generate a resource allocation strategy based on the current cloud warehouse status and the predicted congestion level; the resource allocation strategy includes a single toll station congestion scheduling strategy and a multi-toll station congestion scheduling strategy; S5: The cloud warehouse performs special handling according to the resource allocation strategy and enters S1 according to the preset trigger conditions.

[0010] As a preferred embodiment of the present invention, step S1 includes the following steps: S11: Obtain real-time toll station data for the area to be dispatched; The real-time toll station data includes real-time demand data, periodic data, and static data; The real-time data requirements include gantry data, real-time weather, and traffic control data. The periodic data includes toll record data and ETC / CPC data; The static data includes road segment information and holiday schedule data; S12: Preprocess the real-time toll station data; the preprocessing includes data cleaning and standardization.

[0011] As a preferred embodiment of the present invention, the data cleaning includes missing value processing, outlier processing, and format standardization processing; The missing value handling includes the following steps: Get the percentage of missing data in the current data; When the missing percentage is less than the first missing threshold, linear interpolation is used to process the current data. When the first missing threshold ≤ missing proportion < second missing threshold; KNN interpolation is used, and the current data is weighted and imputed according to the principles of temporal proximity and feature similarity. When the second missing threshold is less than or equal to the missing proportion; if the current data is a non-core feature as defined, the current data is deleted directly; if the current data is a core feature as defined, a random forest model is used to predict and impute it. The outlier handling includes the following steps: Outlier detection is performed on each data point sequentially; if the current data is numerical, the interquartile range method is used for detection; if the current data is high-dimensional, the Isolation Forest algorithm is used for detection. Determine the type of detected outlier; if the current outlier is a device malfunction, use interpolation correction; if the current outlier is a genuine outlier caused by extreme weather and / or holidays, retain the current outlier and mark it as an outlier. The standardization includes data standardization processing or data normalization processing; The data standardization process is used for data that follows a normal distribution, and its expression is as follows: , in, The data is standardized, x is the input data, μ is the mean of the training set, and σ is the standard deviation of the training set. The data normalization process is used for non-normally distributed data, and its processing expression is as follows: , Where, x min x is the minimum value of the current data type. max This represents the maximum value of the current data type.

[0012] As a preferred embodiment of the present invention, the prediction results of the traffic flow tidal prediction model include: Entrance traffic flow : , Where τ is the instantaneous time point, for Instantaneous traffic flow at the entrance. For prediction time granularity; Export traffic volume : , in, for Instantaneous traffic flow at the exit; Tidal peak flow : , in, ∈ or , This is a collection of tidal periods.

[0013] As a preferred embodiment of the present invention, the congestion prediction model adopts the XGBoost model, and uses grid search and 5-fold cross-validation to iteratively fine-tune the hyperparameters; during the training process, the congestion prediction model uses the SMOTE algorithm to handle sample imbalance, and removes redundant features with weights below the threshold based on the built-in feature importance score results.

[0014] As a preferred embodiment of the present invention, the congestion level is evaluated using a comprehensive congestion index; its expression is: , Among them, w i Assign weights to the i-th type of indicator. is the normalized value of the i-th type of index; The evaluation indicators for the congestion level include: Average travel time for vehicles : , in, is the passage time for a single vehicle, and n is the total number of vehicles in the statistical period; Lane utilization : , in, The lane occupancy time for a single vehicle is T, the total time of the statistical period is m, and the total number of lanes is m. Average queue length : , Where L is the average vehicle length; Vehicle traffic delay rate : , in, The average travel time for free flow. For free circulation time.

[0015] As a preferred embodiment of the present invention, the single toll station congestion scheduling strategy is applicable to congestion at a single toll station and includes the following steps: Quantitative decision-making factors for determining resource allocation strategies; The quantitative decision factors for the resource allocation strategy include: cloud agent configuration: including the number of agents and average skill scores; cloud warehouse busyness: including the number of special cases currently being handled and the percentage of idle agents; network latency: including the average data transmission latency between the cloud warehouse and the toll station; historical processing records: including the percentage of times the target toll station was processed, the average processing time, and the processing accuracy. Construct a priority scoring function: , in, The priority score for the j-th cloud warehouse. Let the initial weight of the k-th factor be... Let be the standardized value of the j-th cloud warehouse for the k-th factor; Adjust scheduling rules: When the current toll station is experiencing mild congestion: increase the weighting of cloud warehouse busy level and network latency; When the current toll station is under moderate congestion: increase the weight of cloud seat configuration and historical processing records; When the current toll station is severely congested: increase the weight of cloud warehouse busy level and historical processing records; Calculate the priority score for each cloud warehouse based on the adjusted weights, and select... The highest-level cloud warehouse manages the current toll stations.

[0016] As a preferred embodiment of the present invention, the multi-toll station congestion scheduling strategy is applicable to situations where congestion occurs at more than one toll station, and its cloud warehouse management scheduling function is: Establish a multi-objective optimization model: , Constraints: , Where m represents the number of congested toll booths, and n represents the number of cloud warehouses. The sum of network latency and processing time when cloud warehouse j processes toll station i. In order to handle the revenue, The number of available seats in cloud warehouse j is multiplied by a set coefficient. To assign identifiers, =1 indicates allocation. =0 means no allocation.

[0017] As a preferred embodiment of the present invention, the preset triggering conditions include: setting an update time, the cloud warehouse processing efficiency decreasing beyond a set risk threshold, and the actual congestion level at the toll station being higher than the predicted congestion level.

[0018] A toll station cloud warehouse management and scheduling system based on tidal prediction, characterized in that the system is used to execute the toll station cloud warehouse management and scheduling method based on tidal prediction as described above; it includes a cloud warehouse platform and several edge nodes that are communicatively connected to the cloud warehouse platform; The edge nodes are deployed locally at each toll station to acquire real-time toll station data and perform preprocessing. The cloud warehouse platform is deployed at the traffic control center and is used to generate resource allocation strategies based on the real-time toll station data.

[0019] A toll station cloud warehouse management and scheduling device based on tidal prediction includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute any of the above-described toll station cloud warehouse management and scheduling methods based on tidal prediction.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: This application ensures data quality by acquiring and preprocessing real-time data, laying the foundation for accurate prediction. It then utilizes a tidal prediction model to accurately capture traffic flow fluctuations, outputting highly reliable future traffic volumes. Next, a congestion prediction model transforms traffic flow data into quantified congestion levels, achieving a scientific mapping from traffic volume to schedulable status. Combining real-time cloud warehouse status and prediction levels, priority scoring or multi-objective optimized resource allocation strategies are generated for single-station or multi-station congestion, ensuring precise resource allocation and globally optimal configuration. Finally, a closed-loop feedback mechanism dynamically adjusts the strategy based on the processing results, forming an adaptive and optimized scheduling loop. Therefore, this method achieves early prediction of toll station congestion risks and dynamic and accurate scheduling of cloud warehouse resources, significantly improving traffic efficiency and resource utilization, and effectively solving core problems in existing technologies such as insufficient prediction accuracy, disconnect between assessment and scheduling, and delayed response. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a toll station cloud warehouse management and scheduling method based on tidal prediction as described in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the structure of a toll station cloud warehouse management and scheduling system based on tidal prediction as described in Embodiment 3 of the present invention; Figure 3 This is a schematic diagram of the structure of a toll station cloud warehouse management and scheduling device based on tidal prediction as described in Embodiment 4 of the present invention. Detailed Implementation

[0022] The present invention will be further described in detail below with reference to experimental examples and specific embodiments. However, this should not be construed as limiting the scope of the above-mentioned subject matter of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0023] Example 1 like Figure 1 As shown, a toll station cloud warehouse management and scheduling method based on tidal prediction includes the following steps: S1: Obtain real-time toll station data and perform preprocessing.

[0024] S2: Input the real-time toll station data into the pre-trained traffic flow tidal prediction model to generate the predicted traffic flow data for the corresponding toll station after a set time.

[0025] S3: Input the predicted traffic flow data into the pre-trained congestion prediction model to generate the predicted congestion level for the corresponding toll station after a set time.

[0026] S4: Generate a resource allocation strategy based on the current cloud warehouse status and the predicted congestion level; the resource allocation strategy includes a single toll station congestion scheduling strategy and a multi-toll station congestion scheduling strategy.

[0027] S5: The cloud warehouse performs special handling according to the resource allocation strategy and enters S1 according to the preset trigger conditions.

[0028] Example 2 This embodiment is a specific implementation of the toll station cloud warehouse management and scheduling method based on tidal prediction described in Embodiment 1, including the following steps: S1: Obtain real-time toll station data and perform preprocessing.

[0029] Furthermore, this embodiment, based on the actual needs of toll station scheduling, defines the core prediction indicators as entrance traffic flow, exit traffic flow, and peak traffic flow during tidal periods; specifically, the quantitative prediction results of the traffic flow tidal prediction model include: Entrance traffic flow : , Where τ is the instantaneous time point, for Instantaneous traffic flow at the time entrance (unit: pcu / s). The prediction time granularity is set to 15 minutes in this embodiment, in conjunction with the scheduling response time.

[0030] Export traffic volume : , in, for Instantaneous traffic flow at the exit (unit: pcu / s).

[0031] Tidal peak flow : , in, ∈ or , This is a set of tidal periods (in this embodiment, the set is based on historical data statistics, covering the morning peak from 7:00 to 9:00, the evening peak from 17:00 to 19:00, and the peak travel / return period during holidays).

[0032] Furthermore, this embodiment combines the actual operation of toll stations to screen and quantify the influencing factors in a scenario-based manner.

[0033] I. Factors Affecting Export Vehicle Traffic (1) Characteristics of highway sections (F1): Quantified as design speed (km / h), number of lanes, type of core scene along the route (industrial park / commercial area / tourist area, using one-hot encoding), and distance between entrances and exits (km), which directly affect the spatiotemporal distribution of traffic demand.

[0034] (2) ETC / CPC card passage data (F2): Extract the OD matrix, route selection probability, and passage frequency (times / month) for the past 3 months to accurately match the source and destination of traffic flow.

[0035] (3) Gantry system monitoring data (F3): Select the vehicle passage time and vehicle type ratio of the gantry 5km, 10km and 20km upstream of the toll station, and estimate the time distribution of vehicles arriving at the exit by the road segment travel time.

[0036] (4) Path identification information (F4): Quantify the path allocation ratio (%) of different OD pairs to adapt to the traffic diversion characteristics under complex road networks.

[0037] (5) Distribution of entrance toll stations (F5): Statistically count the number of entrance stations (units) and average distance (km) within the service area of ​​the exit, and match the scale of traffic supply with the time distribution.

[0038] (6) Historical traffic volume (F6): Extract the average, peak and fluctuation coefficient of traffic volume in the same period of the past year (same day, same time period, same holiday type) to construct the time series benchmark features.

[0039] (7) Real-time weather (F7): quantified as precipitation (mm), visibility (km), and weather type (sunny / rainy / fog / snow, using numerical coding), adapting the impact of weather on travel intentions and driving speed.

[0040] II. Factors Affecting Entrance Traffic Flow (1) Historical traffic volume scale (F6′): Extract the intraday distribution, intraweek differences and seasonal fluctuation characteristics of the same period of the past year to construct a time-dependent benchmark.

[0041] (2) Real-time weather (F7′): Adopting the same quantitative standard as the exit traffic flow, focusing on adapting to the direct impact of weather on potential travel demand.

[0042] Specifically, S1 includes the following steps: S11: Obtain real-time toll station data for the area to be scheduled. The real-time toll station data includes real-time demand data, periodic data, and static data.

[0043] The real-time data requirements include gantry data, real-time weather, and traffic control data. In this embodiment, gantry data is collected in seconds, and meteorological and traffic control information is collected in hours. 5G + fiber optic dual-link transmission is used to ensure a latency of ≤5ms.

[0044] The periodic data includes toll record data and ETC / CPC data; in this embodiment, data is collected in real time according to events, and summarized into 15-minute-level structured data daily, which is then transmitted in batches using the FTP protocol.

[0045] The static data includes road segment information and holiday schedule data: in this embodiment, it is updated quarterly / annually and synchronized to the prediction system database via the intranet.

[0046] Specifically, the data source and scenario adaptation table for the real-time toll station data described in this embodiment is shown below.

[0047] Table 1 Data Sources and Scenario Adaptation Table Data types source Core collection fields Scene adaptation instructions Historical passage records of toll stations Toll collection system database Entry station number, exit station number, passage time, vehicle type, ETC / CPC card number Covering one year of data ensures the capture of seasonal tidal characteristics. Gantry system data Gantry monitoring equipment along the line Passage time, license plate, vehicle type, and direction of travel Focusing on key upstream sections (5 / 10 / 20km), data is collected in seconds. ETC / CPC card data Provincial management platform Travel route, transaction time, vehicle type Covering 3 years of data to support long-term travel characteristic analysis Meteorological data Local meteorological observation stations Real-time precipitation, temperature, visibility, and weather type Updated hourly, suitable for short-term forecasting needs. Basic information of road section GIS database of highway management department Design speed, number of lanes, roadside features, and distance between entrances and exits. Quarterly updates to ensure consistency with actual road conditions. Auxiliary data Government Affairs Platform / Traffic Control System Holiday schedule, road occupancy notices, number of motor vehicles Annual / monthly updates explaining abnormal traffic fluctuations. S12: Preprocess the real-time toll station data; the preprocessing includes data cleaning, standardization, and feature engineering. (1) Data cleaning: The data cleaning includes missing value processing, outlier processing and format standardization processing; ① Handling missing values: Get the percentage of missing data in the current data; When the missing percentage is less than the first missing threshold, linear interpolation is used to process the current data. When the first missing threshold ≤ missing proportion < second missing threshold; KNN interpolation is used, and the current data is weighted and imputed according to the principles of temporal proximity and feature similarity. When the second missing threshold is less than or equal to the missing proportion; if the current data is a non-core feature as defined, the current data is deleted directly; if the current data is a core feature as defined, a random forest model is used to predict and impute it.

[0048] Furthermore, this embodiment uses a first missing threshold of 5% and a second missing threshold of 20% as an example: Minor missing values ​​(<5%): Linear interpolation method is used, formula is as follows It is suitable for continuous data such as traffic flow and temperature. For the i-th data, Let i be the time point corresponding to the i-th data.

[0049] Moderate missing data (5% ≤ proportion < 20%): KNN interpolation (K=5) is used to fill in the missing data based on the principle of "temporal proximity + feature similarity" and to adapt to related data such as gantry traffic and passage time.

[0050] Large number of missing features (≥20%): If they are non-core features (such as some auxiliary data), they are deleted directly; if they are core features (such as historical traffic flow), they are filled using a random forest model. The input features include correlation factors such as time, weather, and road segment characteristics.

[0051] ② Outlier handling: Outlier detection is performed on each data point in turn; If the current data is numerical, use the interquartile range method (...). ) to be tested; among them, The first quartile is denoted by IQR, which stands for interquartile range. It is the third quartile.

[0052] If the current data is high-dimensional (such as multi-gantry joint data), the Isolation Forest algorithm is used for detection (contamination=0.05 in this embodiment).

[0053] Determine the type of detected outlier; if the current outlier is a device malfunction (i.e., an erroneous outlier caused by device malfunction, such as a flow rate exceeding a reasonable range), use interpolation correction; if the current outlier is a real outlier caused by extreme weather and / or holidays (such as peak flow), retain the current outlier and mark it as an outlier (e.g., mark: 0 = normal, 1 = outlier).

[0054] ③ Format standardization: The format standardization described in this embodiment includes: The timestamps are standardized to “YYYY-MM-DD HH:MM:SS”, the traffic flow unit is standardized to “pcu / 15min”, and the vehicle speed unit is standardized to “km / h”.

[0055] Correct logical errors (such as exit time being earlier than entry time, or mismatch between vehicle type and toll amount) and perform cross-validation corrections based on toll standards and traffic rules.

[0056] (2) Standardization: The standardization includes data standardization processing or data normalization processing; Z-Score Standardization: The data standardization process is used for data that follows a normal distribution (such as historical traffic flow averages and temperatures). The expression for this process is: , in, The data is standardized, where x is the input data, μ is the mean of the training set, and σ is the standard deviation of the training set, to avoid data leakage.

[0057] Normalization (Min-Max): The data normalization process is used for non-normally distributed data (such as wind speed and lane utilization). The processing expression is as follows: , Where, x min x is the minimum value of the current data type. max This is the maximum value of the current data type. By normalizing, the data is mapped to the [0,1] interval, which improves the convergence speed of the model.

[0058] S2: Input the real-time toll station data into the pre-trained traffic flow tidal prediction model to generate the predicted traffic flow data for the corresponding toll station after a set time.

[0059] Traffic flow tidal prediction is the core prerequisite for dynamic scheduling. Its core objective is to accurately output the inlet and outlet traffic flow (including tidal peaks) based on multi-source data, providing highly reliable data support for subsequent congestion prediction and scheduling decisions.

[0060] Furthermore, the traffic flow tidal prediction model described in this embodiment adopts the Attention-LSTM model, which adapts to the long-term dependence and tidal fluctuation characteristics of traffic flow time series, and focuses on optimizing the input-output adaptation of the model in the toll station scenario.

[0061] Furthermore, the model training and attributes in this embodiment are as follows: Input features: 20 core features including time features, traffic flow features, and interaction features are selected and constructed into a 64-step input sequence (corresponding to 16 hours of historical data) at a 15-minute granularity.

[0062] Output layer: Single output node, directly outputs the traffic flow prediction value (pcu / 15min) for the next 15 minutes, adapting to the time granularity requirements of scheduling decisions.

[0063] Dataset partitioning: In this embodiment, the dataset is divided into training set (70%), validation set (15%), and test set (15%) in chronological order, spanning from January 1, 2023 to December 31, 2023, to ensure the temporal continuity of the data and avoid feature leakage caused by random partitioning.

[0064] Hyperparameters: Initial hyperparameter settings and scene adaptation are shown in the table below: Table 2 Initial Hyperparameter Settings and Scene Adaptation Table Hyperparameter type Parameter name initial value Scene adaptation instructions Network Structure Input sequence length 64 Covering 16 hours of historical data, capturing tidal cycles Network Structure Number of cells in an LSTM layer [256, 128, 64] A 3-layer structure that balances feature extraction capability and computational efficiency. Network Structure Number of attention heads 8 Strengthen the feature weights of key historical periods (such as previous peaks). Network Structure Dropout rate 0.2 Suppress overfitting and adapt to the fluctuating characteristics of traffic data Training parameters Optimizer Adam Fast convergence speed, suitable for batch training needs Training parameters Initial learning rate 0.001 Balancing convergence speed and accuracy Training parameters Batch size 64 Adapt to GPU computing power to ensure training efficiency Training parameters Number of training rounds 100 Combined with an early stopping strategy, overfitting can be avoided. Hyperparameter tuning: This embodiment uses the Bayesian optimization method to iteratively tune the hyperparameters of the toll station scenario. The core steps are as follows: (1) Define the search space: input sequence length [32, 64, 128], number of LSTM units [[128,64,32],[256,128,64], [512,256,128]], dropout rate [0.1, 0.2, 0.3, 0.4], learning rate [0.0001,0.001, 0.01], batch size [32, 64, 128].

[0065] (2) Proxy model and acquisition function: A Gaussian process is used as the proxy model, and the expected improvement function is (EI(x)=E[max(f(x)]). f(x*),0)]) guides the selection of hyperparameters, where x* is the current optimal combination of hyperparameters.

[0066] (3) Iterative optimization: Initially, 5 sets of hyperparameters were selected for training, and the process was iterated 30 times. The goal was to minimize the MAPE of the validation set. Finally, the optimal combination of hyperparameters was determined: input sequence length 64, number of LSTM units [256,128,64], dropout rate 0.2, learning rate 0.001, and batch size 64.

[0067] Furthermore, the traffic flow tidal prediction model also includes model validation and scenario application effect evaluation.

[0068] (1) Evaluation indicators This embodiment focuses on practical application value and uses four indicators: RMSE (Root Mean Squared Error), MAE (Mean Absolute Deviation), MAPE (Mean Absolute Percentage Error), and R2 (R-Squared Coefficient of Determination). Among them, MAPE is the core evaluation indicator (directly reflecting relative error and adapting to the accuracy requirements of scheduling decisions), and its expression is as follows:

[0069]

[0070]

[0071]

[0072] in, Let i be the true value of the i-th sample. Let be the predicted value of the i-th sample, and N be the total number of samples.

[0073] 1.3.3.2 Verification Results and Scenario Adaptation Analysis The results of the test set are shown in the table below, and the comparison and validation were performed with ARIMA, Random Forest, and traditional LSTM models.

[0074] Table 3 Comparison and Verification Table Model RMSE (pcu / 15min) MAE (pcu / 15min) MAPE (%) R2 ARIMA 32.68 25.42 17.35 0.708 Random Forest 24.15 18.76 12.89 0.823 Traditional LSTM 18.32 14.51 7.64 0.915 Attention-LSTM (after optimization) 12.56 9.87 5.32 0.962 Contextual analysis: Off-peak hours (non-tidal): MAPE=3.85%, R2=0.978, suitable for accurate forecasting of low-fluctuation flow.

[0075] Peak hours (non-peak): MAPE=5.12%, R2=0.965, suitable for forecasting demand for moderately fluctuating flow rates.

[0076] During peak tidal periods: MAPE=7.89%, R2=0.934. Although the error is slightly high, it meets the threshold requirement for congestion prediction (MAPE<10%) and can effectively support scheduling decisions.

[0077] This invention overcomes the limitations of existing technologies that rely on single data sources and lack consideration of tidal characteristics by integrating multi-source data fusion and scenario-based feature engineering. It integrates multi-dimensional influencing factors such as highway segment characteristics, ETC / CPC traffic data, gantry monitoring information, historical traffic flow, and real-time weather, combined with the long-term dependency capture capability of the Attention-LSTM model and Bayesian optimization, to achieve accurate prediction of entrance and exit traffic flow and tidal peaks. It effectively adapts to both low-fluctuation traffic flow during off-peak hours and complex traffic changes during peak hours and holidays, avoiding the accuracy shortcomings in cross-regional traffic flow prediction. Accurate traffic flow prediction provides highly reliable data support for subsequent congestion assessment, anticipates congestion risks in advance, and allows sufficient preparation time for scheduling decisions, thus improving the foresight of toll station operation and management from the source.

[0078] S3: Input the predicted traffic flow data into the pre-trained congestion prediction model to generate the predicted congestion level for the corresponding toll station after a set time.

[0079] Congestion level prediction is crucial for connecting traffic flow forecasting and cloud-based scheduling. Its core objective is to output the congestion level of toll stations based on multiple influencing factors, providing a direct basis for scheduling strategy formulation. This section focuses on the application and optimization of the model in congestion assessment scenarios.

[0080] Furthermore, the evaluation indicators for the congestion level include: Average travel time for vehicles : , in, The time taken for a single vehicle to pass through is denoted as (seconds / vehicle), and n is the total number of vehicles in the statistical period. ETC and MTC lanes are counted separately.

[0081] Lane utilization : , in, The lane occupancy time for a single vehicle is denoted as T (seconds), the total time of the statistical period is denoted as m (seconds), and the total number of lanes is denoted as m. ETC and MTC lanes are counted separately.

[0082] Average queue length : , Where L is the average vehicle length.

[0083] Vehicle traffic delay rate : , in, The average travel time for free flow. Free passage time (3-5 seconds / vehicle in ETC lane, 5-15 seconds / vehicle in MTC lane).

[0084] Furthermore, the congestion level is evaluated using a comprehensive congestion index; to adapt to the scheduling decision-making requirements, the Analytic Hierarchy Process (AHP) is used to determine the weights of the indicators (average vehicle travel time 0.42, lane utilization 0.18, average queue length 0.25, and traffic delay rate 0.15). Based on operational experience, five congestion levels are divided, as shown in the table below.

[0085] Table 4 Congestion Level Classification Table Congestion Level Average vehicle travel time (seconds / vehicle) Lane utilization Average queue length (meters) Traffic delay rate (%) Comprehensive Congestion Index (CCI) Smooth ETC≤5, MTC≤8 ≤0.3 ≤50 ≤10 0≤CCI≤20 Basically unobstructed 5<ETC≤8,8<MTC≤12 0.3<η≤0.5 50<L≤100 10<Dr≤30 20<CCI≤40 Mild congestion 8<ETC≤12,12<MTC≤20 0.5<η≤0.7 100<L≤200 30<Dr≤50 40<CCI≤60 Moderate congestion 12<ETC≤20,20<MTC≤40 0.7<η≤0.9 200<L≤300 50<Dr≤80 60<CCI≤80 Severe congestion ETC>20, MTC>40 >0.9 >300 >80 CCI>80 Overall Congestion Index: , Among them, w i Assign weights to the i-th type of indicator. is the normalized value of the i-th type of index (mapped to [0,100]).

[0086] Based on the actual operation of toll stations, we identified the core influencing factors and quantified their application: (1) Traffic flow (F10): The prediction result Q using the Attention-LSTM model in (t), Q out (t) and real-time traffic flow data, input in 15-minute increments.

[0087] (2) Lane operation status (F11): quantified as the number of ETC lanes, the number of MTC lanes, the percentage of ETC lanes (%), the number of faulty lanes, and the duration of the fault (minutes).

[0088] (3) Historical data factors (F12): Extract the congestion level, average duration and frequency of occurrence in the same period of the past 3 months to construct historical baseline features.

[0089] (4) Time period characteristics (F13): Quantified into time period type (peak / off-peak / valley, code 0-2), date type (weekday / weekend / holiday, code 0-2), and whether it is a tidal peak period (0 / 1).

[0090] (5) On-site staffing (F14): Number of toll collectors (persons), average years of service (years), number of traffic control personnel (persons).

[0091] (6) Site conditions (F15): Toll plaza area (㎡), entrance ramp length (m), connecting line width (m).

[0092] (7) Real-time weather (F16): Uses the same quantification standard as traffic flow forecast.

[0093] Furthermore, the congestion prediction model described in this embodiment adopts the XGBoost model, which is suitable for congestion prediction scenarios with multi-factor nonlinear coupling, and focuses on optimizing the model's application effect in toll station scenarios. The model training and attributes in this embodiment are as follows: Input features: The above 7 categories of influencing factors have a total of 32 features, and the input samples are constructed at a 15-minute granularity.

[0094] Output results: Congestion level (0 = smooth, 1 = basically smooth, 2 = light congestion, 3 = moderate congestion, 4 = severe congestion), which directly provides a classification basis for scheduling decisions.

[0095] Hyperparameter tuning: In this embodiment, grid search + 5-fold cross-validation is used. Specific examples are shown in the table below.

[0096] Table 5 Hyperparameter Optimization and Scene Adaptation Table Hyperparameters Search scope optimal value Scene adaptation instructions Learning rate [0.01, 0.05, 0.1] 0.05 Balancing convergence speed and generalization ability Tree depth [3, 5, 7, 9] 5 To avoid overfitting, adapt to medium-dimensional features. Number of leaf nodes [10, 20, 30] 20 Balancing model complexity and computational efficiency Regularization coefficient (λ) [0.1, 1, 10] 1 Suppress overfitting and adapt to the fluctuating characteristics of congestion data. Training process optimization Data balancing: The SMOTE algorithm is used to handle sample imbalance (low proportion of severely congested samples) and improve the prediction accuracy of minority class samples.

[0097] Feature importance screening: Based on the built-in feature importance score of XGBoost, redundant features with a weight <0.01 are removed, and finally 28 core features are retained.

[0098] Model Validation and Scenario Application Effects The test set results were evaluated using accuracy, precision, recall, and F1 score, as shown in the table below: Table 6. Test Results of the Congestion Prediction Model Congestion Level Accuracy Recall rate F1 score Application value of scenarios Smooth 0.95 0.93 0.94 Accurately identify low-load conditions to avoid over-scheduling. Basically unobstructed 0.92 0.90 0.91 Adapting to resource maintenance strategies under normal conditions Mild congestion 0.89 0.88 0.88 Core scheduling trigger thresholds allow for timely intervention to alleviate congestion. Moderate congestion 0.87 0.86 0.86 Adaptive resource enhancement scheduling strategy Severe congestion 0.84 0.82 0.83 Adapt to emergency dispatch strategies to prevent congestion from spreading. average 0.89 0.87 0.88 Meet the accuracy requirements of scheduling decisions Scenario-based conclusions: The model has a recall rate of ≥82% for light and severe congestion, which can trigger scheduling strategies in a timely manner; the accuracy rate for smooth traffic reaches 95%, which can avoid the waste of ineffective scheduling resources and fully meet the actual needs of toll station congestion prediction.

[0099] This invention, based on a multi-dimensional evaluation index system and the XGBoost model, constructs a congestion assessment mechanism that breaks through the limitations of existing technologies that rely on a single dimension to judge congestion. It integrates four core indicators—average vehicle travel time, lane utilization, average queue length, and traffic delay rate—with scientific weights determined by the analytic hierarchy process (AHP), achieving quantitative classification and accurate identification of congestion levels. The model fully considers actual operational factors such as lane conditions, on-site personnel configuration, and site conditions. Through data balancing and feature selection optimization, it effectively improves the sensitivity of identifying mild to severe congestion while ensuring the accuracy of judging smooth traffic flow. This avoids scheduling delays caused by missed congestion assessments and eliminates resource waste caused by ineffective scheduling. As a key link connecting traffic prediction and cloud warehouse scheduling, congestion assessment provides a clear basis for matching corresponding scheduling strategies to different levels of congestion, achieving an orderly connection between "accurate assessment and targeted scheduling."

[0100] S4: Generate resource allocation strategies based on the current cloud warehouse status and predicted congestion levels.

[0101] Furthermore, the quantitative decision-making factors for the resource allocation strategy include: Cloud seat configuration ( This includes the number of seats (people) and the average skill score (evaluated based on assessment results, ranging from 0 to 100 points). Cloud warehouse busy level ( This includes the number of special cases currently being handled and the percentage of available seats (%). Network latency ( ): Includes the average data transmission latency (ms) between the cloud warehouse and the toll station; Historical processing records ( ): This includes the percentage of times the target toll station was processed (%), the average processing time (seconds), and the processing accuracy rate (%).

[0102] In this embodiment, the initial weight of each decision factor is calculated using the Analytic Hierarchy Process (AHP), including the following steps: (1) Construct a judgment matrix (based on scores from 10 experts, adapted to scheduling decision priorities):

[0103] (2) Weight calculation: After the feature vector is normalized, the weight vector is [0.35, 0.25, 0.20, 0.20]. The consistency test CR=0.045<0.1, and the weight is reasonable.

[0104] Resource allocation strategy: The resource allocation strategies include single-toll-station congestion scheduling strategies and multi-toll-station congestion scheduling strategies.

[0105] Furthermore, the single toll station congestion scheduling strategy is applicable to congestion at a single toll station, and includes the following steps: Construct a priority scoring function: , in, The priority score for the j-th cloud warehouse. Let the initial weight of the k-th factor be... Let be the standardized value of the j-th cloud warehouse for the k-th factor; Adjust scheduling rules: When the current toll station is in a state of mild congestion (e.g., CCI∈[40-60]): increase the weight of cloud warehouse busyness and network latency; for example, prioritize scheduling cloud warehouses with idle seat ratio ≥50% and network latency ≤50ms.

[0106] When the current toll station is under moderate congestion (e.g., CCI∈[60-80]): increase the weight of cloud seat configuration and historical processing records; for example, prioritize cloud warehouses with a skill score ≥80 and a historical processing accuracy ≥90%.

[0107] When the current toll station is severely congested (e.g., CCI > 80): increase the weight of cloud warehouse busyness and historical processing records; for example, schedule cloud warehouses with idle seat ratio ≥ 30% + historical processing time ≤ 60 seconds, and activate the emergency support mechanism at the same time.

[0108] Calculate the priority score for each cloud warehouse based on the adjusted weights, and select... The highest-level cloud warehouse manages the current toll stations.

[0109] Furthermore, the multi-toll station congestion scheduling strategy is applicable when congestion occurs at more than one toll station, and its cloud warehouse management scheduling function is: Establish a multi-objective optimization model: , Constraints: , Where m represents the number of congested toll booths, and n represents the number of cloud warehouses. The sum of network latency and processing time when cloud warehouse j processes toll station i. To process the benefits (i.e., the degree of congestion relief, determined based on the magnitude of the CCI decrease). The number of available seats in cloud warehouse j is multiplied by a set coefficient (in this embodiment, the set coefficient is 2). To assign identifiers, =1 indicates allocation. =0 means no allocation.

[0110] S5: The cloud warehouse performs special handling according to the resource allocation strategy, and enters S1 according to the preset trigger conditions, and dynamically adjusts the strategy based on the real-time congestion level of the toll station (CCI update value) and the cloud warehouse status (number of idle seats, processing efficiency).

[0111] Furthermore, the preset triggering conditions include: setting an update time (e.g., 15 minutes), cloud warehouse processing efficiency decreasing beyond a set risk threshold, and the actual congestion level at the toll station being higher than the predicted congestion level.

[0112] This invention overcomes the limitations of existing technologies that rely on a single data source and ignore the tidal fluctuations in traffic flow. It integrates multiple influencing factors, including highway segment characteristics, ETC / CPC traffic data, gantry monitoring information, historical traffic flow, and real-time weather. Through scenario-based feature engineering, it strengthens the feature representation of key scenarios and leverages the Attention-LSTM model's ability to capture long-term dependencies in time series data to achieve accurate predictions of entrance and exit traffic flow and tidal peaks. This technology is adaptable to different traffic flow fluctuation scenarios, including off-peak, peak, and tidal peak periods, effectively compensating for the accuracy shortcomings of cross-regional traffic flow prediction. It provides highly reliable data support for subsequent congestion prediction, improving the foresight of toll station operation and management from the source.

[0113] Simultaneously, it abandons the traditional single-dimensional approach to congestion assessment, constructing a core evaluation index system encompassing average vehicle travel time, lane utilization, average queue length, and traffic delay rate. Scientific methods are used to determine the weights of each index, and the XGBoost model is integrated to adapt to congestion scenarios involving multiple factors and nonlinear coupling. Through data balancing and feature selection optimization, the model can accurately output congestion classification results, improving the sensitivity of identifying mild to severe congestion while ensuring the accuracy of judging smooth traffic flow. This effectively avoids scheduling delays caused by missed congestion assessments and resource waste due to ineffective scheduling, becoming a crucial bridge connecting traffic flow prediction and cloud warehouse scheduling, providing a clear basis for targeted scheduling.

[0114] Finally, this invention focuses on the diverse needs of actual toll station operations, providing customized optimizations for scenarios such as traffic fluctuations at different times, overlapping congestion at multiple toll stations, and equipment failures. Targeted data preprocessing ensures data quality, and scenario-based optimization of model parameters improves adaptability, ensuring the technical solution can flexibly handle different operational scenarios such as off-peak, peak, and holiday periods. Simultaneously, the combination of a tiered scheduling strategy and an emergency adjustment mechanism further enhances the feasibility and stability of the technical solution, avoiding ineffective designs divorced from actual operational needs, and providing a practical and intelligent solution for toll station management in smart transportation scenarios.

[0115] Example 3 like Figure 2 As shown, a toll station cloud warehouse management and scheduling system based on tidal prediction is disclosed. The system is used to execute the toll station cloud warehouse management and scheduling method based on tidal prediction described in Embodiment 2. This system adopts a hybrid architecture of "edge computing + cloud computing" to adapt to the distributed deployment and real-time scheduling requirements of toll stations. It includes a cloud warehouse platform and several edge nodes that are communicatively connected to the cloud warehouse platform.

[0116] The edge nodes are deployed locally at each toll station to acquire real-time toll station data and perform preprocessing (data cleaning, feature extraction) to reduce transmission pressure.

[0117] The cloud warehouse platform is deployed in the traffic control center (or regional traffic cloud center) and is used to generate resource allocation strategies based on the real-time toll station data. This cloud warehouse platform integrates a predictive model, a scheduling module, and a special situation handling system, supporting parallel management of multiple toll stations.

[0118] Furthermore, the system described in this embodiment uses a 5G + fiber optic dual-link transmission network, employing TCP / IP (batch data) + MQTT (real-time data) protocols to ensure data transmission latency ≤100ms and reliability ≥99.9%.

[0119] Furthermore, the scope of management and the division of permissions in the system are as follows: Management scope: Based on geographical location and traffic volume, the 30 toll stations in the region are divided into 3 management areas, with each cloud warehouse responsible for 10 toll stations, ensuring that the dispatch response time is ≤3 minutes.

[0120] Access control is implemented using the RBAC model, with cloud warehouse administrators (monitoring, scheduling, and control permissions), scheduling specialists (scheduling permissions), and cloud agents (special situation handling permissions) providing hierarchical management to avoid operational risks.

[0121] The actual operation process of the toll station cloud warehouse management and scheduling system based on tidal prediction is as follows: Data acquisition: Edge nodes (toll collection equipment, monitoring, and weather sensors at the toll station site) collect data in real time and perform preprocessing.

[0122] Status Prediction: The cloud warehouse platform calls the traffic flow and congestion prediction model to output the congestion level 15 minutes later.

[0123] Scheduling decision: The cloud warehouse platform generates resource allocation strategies based on congestion levels and cloud warehouse status.

[0124] Special situation handling: Cloud-based agents execute special situation handling according to the policy (remote operation / on-site guidance).

[0125] Results Feedback: The on-site system will send the processing results (travel time, degree of congestion relief) to the cloud warehouse to update the scheduling strategy.

[0126] This invention employs a hybrid architecture of "edge computing + cloud computing" and a dual-link transmission design of 5G + fiber optics, significantly reducing data transmission latency and ensuring rapid implementation of scheduling commands. By rationally dividing the management scope and implementing RBAC hierarchical permission control, centralized coordination and secure management of toll station resources within the region are achieved, significantly improving the scheduling response efficiency of each cloud warehouse for its assigned toll stations. For different scenarios of simultaneous congestion at a single toll station and multiple toll stations, priority scoring mechanisms and multi-objective optimization models are designed respectively. Combining core factors such as cloud agent configuration, cloud warehouse busyness, and historical processing records, optimal resource allocation is achieved. The real-time adjustment mechanism ensures that the scheduling strategy can be dynamically optimized based on actual congestion changes and cloud warehouse status, effectively improving cloud warehouse resource utilization. In cases of insufficient special handling personnel or overlapping congestion at multiple stations, idle resources can be quickly integrated and emergency support activated, significantly shortening congestion relief time.

[0127] Meanwhile, the "traffic flow tidal prediction - congestion assessment - cloud warehouse dynamic scheduling" full-process technical system constructed by this invention completely solves the problem of disconnect between various links in existing technologies, forming a complete closed loop of "data collection - status prediction - scheduling execution - result feedback". Through the coordinated linkage of various modules, the toll station's traffic efficiency is significantly improved, vehicle queuing time and traffic delays are greatly reduced, and the travel experience of drivers is continuously improved. At the same time, intelligent scheduling reduces the ineffective consumption of resources such as manpower and equipment, and lowers operation and management costs. The combination of hierarchical scheduling and emergency adjustment mechanisms enhances the toll station's ability to respond to special situations such as sudden congestion and equipment failure, ensuring the stability and flexibility of operation and management. This technical solution provides a feasible intelligent solution for toll station management in smart transportation scenarios, achieving a triple improvement in traffic efficiency, service quality, and resource utilization, and has significant engineering application value and socio-economic benefits.

[0128] Example 4 like Figure 3 As shown, a toll station cloud warehouse management and scheduling device based on tidal prediction includes at least one processor, a memory communicatively connected to the at least one processor, and at least one input / output interface communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, enables the at least one processor to perform the toll station cloud warehouse management and scheduling method based on tidal prediction described in the foregoing embodiments. The input / output interface may include a display, keyboard, mouse, and USB interface for inputting and outputting data.

[0129] Furthermore, the tidal prediction-based toll station cloud warehouse management and scheduling equipment can be a desktop computer, mobile phone, tablet computer, wearable tidal prediction-based toll station cloud warehouse management and scheduling equipment, etc., capable of deep information recognition.

[0130] Furthermore, the processor may include one or more processing cores. The processor connects to various parts of the tidal prediction-based toll station cloud warehouse management and scheduling equipment using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory, and by calling data stored in memory. Optionally, the processor may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor may integrate one or more of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also be implemented separately as a communication chip, without being integrated into the processor.

[0131] The memory may include random access memory (RAM) or read-only memory (ROM). The memory can be used to store instructions, programs, code, code sets, or instruction sets, such as instructions or code sets used to implement the tidal prediction-based toll station cloud warehouse management and scheduling method provided in this application embodiment. The memory may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the various method embodiments described above, etc. The data storage area may also store data created during the use of the tidal prediction-based toll station cloud warehouse management and scheduling device (such as a modulation sequence-depth mapping table, image data, spectrogram data, etc.).

[0132] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory, magnetic disks, or optical disks.

[0133] When the integrated units of the present invention are implemented as software functional units and sold or used as independent products, they can also be stored in a computer-readable storage medium. The computer-readable storage medium stores program code, which can be called by a processor to execute the methods described in the above method embodiments. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes electronic memories such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, the computer-readable storage medium includes a non-transitory computer-readable storage medium. The computer-readable storage medium has storage space for program code that executes any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code can be compressed, for example, in an appropriate form.

[0134] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A toll station cloud warehouse management and scheduling method based on tidal prediction, characterized in that, Includes the following steps: S1: Obtain real-time toll station data and perform preprocessing; S2: Input the real-time toll station data into the pre-trained traffic flow tidal prediction model to generate the predicted traffic flow data for the corresponding toll station after a set time. S3: Input the predicted traffic flow data into the pre-trained congestion prediction model to generate the predicted congestion level for the corresponding toll station after a set time. S4: Generate a resource allocation strategy based on the current cloud warehouse status and the predicted congestion level; the resource allocation strategy includes a single toll station congestion scheduling strategy and a multi-toll station congestion scheduling strategy; S5: The cloud warehouse performs special handling according to the resource allocation strategy and enters S1 according to the preset trigger conditions.

2. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 1, characterized in that, S1 includes the following steps: S11: Obtain real-time toll station data for the area to be dispatched; The real-time toll station data includes real-time demand data, periodic data, and static data; The real-time data requirements include gantry data, real-time weather, and traffic control data. The periodic data includes toll record data and ETC / CPC data; The static data includes road segment information and holiday schedule data; S12: Preprocess the real-time toll station data; the preprocessing includes data cleaning and standardization.

3. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 2, characterized in that, The data cleaning includes missing value processing, outlier processing, and format standardization. The missing value handling includes the following steps: Get the percentage of missing data in the current data; When the missing percentage is less than the first missing threshold, linear interpolation is used to process the current data. When the first missing threshold ≤ missing proportion < second missing threshold; KNN interpolation is used, and the current data is weighted and imputed according to the principles of temporal proximity and feature similarity. When the second missing threshold is less than or equal to the missing proportion; if the current data is a non-core feature as defined, the current data is deleted directly; if the current data is a core feature as defined, a random forest model is used to predict and impute it. The outlier handling includes the following steps: Outlier detection is performed on each data point sequentially; if the current data is numerical, the interquartile range method is used for detection; if the current data is high-dimensional, the Isolation Forest algorithm is used for detection. Determine the type of detected outlier; if the current outlier is a device malfunction, use interpolation correction; if the current outlier is a genuine outlier caused by extreme weather and / or holidays, retain the current outlier and mark it as an outlier. The standardization includes data standardization processing or data normalization processing; The data standardization process is used for data that follows a normal distribution, and its expression is as follows: ， in, The data is standardized, x is the input data, μ is the mean of the training set, and σ is the standard deviation of the training set. The data normalization process is used for non-normally distributed data, and its processing expression is as follows: ， Where, x min x is the minimum value of the current data type. max This represents the maximum value of the current data type.

4. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 2, characterized in that, The prediction results of the traffic flow tidal prediction model include: Entrance traffic flow : , Where τ is the instantaneous time point, for Instantaneous traffic flow at the entrance. For prediction time granularity; Export traffic volume : , in, for Instantaneous traffic flow at the exit; Tidal peak flow : , in, ∈ or , This is a collection of tidal periods.

5. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 2, characterized in that, The congestion prediction model adopts the XGBoost model, and uses grid search and 5-fold cross-validation to iteratively fine-tune the hyperparameters. During the training process, the congestion prediction model uses the SMOTE algorithm to handle sample imbalance and removes redundant features with weights below the threshold based on the built-in feature importance score results.

6. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 5, characterized in that, The congestion level is evaluated using a comprehensive congestion index; its expression is: ， Among them, w i Assign weights to the i-th type of indicator. is the normalized value of the i-th type of index; The evaluation indicators for the congestion level include: Average travel time for vehicles : , in, is the passage time for a single vehicle, and n is the total number of vehicles in the statistical period; Lane utilization : , in, The lane occupancy time for a single vehicle is T, the total time of the statistical period is m, and the total number of lanes is m. Average queue length : , Where L is the average vehicle length; Vehicle traffic delay rate : , in, The average travel time for free flow. For free circulation time.

7. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 6, characterized in that, The single toll station congestion scheduling strategy is applicable to situations where congestion occurs at a single toll station, and includes the following steps: Quantitative decision-making factors for determining resource allocation strategies; The quantitative decision factors for the resource allocation strategy include: cloud agent configuration: including the number of agents and average skill scores; cloud warehouse busyness: including the number of special cases currently being handled and the percentage of idle agents; network latency: including the average data transmission latency between the cloud warehouse and the toll station; historical processing records: including the percentage of times the target toll station was processed, the average processing time, and the processing accuracy. Construct a priority scoring function: ， in, The priority score for the j-th cloud warehouse. Let the initial weight of the k-th factor be... Let be the standardized value of the j-th cloud warehouse for the k-th factor; Adjust scheduling rules: When the current toll station is experiencing mild congestion: increase the weighting of cloud warehouse busy level and network latency; When the current toll station is under moderate congestion: increase the weight of cloud seat configuration and historical processing records; When the current toll station is severely congested: increase the weight of cloud warehouse busy level and historical processing records; Calculate the priority score for each cloud warehouse based on the adjusted weights, and select... The highest-level cloud warehouse manages the current toll stations.

8. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 7, characterized in that, The multi-toll station congestion scheduling strategy is applicable when congestion occurs at more than one toll station, and its cloud warehouse management scheduling function is: Establish a multi-objective optimization model: , Constraints: , Where m represents the number of congested toll booths, and n represents the number of cloud warehouses. The sum of network latency and processing time when cloud warehouse j processes toll station i. In order to handle the revenue, The number of available seats in cloud warehouse j is multiplied by a set coefficient. To assign identifiers, =1 indicates allocation. =0 means no allocation.

9. The toll station cloud warehouse management and scheduling method based on tidal prediction according to claim 1, characterized in that, The preset trigger conditions include: setting an update time, cloud warehouse processing efficiency decreasing beyond a set risk threshold, and the actual congestion level at the toll station being higher than the predicted congestion level.

10. A toll station cloud warehouse management and scheduling system based on tidal prediction, characterized in that, The system is used to execute a toll station cloud warehouse management and scheduling method based on tidal prediction as described in any one of claims 1 to 9; it includes a cloud warehouse platform and several edge nodes that are communicatively connected to the cloud warehouse platform; The edge nodes are deployed locally at each toll station to acquire real-time toll station data and perform preprocessing. The cloud warehouse platform is deployed at the traffic control center and is used to generate resource allocation strategies based on the real-time toll station data.

11. A toll station cloud warehouse management and scheduling device based on tidal prediction, characterized in that, It includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a toll station cloud warehouse management and scheduling method based on traffic flow tidal prediction as described in any one of claims 1 to 9.

Images