A base station resource scheduling method based on traffic and communication characteristic complementary prediction

Abstract

This invention discloses a base station resource scheduling method based on complementary prediction of traffic and communication features. It proposes a traffic flow and data flow prediction optimization model based on dynamic time programming. Vehicle connection data and vehicle communication volume data are acquired between base stations. The prediction model is trained using DDPG to predict the number of vehicles connected to each base station and the communication volume in the future. Based on the prediction data, a bandwidth allocation optimization model considering overall system latency and the equality of resource allocation at different times is designed. The optimal solution is found through projected gradient descent, achieving resource scheduling for future traffic flow. This method overcomes the problem of base stations being unable to dynamically adjust bandwidth allocation strategies according to traffic flow changes, leading to resource waste or shortages. By predicting traffic flow and communication demand, the method allows base stations to allocate more resources when traffic flow is expected to increase and less resources when traffic flow is expected to decrease, achieving efficient resource utilization and improving bandwidth resource utilization in intelligent transportation systems.

Description

Technical Field

[0001] This invention belongs to the field of vehicle-to-everything (V2X) communication technology, specifically relating to a base station resource scheduling method based on complementary prediction of traffic and communication features. Background Technology

[0002] With the gradual maturation of vehicle-to-everything (V2X) technology and the arrival of the big data era, more and more vehicles are equipped with intelligent services, such as autonomous driving and environmental perception. These services have led to a sharp increase in the demand for communication resources, resulting in a decline in the efficiency of V2X spectrum resource utilization. To address this challenge, it is necessary to improve the utilization rate of V2X spectrum resources through technologies such as traffic flow prediction and communication resource reservation. Existing traffic flow prediction methods mainly focus on the distribution of traffic flow in different areas at the same time, improving resource utilization efficiency through inter-regional resource scheduling, or indirectly balancing inter-regional communication demand through traffic flow scheduling. However, these methods ignore the temporal resource scheduling within the same region and cannot reserve resources for upcoming traffic flows. At the same time, traffic flow and communication flow are interconnected but have different characteristics. Traffic flow is constrained by the road network and changes gradually, while communication flow, although affected by traffic flow, exhibits drastic fluctuations due to the randomness of information transmission. This also poses a significant challenge to intelligent scheduling of communication resources based on traffic flow prediction. Improving the comprehensiveness, spatiotemporal correlation, and data complementarity of road network traffic flow and user communication characteristics in terms of traffic communication flow prediction is a feasible direction.

[0003] In recent years, traffic flow prediction techniques based on Dynamic Time Warping (DTW) algorithms have been widely applied. This method helps extract the temporal characteristics of traffic flow, thereby making accurate predictions of upcoming traffic flow and communication demands. On the other hand, deep learning has also achieved widespread application in recent years, among which Deep Deterministic Policy Gradient (DDPG) has become a popular technique, capable of intelligently scheduling current and future communication resources based on traffic flow and communication demand predictions.

[0004] In summary, current vehicle-to-everything (V2X) networks cannot efficiently schedule spectrum resources across time sequences, preventing base stations from intelligently allocating spectrum resources based on potential future traffic flow. Secondly, traffic flow prediction does not directly reflect communication flow. Traffic flow, constrained by the road network, changes gradually, while communication flow, although affected by traffic flow, exhibits significant fluctuations due to the randomness of information transmission. This means that communication flow cannot be directly predicted from traffic flow; a comprehensive consideration of the correlation and complementarity between the two is necessary. Traffic flow prediction and communication resource reservation are crucial techniques for improving spectrum resource utilization. Traffic flow prediction technology based on DDPG and dynamic time planning algorithms can help extract temporal features and predict future traffic flow and communication demands. Simultaneously, base stations can utilize these prediction results to make optimal resource scheduling decisions under resource constraints. The combination of these technologies can better meet the needs of vehicle communication, particularly by achieving reasonable resource reservation to satisfy the overall communication needs of currently arriving and future arriving vehicles. Summary of the Invention

[0005] This invention proposes a base station resource scheduling method based on complementary prediction of traffic and communication features. It obtains time-series data of two different changing characteristics of historical connected vehicles and communication demand of surrounding base stations, complementing each other to establish the spatiotemporal correlation between road network traffic flow and user communication features. The similarity of the changing features of the two is extracted by dynamic time planning algorithm to establish a more comprehensive prediction model to improve prediction accuracy. The optimization model is solved using DDPG to obtain the number of vehicles and communication demand that may arrive in the future. Based on the prediction results, the base station resource reservation is guided to minimize the total vehicle delay and time delay variance, and the problem is optimized using the projected gradient descent method.

[0006] The technical solution adopted in this invention is: a base station resource scheduling method based on complementary prediction of traffic and communication features, the specific steps of which are as follows:

[0007] S1, Base station sampling;

[0008] Base station time step For current data sampling, a set of sampled data includes: the current time. Connected to base station Number of vehicles The sum of the communication needs generated by these vehicles .

[0009] in, Indicates in Time vehicle Is it connected to the base station? superior, Indicates a connection. This indicates that the connection is not established. express Vehicles within the time period At the base station The communication needs generated above; This represents the total number of all vehicles in the system.

[0010] The base station stores the sampled data in a historical data pool and saves it. Group data Determine if the sampling exceeds Delete more than 100 data entries. If the oldest historical data is not older than the limit, sampling will continue while data will not be made available to other base stations until the sampling limit is reached. This is a request from a neighboring base station. When processing group data, the latest data is transmitted directly through inter-base station communication.

[0011] in, Indicates the latest A vehicle count vector composed of sampled vehicle counts. Indicates the latest The communication demand vector, composed of samples of vehicle communication demands, together with the traffic flow and user communication characteristic data of the complete road network. .

[0012] Simultaneously, the base station records the remaining bandwidth of the base station and the bandwidth allocated to vehicles, which is used for calculating the base station's communication resource reservation. At any moment The remaining bandwidth is represented as: .

[0013] in, Indicates vehicle time At the base station The actual bandwidth allocated. Indicates base station Total bandwidth.

[0014] S2, base station Obtain the number of vehicles accessing adjacent base stations and the time-series data of communication volume of adjacent base stations;

[0015] base station Request traffic and communication data from neighboring base stations to obtain information about several neighboring base stations. data , , .

[0016] S3. Predicting traffic flow and communication demand based on dynamic time planning algorithm and DDPG;

[0017] Base station The number of vehicles accessing adjacent base stations and their communication traffic time-series data are input into the optimization model to obtain future... time step Predicted traffic flow and communication demand data .

[0018] in, This represents predicted traffic flow data. This represents data predicting communication demand. Indicates base station The start timestamp of the predicted data.

[0019] The optimization model is specifically as follows:

[0020]

[0021] in, Represents the predictive model; Indicates the data transmission time between adjacent base stations. This indicates the historical data of adjacent base stations and the target base station. DTW average similarity of the predicted data Indicates the target base station The error between predicted and actual data; Indicates time base station Maximum number of vehicles that can be connected; Indicates vehicle time At the base station Service transmission latency Indicates vehicle time At the base station The maximum acceptable service transmission latency Indicates vehicle time At the base station The minimum acceptable bandwidth requirement; Indicates base station and base stations The amount of data transmitted between them Indicates base station and base stations The transmission rate between them; Indicates optimizing model training accuracy, This represents the minimum acceptable training accuracy of the optimized model.

[0022] This indicates that the number of vehicles connected to the base station does not exceed the maximum number of vehicles that can be served. This indicates that the actual latency of the vehicle is no greater than the maximum acceptable latency, from which the minimum bandwidth of the vehicle can be obtained. , This indicates that the historical data transmission time needs to be much less than A specific time step is needed; otherwise, the data will be too old and lose its meaning. This indicates that the training accuracy is greater than the minimum accuracy.

[0023] Then the DDPG algorithm is used to optimize the prediction model.

[0024] S4. Minimize the delay and delay variance, and use the projective gradient descent method to solve for the optimal bandwidth allocation scheme;

[0025] After step S3, the base station Predictive data is obtained by requesting data from neighboring base stations and by performing model calculations. . Indicates base station The start timestamp of the prediction data is important because the transmission and prediction process takes time. Therefore, the start time of the predicted data is earlier than the time when the predicted data is obtained, that is, the time when the predicted data is obtained is... .Will Number of vehicles arriving within and communication volume Replace with forecast data In the process, mixed prediction data is obtained. At this point, the vehicle arrival time is specifically predicted as follows:

[0026]

[0027]

[0028] in, and Represents real data. and This indicates the predicted data, where the first subscript represents the data timestamp and the second represents the base station of the predicted data.

[0029] base station Based on mixed forecast data We construct an optimization problem and solve it using the projective gradient descent method to obtain the optimal bandwidth allocation scheme, which is the bandwidth allocation scheme for the current connected vehicles and the future connected vehicles.

[0030] The optimization problem is that at each time step... The calculation of the average vehicle latency within the base station and the squared difference between the mean of this average latency at all time steps is performed over the total time. Inner summation is obtained Solve this optimization problem using gradient descent:

[0031]

[0032] in, Indicates base station At any moment The total bandwidth allocated to vehicles. Indicates time At the base station Available bandwidth on This represents an optimization problem; the first constraint represents the bandwidth allocated at each time step. No more than available bandwidth The second constraint represents the bandwidth of the vehicle segment. To exceed the acceptable minimum bandwidth requirement, optimization problem It consists of two parts: the sum of the system's average delays and the variance of the system's average delays, specifically:

[0033]

[0034] in, This represents the sum of the system's average latency, ensuring that the communication resource allocation scheme has low latency. Indicates base station At any moment The total bandwidth allocated to vehicles. Indicates a guarantee Small positive numbers with a non-zero denominator. , This represents the sum of time delays and the time delay variance weighting factor. This represents the total time taken to run the gradient descent method after obtaining the predicted data, and .

[0035] , Indicates base station Will Time bandwidth The remaining bandwidth after allocation Indicates base station At any moment The total bandwidth allocated to vehicles, the third item This indicates the recovery status of previously allocated bandwidth. This indicates the reclaiming of the sliding window, from the last time it was not reclaimed to the current time. This is a bandwidth reclamation indicator; 1 indicates that the bandwidth has not been reclaimed, and 0 indicates that it has been reclaimed. Indicates a delay indicator, when hour, Otherwise, it is 0. express The average latency of service transmission corresponding to this portion of bandwidth Therefore, bandwidth is allocated. Front Base Station exist Available bandwidth at any time is Only with It is related to the allocation situation before the time. Once the bandwidth allocation is complete, the available bandwidth for the next moment can be calculated directly.

[0036] S5, base station According to the optimal bandwidth allocation scheme Allocate bandwidth resources, take into account subsequent traffic flow, and realize resource scheduling for future traffic flow;

[0037] Among them, vehicles With time At the base station Traffic generated on Assign weights, distribute ,Right now .

[0038] Furthermore, in step S3, the optimization problem includes the data transmission time between adjacent base stations, historical data of adjacent base stations, and the target base station. DTW average similarity of predicted data, target base station The errors between the predicted and actual data are as follows:

[0039] (1) Represents the set of adjacent base stations to base station The sum of communication delays for transmitting data;

[0040] in, Indicates base station to base station Data size to be transmitted Indicates base station to base station Bandwidth for data transmission Indicates base station to base station Transmission power for transmitting data Indicates base station to base station Channel gain for data transmission Indicates noise power. Indicates base station Base stations that generate noise from outside sources.

[0041] (2) Indicates adjacent base stations Historical data and target base stations DTW average similarity of the predicted data;

[0042] in, This represents the DTW average similarity function, used to calculate the similarity between two time series data points; the smaller the value, the more similar they are. This represents the weight of each historical data point, which decays exponentially with the freshness of the data over time. , and These represent the attenuation factor and the attenuation exponent factor, respectively.

[0043] (3) Indicates the target base station The error between predicted and actual data;

[0044] in, This represents the sum of time delays and the time delay variance weighting factor. Indicates the target base station The error between the predicted and actual number of vehicles connected. Indicates the target base station Error between predicted and actual communication demand data This indicates the first and last timestamps of the predicted data.

[0045] Furthermore, in step S3, the method for calculating the DTW average similarity is as follows:

[0046] A1. Define a cost matrix. That is, the historical data of adjacent base stations and the target base station The predicted data constitutes a two-dimensional matrix, where any position within the matrix... The value is the historical data of adjacent base stations and the target base station. The Euclidean distance between the predicted data and their corresponding indices is , This indicates that this element is located at the leftmost position in the cost matrix. List, This indicates that this element is located at the top of the cost matrix from bottom to top. OK;

[0047] A2. Define a dynamic time-planning shortest path. It meets the following conditions:

[0048] a1、 The number of elements is between the longest side of the matrix and the number of elements in the matrix;

[0049] a2、 With the beginning and end fixed, the shortest path can only be from the bottom left corner to the top right corner of the matrix;

[0050] a3, if , ,but ,and Each step of the shortest path can only be taken in three directions: right, up, or upper right.

[0051] A3. In the cost matrix In the middle, the search proceeds from the bottom left corner to the top right corner to find the shortest path that meets the requirements. At this point, the adjacent base stations... Historical data and target base stations The DTW average similarity function for the predicted data is:

[0052]

[0053] Furthermore, in step S3, the optimization problem is solved using the projection gradient method, as detailed below:

[0054] B1. Initialization: Randomly generate bandwidth allocation scheme. Set the number of iterations ,get ;

[0055] B2. Calculate the average delay Calculate the optimization function exist Gradient ;

[0056] Specifically, the gradient is:

[0057]

[0058] B3. Based on gradient direction and step size renew ,get ;

[0059] B4. The updated version Projected onto the feasible region ,Right now In the middle, the distance in the feasible region is obtained. nearest point The specific projection method is as follows:

[0060]

[0061] in, Represents the projection operator. Indicates the first During the next iteration, the base station exist Available bandwidth at any given time.

[0062] B5. Determine if convergence has occurred: If the decrease in the optimization function is less than the threshold... ,Right now Or the number of iterations is greater than the maximum number of iterations, i.e. Then output the current solution. ,otherwise Return to step B2.

[0063] The beneficial effects of this invention are as follows: This invention proposes a traffic flow and data flow prediction optimization model based on dynamic time planning. It acquires vehicle connection data and vehicle communication volume data between base stations, trains the prediction model using DDPG, and predicts the number of vehicles connected to each base station and the communication volume in the future. Based on the predicted data, a bandwidth allocation optimization model considering overall system latency and the equality of resource allocation at each time point is designed. The optimal solution is found through projective gradient descent, achieving resource scheduling for future traffic flow. This invention overcomes the problem of base stations being unable to dynamically adjust bandwidth allocation strategies according to changes in traffic flow, leading to resource waste or shortage. By predicting traffic flow and communication demand, base stations allocate more resources when traffic flow is expected to increase and less resources when traffic flow is expected to decrease, achieving efficient resource utilization and improving bandwidth resource utilization in intelligent transportation systems. This invention also considers the gains in information comprehensiveness, spatiotemporal correlation, and data complementarity between road network traffic flow and user communication characteristics in traffic communication flow prediction, thereby improving the comprehensiveness and accuracy of the prediction model. Based on the dynamic time warping algorithm and DDPG, the method of this invention uses complementary prediction of road network traffic flow and user communication characteristics to obtain the upcoming traffic flow and communication demand in a specific area. By using the gradient descent method, the prediction results are used to calculate the optimal bandwidth allocation scheme, thereby directly guiding the base station to reserve bandwidth resources. Attached Figure Description

[0064] Figure 1 This is a flowchart of a base station resource scheduling method based on complementary prediction of traffic and communication features according to the present invention.

[0065] Figure 2 This is a schematic diagram of road network modeling in an embodiment of the present invention.

[0066] Figure 3 This is a schematic diagram of the cost matrix in dynamic time planning in an embodiment of the present invention. Detailed Implementation

[0067] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0068] like Figure 1 The flowchart of a base station resource scheduling method based on complementary prediction of traffic and communication features of the present invention is shown below. The specific steps are as follows:

[0069] like Figure 2As shown in the diagram, the application scenario of this embodiment is a road network modeling diagram. The road network consists of several roads; several base stations (also known as roadside units, RSUs) cover the road network and provide vehicle-to-everything (V2I) services to vehicles in the road network; vehicles establish V2I (vehicle-to-infrastructure) connections with base stations; by modeling the real road network, the historical number of connected vehicles and communication requirements of surrounding base stations are obtained.

[0070] S1, Base station sampling;

[0071] Base station time step For current data sampling, a set of sampled data includes: the current time. Number of vehicles connected to the base station The sum of the average communication demands generated by these vehicles .

[0072] in, Indicates in Time vehicle Is it connected to the base station? superior, Indicates a connection. This indicates that the connection is not established. express Vehicles within the time period At the base station The communication needs generated above; This represents the total number of all vehicles in the system.

[0073] The base station stores the sampled data in a historical data pool and saves it. Group data Determine if the sampling exceeds Delete more than 100 data entries. If the oldest historical data is not older than the limit, sampling will continue while data will not be made available to other base stations until the sampling limit is reached. This is a request from a neighboring base station. When processing group data, the latest data is transmitted directly through inter-base station communication.

[0074] in, Indicates the latest A vehicle count vector composed of sampled vehicle counts. Indicates the latest The communication demand vector, composed of samples of vehicle communication demands, together with the traffic flow and user communication characteristic data of the complete road network. .

[0075] Simultaneously, the base station records the remaining bandwidth of the base station and the bandwidth allocated to vehicles, which is used for calculating the base station's communication resource reservation. At any moment The remaining bandwidth is represented as: .

[0076] in, Indicates vehicle time At the base station The actual bandwidth allocated. Indicates base station Total bandwidth.

[0077] S2, base station Obtain the number of vehicles accessing adjacent base stations and the time-series data of communication volume of adjacent base stations;

[0078] base station Request traffic and communication data from neighboring base stations to obtain information about several neighboring base stations. data , , .

[0079] Each base station sends data immediately upon receiving a data request. In fact, the more distant the base station and the longer the communication latency, the fresher the data is in the same round of requests. In other words, the communication latency between base stations is different, and the timestamp of the latest historical data may not be the same.

[0080] S3. Predicting traffic flow and communication demand based on dynamic time planning algorithm and DDPG;

[0081] Base station The number of vehicles accessing adjacent base stations and their communication traffic time-series data are input into the optimization model to obtain future... time step Predicted traffic flow and communication demand data .

[0082] in, This represents predicted traffic flow data. This represents data predicting communication demand. Indicates base station The start timestamp of the predicted data.

[0083] The optimization model described herein is a problem that minimizes the optimization problem in order to solve for the optimal prediction model. The optimization problem includes the data transmission time between adjacent base stations, the historical data of adjacent base stations, and the target base station. DTW average similarity of predicted data, target base station The error between the predicted data and the actual data is as follows:

[0084]

[0085] in, Represents the predictive model; Indicates the data transmission time between adjacent base stations. This indicates the historical data of adjacent base stations and the target base station. DTW average similarity of the predicted data Indicates the target base station The error between predicted and actual data; Indicates time base station Maximum number of vehicles that can be connected; Indicates vehicle time At the base station Service transmission latency Indicates vehicle time At the base station The maximum acceptable service transmission latency Indicates vehicle time At the base station The minimum acceptable bandwidth requirement; Indicates base station and base stations The amount of data transmitted between them Indicates base station and base stations The transmission rate between them; Indicates optimizing model training accuracy, This represents the minimum acceptable training accuracy of the optimized model.

[0086] Among the constraints, This indicates that the number of vehicles connected to the base station does not exceed the maximum number of vehicles that can be served. This indicates that the actual latency of the vehicle is no greater than the maximum acceptable latency, from which the minimum bandwidth of the vehicle can be obtained. , This indicates that the historical data transmission time needs to be much less than A specific time step is needed; otherwise, the data will be too old and lose its meaning. This indicates that the training accuracy is greater than the minimum accuracy.

[0087] The optimization problem consists of three parts: data transmission time between adjacent base stations. Historical data of adjacent base stations and target base station DTW average similarity of predicted data Target base station Error between predicted and actual data Specifically:

[0088] (1) Represents the set of adjacent base stations to base station The sum of communication delays for transmitting data.

[0089] in, Indicates base station to base station Data size to be transmitted Indicates base station to base station Bandwidth for data transmission Indicates base station to base station Transmission power for transmitting data Indicates base station to base station Channel gain for data transmission Indicates noise power.

[0090] (2) Indicates adjacent base stations Historical data and target base stations DTW average similarity of the predicted data.

[0091] in, This represents the DTW average similarity function, used to calculate the similarity between two time series data. A lower similarity indicates greater similarity. This represents the weight of each historical data point, which decays exponentially with the freshness of the data over time. , and These represent the attenuation factor and the attenuation exponent factor, respectively.

[0092] Traffic flow determines the upper limit of communication flow, and changes in communication flow can be seen as changes based on the numerical value of traffic flow. This invention combines traffic flow and communication flow, so that the combined data exhibits slow change characteristics similar to traffic flow on a large scale and jitter characteristics similar to communication flow on a small scale. The DTW algorithm can be used to evaluate the similarity between the combined data.

[0093] The DTW average similarity calculation method is as follows:

[0094] A1, such as Figure 3 As shown, a cost matrix is ​​defined. That is, the historical data of adjacent base stations and the target base station The predicted data constitutes a two-dimensional matrix, where any position within the matrix... The value is the historical data of adjacent base stations and the target base station. The Euclidean distance between the predicted data and their corresponding indices is , This indicates that this element is located at the leftmost position in the cost matrix. List, This indicates that this element is located at the top of the cost matrix from bottom to top. OK.

[0095] A2. Define a dynamic time-planning shortest path. It meets the following conditions:

[0096] a1、 The number of elements is between the longest side of the matrix and the number of elements in the matrix;

[0097] a2、 With the beginning and end fixed, the shortest path can only be from the bottom left corner to the top right corner of the matrix;

[0098] a3, if , ,but ,and Each step of the shortest path can only be taken in three directions: right, up, or upper right.

[0099] A3. In the cost matrix In the middle, the search proceeds from the bottom left corner to the top right corner to find the shortest path that meets the requirements. At this point, the adjacent base stations... Historical data and target base stations The DTW average similarity function for the predicted data is:

[0100]

[0101] (3) Indicates the target base station The error between predicted data and actual data.

[0102] in, This represents the sum of time delays and the time delay variance weighting factor. Indicates the target base station The error between the predicted and actual number of vehicles connected. Indicates the target base station Error between predicted and actual communication demand data This indicates the first and last timestamps of the predicted data.

[0103] In this embodiment, as shown in Table 1, the optimization model uses the DDPG algorithm to optimize the prediction model. Specifically, a deep deterministic policy gradient can be adopted, using the number of vehicles and traffic at the base station as the state space and the predictions of the base station as the action space. The optimal prediction policy is obtained by training an agent to maximize the reward. .

[0104] Table 1

[0105]

[0106] S4. Minimize the delay and delay variance, and use the projective gradient descent method to solve for the optimal bandwidth allocation scheme;

[0107] After step S3, the base station Predictive data is obtained by requesting data from neighboring base stations and by performing model calculations. . Indicates base station The start timestamp of the prediction data is important because the transmission and prediction process takes time. Therefore, the start time of the predicted data is earlier than the time when the predicted data is obtained, that is, the time when the predicted data is obtained is... .Will Number of vehicles arriving within and communication volume Replace with forecast data In the process, mixed prediction data is obtained. At this point, the vehicle arrival time is specifically predicted as follows:

[0108]

[0109]

[0110] in, and Represents real data. and This indicates the predicted data, where the first subscript represents the data timestamp and the second represents the base station of the predicted data.

[0111] base station Based on mixed forecast data We construct an optimization problem with the goal of minimizing system latency and latency variance at each time step. We use the projective gradient descent method to solve the optimization problem and obtain the optimal bandwidth allocation scheme, which is the bandwidth allocation scheme for currently connected vehicles and future connected vehicles.

[0112] The optimization problem is that at each time step... The total time to run the gradient descent method after obtaining the prediction data is calculated by internally calculating the average vehicle latency within the base station and the squared difference between the mean of this average latency over all time steps. Inner summation is obtained Solve this optimization problem using gradient descent:

[0113]

[0114] in, Indicates base station At any moment The total bandwidth allocated to vehicles. Indicates time At the base station Available bandwidth on This represents an optimization problem; the first constraint represents the bandwidth allocated at each time step. No more than available bandwidth The second constraint represents the bandwidth of the vehicle segment. To exceed the acceptable minimum bandwidth requirement, optimization problem It consists of two parts: the sum of the system's average delays and the variance of the system's average delays, specifically:

[0115] in, This represents the sum of the system's average latency, ensuring that the communication resource allocation scheme has low latency. Indicates base station At any moment The total bandwidth allocated to vehicles. Indicates a guarantee Small positive numbers with a non-zero denominator. , This represents the sum of time delays and the time delay variance weighting factor. This represents the total time to run the gradient descent method after obtaining the predicted data. Due to the time delay in the transmission and prediction processes, the start time of the predicted data is earlier than the time when the predicted data is obtained. In this case, the total time to run the gradient descent method is no greater than the time span of the predicted data. .

[0116] , Indicates base station Will Time bandwidth The remaining bandwidth after allocation Indicates base station At any moment The total bandwidth allocated to vehicles, the third item This indicates the recovery status of previously allocated bandwidth. This indicates the reclaiming of the sliding window, from the last time it was not reclaimed to the current time. This is a bandwidth reclamation indicator; 1 indicates that the bandwidth has not been reclaimed, and 0 indicates that it has been reclaimed. Indicates a delay indicator, when hour, Otherwise, it is 0. express The average latency of service transmission corresponding to this portion of bandwidth Therefore, bandwidth is allocated. Front Base Station exist Available bandwidth at any time is Only with It is related to the allocation situation before the time. Once the bandwidth allocation is complete, the available bandwidth for the next moment can be calculated directly.

[0117] This optimization problem is solved using the projected gradient method, as follows:

[0118] B1. Initialization: Randomly generate bandwidth allocation scheme. Set the number of iterations ,get ;

[0119] B2. Calculate the average delay Calculate the optimization function exist Gradient ;

[0120] Specifically, the gradient is:

[0121]

[0122] B3. Based on gradient direction and step size renew ,get ;

[0123] B4. The updated version Projected onto the feasible region ,Right now In the middle, the distance in the feasible region is obtained. nearest point The specific projection method is as follows:

[0124]

[0125] in, Represents the projection operator. Indicates the first During the next iteration, the base station exist Available bandwidth at any given time.

[0126] B5. Determine if convergence has occurred: If the decrease in the optimization function is less than the threshold... ,Right now Or the number of iterations is greater than the maximum number of iterations, i.e. Then output the current solution. ,otherwise Return to step B2.

[0127] S5, base station According to the optimal bandwidth allocation scheme Allocate bandwidth resources, take into account subsequent traffic flow, and realize resource scheduling for future traffic flow;

[0128] Among them, vehicles With time At the base station Traffic generated on Assign weights, distribute ,Right now .

[0129] In summary, the application scenario of the method of this invention is as follows: the road network consists of roads, vehicles, and base stations (Roadside Units, RSUs). Vehicles communicate through RSUs, and RSUs cover the entire road network. The method of this invention combines dynamic time planning algorithms and DDPG to obtain two time-series data points: the historical number of connected vehicles and communication demand of surrounding base stations. Based on this, it predicts the number of vehicles and communication demand that may arrive in the future. By optimizing these predictions, it guides the resource reservation of base stations, thereby maximizing the overall resource utilization rate for the present and future. The resource reservation scheme for base stations is further improved by using gradient descent to optimize the resource reservation scheme. The method of this invention considers road network conditions where communication resource demand increases sharply when vehicles have many intelligent services. It can obtain the gains in traffic communication flow prediction in terms of information comprehensiveness, spatiotemporal correlation, and data complementarity by using traffic flow prediction and communication resource reservation techniques. This establishes a comprehensive and highly accurate traffic communication flow prediction model to improve the utilization rate of vehicle-to-everything (V2X) spectrum resources.

[0130] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A base station resource scheduling method based on complementary prediction of traffic and communication features, the specific steps of which are as follows: S1, Base station sampling; Base station time step For the current data sampling, a set of sampled data includes: Current moment Connected to base station Number of vehicles The sum of the communication needs generated by these vehicles ; in, Indicates in Time vehicle Is it connected to the base station? superior, Indicates a connection. This indicates that the connection is not established. express Vehicles within the time period At the base station The communication needs generated above; This represents the total number of all vehicles in the system. The base station stores the sampled data in a historical data pool and saves it. Group data Determine if the sampling exceeds Delete more than 100 data entries. If the oldest historical data is not older than the limit, sampling will continue while data will not be made available to other base stations until the sampling limit is reached. This; neighboring base stations request this When processing group data, the latest data is transmitted directly through inter-base station communication; in, Indicates the latest A vehicle count vector composed of sampled vehicle counts. Indicates the latest The communication demand vector, composed of samples of vehicle communication demands, together with the traffic flow and user communication characteristic data of the complete road network. ; Simultaneously, the base station records the remaining bandwidth of the base station and the bandwidth allocated to vehicles, which is used for calculating the base station's communication resource reservation. At any moment The remaining bandwidth is represented as: ; in, Indicates vehicle time At the base station The actual bandwidth allocated. Indicates base station Total bandwidth; S2, base station Obtain the number of vehicles accessing adjacent base stations and the time-series data of communication volume of adjacent base stations; base station Request traffic and communication data from neighboring base stations to obtain information about several neighboring base stations. data , , ; S3. Predicting traffic flow and communication demand based on dynamic time planning algorithm and DDPG; Base station The number of vehicles accessing adjacent base stations and their communication traffic time-series data are input into the optimization model to obtain future... time step Predicted traffic flow and communication demand data ; in, This represents predicted traffic flow data. This represents data predicting communication demand. Indicates base station The start timestamp of the predicted data; The optimization model is specifically as follows: ； in, Represents the predictive model; Indicates the data transmission time between adjacent base stations. This indicates the historical data of adjacent base stations and the target base station. DTW average similarity of the predicted data Indicates the target base station The error between predicted and actual data; Indicates time base station Maximum number of vehicles that can be connected; Indicates vehicle time At the base station Service transmission latency Indicates vehicle time At the base station The maximum acceptable service transmission latency Indicates vehicle time At the base station The minimum acceptable bandwidth requirement; Indicates base station and base stations The amount of data transmitted between them Indicates base station and base stations The transmission rate between them; Indicates optimizing model training accuracy, This represents the minimum acceptable training accuracy of the optimized model; This indicates that the number of vehicles connected to the base station does not exceed the maximum number of vehicles that can be served. This indicates that the actual latency of the vehicle is no greater than the maximum acceptable latency, from which the minimum bandwidth of the vehicle can be obtained. , This indicates that the historical data transmission time needs to be much less than A specific time step is needed; otherwise, the data will be too old and lose its meaning. This indicates that the training precision is greater than the minimum precision; Then the DDPG algorithm is used to optimize the prediction model; S4. Minimize the delay and delay variance, and use the projective gradient descent method to solve for the optimal bandwidth allocation scheme; After step S3, the base station Predictive data is obtained by requesting data from neighboring base stations and by performing model calculations. ; Indicates base station The start timestamp of the prediction data is important because the transmission and prediction process takes time. Therefore, the start time of the predicted data is earlier than the time when the predicted data is obtained, that is, the time when the predicted data is obtained is... ;Will Number of vehicles arriving within and communication volume Replace with forecast data In the process, mixed prediction data is obtained. At this point, the vehicle arrival mixed prediction data is as follows: ； ； in, and Represents real data. and This indicates predicted data, where the first subscript represents the data timestamp and the second represents the base station from which the predicted data is generated. base station Based on mixed forecast data An optimization problem is constructed, and the projective gradient descent method is used to solve the optimization problem to obtain the optimal bandwidth allocation scheme, that is, the bandwidth allocation scheme for the current connected vehicles and the future connected vehicles. The optimization problem is that at each time step... The calculation of the average vehicle latency within the base station and the squared difference between the mean of this average latency at all time steps is performed over the total time. Inner summation is obtained Solve this optimization problem using gradient descent: ； in, Indicates base station At any moment The total bandwidth allocated to vehicles. Indicates time At the base station Available bandwidth on This represents an optimization problem; the first constraint represents the bandwidth allocated at each time step. No more than available bandwidth The second constraint represents the bandwidth of the vehicle segment. To exceed the acceptable minimum bandwidth requirement, optimization problem It consists of two parts: the sum of the system's average delays and the variance of the system's average delays, specifically: ； in, This represents the sum of the system's average latency, ensuring that the communication resource allocation scheme has low latency. Indicates base station At any moment The total bandwidth allocated to vehicles. Indicates a guarantee Small positive numbers with a non-zero denominator; , This represents the sum of time delays and the time delay variance weighting factor. This represents the total time taken to run the gradient descent method after obtaining the predicted data, and ; , Indicates base station Will Time bandwidth The remaining bandwidth after allocation Indicates base station At any moment The total bandwidth allocated to vehicles, the third item This indicates the recovery status of previously allocated bandwidth. This indicates the reclaiming of the sliding window, from the last time it was not reclaimed to the current time. This is a bandwidth reclamation indicator; 1 indicates that the bandwidth has not been reclaimed, and 0 indicates that it has been reclaimed. Indicates the delay indicator, when hour, Otherwise, it is 0. express The average latency of service transmission corresponding to this portion of bandwidth Therefore, bandwidth is allocated. Front Base Station exist Available bandwidth at any time is Only with It is related to the allocation situation before the time. Once the allocation is complete, the available bandwidth for the next moment can be calculated directly. S5, base station According to the optimal bandwidth allocation scheme Allocate bandwidth resources, take into account subsequent traffic flow, and realize resource scheduling for future traffic flow; Among them, vehicles With time At the base station Traffic generated on Assign weights, distribute ,Right now .

2. The base station resource scheduling method based on complementary prediction of traffic and communication features according to claim 1, characterized in that, In step S3, the optimization problem includes the data transmission time between adjacent base stations, historical data of adjacent base stations, and the target base station. DTW average similarity of predicted data, target base station The errors between the predicted and actual data are as follows: (1) Represents the set of adjacent base stations to base station The sum of communication delays for transmitting data; in, Indicates base station to base station Data size to be transmitted Indicates base station to base station Bandwidth for data transmission Indicates base station to base station Transmission power for transmitting data Indicates base station to base station Channel gain for data transmission Indicates noise power. Indicates base station Base stations that generate noise externally; (2) Indicates adjacent base stations Historical data and target base stations DTW average similarity of the predicted data; in, This represents the DTW average similarity function, used to calculate the similarity between two time series data points; the smaller the value, the more similar they are. This represents the weight of each historical data point, which decays exponentially with the freshness of the data over time. , and These represent the attenuation factor and the attenuation exponent factor, respectively. (3) Indicates the target base station The error between predicted and actual data; in, This represents the sum of time delays and the time delay variance weighting factor. Indicates the target base station The error between the predicted and actual number of vehicles connected. Indicates the target base station Error between predicted and actual communication demand data This indicates the first and last timestamps of the predicted data.

3. The base station resource scheduling method based on complementary prediction of traffic and communication features according to claim 2, characterized in that, In step S3, the method for calculating the DTW average similarity is as follows: A1. Define a cost matrix. That is, the historical data of adjacent base stations and the target base station The predicted data constitutes a two-dimensional matrix, where any position within the matrix... The value is the historical data of adjacent base stations and the target base station. The Euclidean distance between the predicted data and their corresponding indices is , This indicates that this element is located at the leftmost position in the cost matrix. List, This indicates that this element is located at the top of the cost matrix from bottom to top. OK; A2. Define a dynamic time-planning shortest path. It meets the following conditions: a1、 The number of elements is between the longest side of the matrix and the number of elements in the matrix; a2、 With the beginning and end fixed, the shortest path can only be from the bottom left corner to the top right corner of the matrix; a3, if , ,but ,and Each step of the shortest path can only be taken in three directions: right, up, and upper right. A3. In the cost matrix In the middle, the search proceeds from the bottom left corner to the top right corner to find the shortest path that meets the requirements. At this point, the adjacent base stations... Historical data and target base stations The DTW average similarity function for the predicted data is: 。 4. The base station resource scheduling method based on complementary prediction of traffic and communication features according to claim 1, characterized in that, In step S3, the projection gradient method is used to solve the optimization problem, as follows: B1. Initialization: Randomly generate bandwidth allocation scheme. Set the number of iterations ,get ; B2. Calculate the average delay Calculate the optimization function exist Gradient ; Specifically, the gradient is: ； B3. Based on gradient direction and step size renew ,get ; B4. The updated version Projected onto the feasible region ,Right now In the middle, the distance in the feasible region is obtained. nearest point The specific projection method is as follows: ； in, Represents the projection operator. Indicates the first During the next iteration, the base station exist Available bandwidth at any given time; B5. Determine if convergence has occurred: If the decrease in the optimization function is less than the threshold... ,Right now Or the number of iterations is greater than the maximum number of iterations, i.e. Then output the current solution. ,otherwise Return to step B2.

Images